Neural Information Processing Systems

NIPS*2000 Online Abstracts and Papers

This is a list of the papers that were presented at NIPS*2000. They are listed in (approximately) alphabetical order according to the last name of the first author. The title is a link to a Postscript or PDF version of the paper. BibTeX entries for all papers can be found here.

What Can a Single Neuron Compute? -- Blaise Ag{\"u}era y Arcas, Adrienne L. Fairhall, William Bialek
In this paper we formulate a description of the computation performed by a neuron as a combination of dimensional reduction and nonlinearity. We evaluate this description for the Hodgkin-Huxley model information-theoretically.

Who Does What? A Novel Algorithm to Determine Function Localization -- Ranit Aharonov-Barki, Isaac Meilijson, Eytan Ruppin
We introduce a novel algorithm, termed FDA (Functional Deficit Algorithm), that quantitatively measures the contributions of elements of a neural system to the tasks it performs. The algorithm identifies the neurons or areas which participate in a cognitive or behavioral task, given data about performance decrease in a small set of lesions. It also allows the accurate prediction of deficits due to multi-element lesions. The effectiveness of the new algorithm is demonstrated in two models of recurrent neural networks with complex interactions among the elements. The algorithm is scalable and applicable to the analysis of large neural networks. Given the recent advances in reversible inactivation techniques, it has the potential to significantly contribute to the understanding of the organization of biological nervous systems, and to shed light on the long-lasting debate about local versus distributed computation in the brain.

Programmable Reinforcement Learning Agents -- David Andre, Stuart J. Russell
We present an expressive agent design language for reinforcement learning that allows the user to constrain the policies considered by the learning process by {\em partially} specifying a program for the agent. The language includes standard programmatic features such as parameterized subroutines, temporary interrupts, aborts, and memory variables, and allows for the partial specification of complex agent programs. For learning that which isn't specified, we present provably convergent learning algorithms. We demonstrate that agent programs written in the language are concise as well as modular, which is important to facilitate state abstraction and the transferability of learned skills.

From Mixtures of Mixtures to Adaptive Transform Coding -- Cynthia Archer, Todd K. Leen
Transform codes (e.g. JPEG, wavelet coding) compress data by transforming data into coordinates presumed to be more favorable for compression, allocating coding bits among the new coordinates, and quantizing the transformed coordinate scalar values. Often the coding procedure includes an ad hoc choice of transform (e.g. PCA, DCT) and suboptimal bit allocation and quantizer design. As detrimental as the suboptimal choices for these steps is the fact that designs proceed piecemeal, with interactions between steps typically ignored. This paper does three things: i) We establish a principled framework for adaptive transform coding that starts from a probabilistic latent variable model in the form of a mixture of constrained Gaussian mixtures. A hard-clustering limit of this model reveals a truly optimal transform coding algorithm, which is a constrained version of the generalized Lloyd algorithm for vector quantizer design. ii) We introduce a new transform that is required by the mathematical development. Unlike PCA, DCT, and wavelet transforms, this new transform is explicitly optimized for compression. iii) We exercise the new algorithm on image data. The tested implementation reduces computational complexity, while sustaining a small (estimated $< 0.2$ dB) decrease in signal-to-noise ratio, by using the PCA transform and a greedy bit allocation procedure instead of our new coding transform and optimal bit allocation. Adaptive transform coders designed with our algorithm improve signal-to-noise ratio by 2.5 dB compared to global transform coding and 1 to 3 dB compared to other adaptive transform coders.

Dendritic Compartmentalization Could Underlie Competition and Attentional Biasing of Simultaneous Visual Stimuli -- Kevin A. Archie, Bartlett W. Mel
Neurons in area V4 have relatively large receptive fields (RFs), so multiple visual features are simultaneously ``seen'' by these cells. Recordings from single V4 neurons suggest that simultaneously presented stimuli compete to set the output firing rate, and that attention acts to isolate individual features by biasing the competition in favor of the attended object. We propose that both stimulus competition and attentional biasing arise from the spatial segregation of afferent synapses onto different regions of the excitable dendritic tree of V4 neurons. The pattern of feedforward, stimulus-driven inputs follows from a Hebbian rule: excitatory afferents with similar RFs tend to group together on the dendritic tree, avoiding randomly located inhibitory inputs with similar RFs. The same principle guides the formation of inputs that mediate attentional modulation. Using both biophysically detailed compartmental models and simplified models of computation in single neurons, we demonstrate that such an architecture could account for the response properties and attentional modulation of V4 neurons. Our results suggest an important role for nonlinear dendritic conductances in extrastriate cortical processing.

Place Cells and Spatial Navigation Based on 2D Visual Feature Extraction, Path Integration, and Reinforcement Learning -- Angelo Arleo, Fabrizio Smeraldi, St\'ephane Hug, Wulfram Gerstner
We model hippocampal place cells and head-direction cells by combining allothetic (visual) and idiothetic (proprioceptive) stimuli. Visual input, provided by a video camera on a miniature robot, is preprocessed by a set of Gabor filters on 31 nodes of a log-polar retinotopic graph. Unsupervised Hebbian learning is employed to incrementally build a population of localized overlapping place fields. Place cells serve as basis functions for reinforcement learning. Experimental results for goal-oriented navigation of a mobile robot are presented.

Speech Denoising and Dereverberation Using Probabilistic Models -- Hagai Attias, John C. Platt, Alex Acero, Li Deng
This paper presents a unified probabilistic framework for denoising and dereverberation of speech signals. The framework transforms the denoising and dereverberation problems into Bayes-optimal signal estimation. A key point in our approach is the use of a strong speech model as a prior. The speech model is a mixture model pre-trained on a large data set of clean speech. The framework covers both single and multiple microphones. We apply this approach to a noisy reverberant speech signal, and get results substantially better than standard methods.

Combining ICA and Top-Down Attention for Robust Speech Recognition -- Un-Min Bae, Soo-Young Lee
We present an algorithm which compensates for the mismatches between characteristics of real-world problems and assumptions of independent component analysis algorithm. To provide additional information to the ICA network, an MLP classifier is added to the separated signal channel and the error of the classifier is backpropagated to the ICA network. Then, the unmixing matrix is retrained according to a new cost function representing the backpropagated error as well as independence. It modifies the density of recovered signals to the density appropriate for classification. Under a stationary mixing environment, the averaged adaptation of the unmixing system over the total data can roughly approximate the non-ideal mixing characteristics, such as the nonlinearity of microphones. For noisy speech signal recorded in real environments, the algorithm improved the recognition performance and showed robustness against parametric changes.

Modelling Spatial Recall, Mental Imagery and Neglect -- Suzanna Becker, Neil Burgess
We present a computational model of the neural mechanisms in the parietal and temporal lobes that support spatial navigation, imagery, and episodic recall. Long term representations are stored in the hippocampus, and are associated with local spatial and object-related features in the parahippocampal region. Viewer-centered representations are dynamically generated from long term memory in the parietal part of the model. The model thereby simulates recall and imagery of locations and objects in complex environments. After parietal damage, the model exhibits hemispatial neglect in mental imagery that rotates with the imagined perspective of the observer, as in the famous Milan Square experiment reported by Bisiach and Luzatti (1978). Our model makes novel predictions for the neural representations in the parahippocampal and parietal regions and for behavior in healthy volunteers and neuropsychological patients.

Shape Context: A New Descriptor for Shape Matching and Object Recognition -- Serge Belongie, Jitendra Malik, Jan Puzicha
We introduce a new shape descriptor, the {\em shape context}, for correspondence recovery and shape-based object recognition. The shape context at a point captures the distribution over relative positions of other shape points and thus summarizes global shape in a rich, local descriptor. Shape contexts greatly simplify recovery of correspondences between points of two given shapes. Moreover, the shape context leads to a robust score for measuring shape similarity, once shapes are aligned. The shape context descriptor is tolerant to all common shape deformations. As a key advantage no special landmarks or key-points are necessary. It is thus a generic method with applications in object recognition, image registration and point set matching. Using examples involving both handwritten digits and 3D objects, we illustrate its power for object recognition.

Efficient Learning of Linear Perceptrons -- Shai Ben-David, Hans Ulrich Simon
We consider the existence of efficient algorithms for learning the class of half-spaces in ${\bf R}^n$ in the agnostic learning model (i.e., making no prior assumptions on the example-generating distribution). The resulting combinatorial problem - finding the best agreement half-space over an input sample - is NP hard to approximate to within some constant factor. We suggest a way to circumvent this theoretical bound by introducing a new measure of success for such algorithms. An algorithm is $\mu$-successful if the agreement ratio of the half-space it outputs is as good as that of any half-space once training points that are inside the $\mu$-margins of its separating hyper-plane are disregarded. We prove crisp computational complexity results with respect to this success measure; On one hand, for every positive $\mu$, there exist efficient (poly-time) $\mu$-successful learning algorithms. On the other hand, we prove that unless P$=$NP, there is no algorithm that runs in time polynomial in the sample size and in $1/\mu$ that is $\mu$-successful for all $\mu$.

A Support Vector Method for Clustering -- Asa Ben-Hur, David Horn, Hava T. Siegelmann, Vladimir Vapnik
We present a novel method for clustering using the support vector machine approach. Data points are mapped to a high dimensional feature space, where support vectors are used to define a sphere enclosing them. The boundary of the sphere forms in data space a set of closed contours containing the data. As the kernel parameter is varied these contours fit the data more tightly and splitting of contours occurs. The contours are interpreted as cluster boundaries and the points within each disconnected contour are defined as a cluster. Cluster boundaries can take on arbitrary geometrical shapes and clusters are separated by valleys in the underlying probability distribution. As in other SV algorithms, outliers can be dealt with by introducing a soft margin constant leading to smoother cluster boundaries. The hierarchical structure of the data is explored by varying the two parameters. We investigate the dependence of our method on these parameters and apply it to several data sets.

A Neural Probabilistic Language Model -- Yoshua Bengio, R\'ejean Ducharme, Pascal Vincent
Modeling high-dimensional discrete random variables without significant a priori knowledge of conditional independence is intrinsically difficult because of the curse of dimensionality: combinations of values on which the model will be tested are likely to be different from all the combinations of values seen in the training data. Neural networks can capture, with few parameters, dependencies of any order between the variables (but not all of them, of course). We consider here the case in which each variable can take a large number of values, as in statistical language modeling. The proposed model learns the conditional distribution of the next word at the same time as learning a distributed representation for words (in a way that is similar to latent semantic analysis), allowing to generalize from contexts (combinations of words) seen in the training set to new combinations of words, because of the similarity of their internal representations. Successful comparisons against smoothed trigrams are reported.

A Variational Mean-Field Theory for Sigmoidal Belief Networks -- Chiranjib Bhattacharyya, S. Sathiya Keerthi
A new variational derivation of Plefka's mean-field theory is presented. This theory is then applied to sigmoidal belief networks with the aid of further approximations. Empirical evaluation on small scale networks shows that the proposed approximations are quite competitive with other mean field approximations proposed in the literature.

Stability and Noise in Biochemical Switches -- William Bialek
Many processes in biology, from the regulation of gene expression in bacteria to memory in the brain, involve switches constructed from networks of biochemical reactions. Crucial molecules are present in small numbers, raising questions about noise and stability. Analysis of noise in simple reaction schemes indicates that switches stable for years and switchable in milliseconds can be built from fewer than one hundred molecules. Prospects for direct tests of this prediction, as well as implications, are discussed.

Emergence of Movement Sensitive Neurons' Properties by Learning a Sparse Code for Natural Moving Images -- Rafal Bogacz, Malcolm W. Brown, Christophe Giraud-Carrier
Olshausen \& Field demonstrated that a learning algorithm that attempts to generate a sparse code for natural scenes develops a complete family of localised, oriented, bandpass receptive fields, similar to those of 'simple cells' in V1. This paper describes an algorithm which finds a sparse code for sequences of images that preserves information about the input. This algorithm when trained on natural video sequences develops bases representing the movement in particular direction with particular speed, similar to the receptive fields of the movement-sensitive cells observed in cortical visual areas. Furthermore, in contrast to previous approaches to learning direction selectivity, the timing of neuronal activity encodes the phase of the movement, so the precise timing of spikes is crucially important to the information encoding.

New Approaches Towards Robust and Adaptive Speech Recognition {\rm (invited paper)} -- Herv\'e Bourlard, Samy Bengio, Katrin Weber
In this paper, we discuss some new research directions in automatic speech recognition (ASR), and which somewhat deviate from the usual approaches. More specifically, we will motivate and briefly describe new approaches based on multi-stream and multi/band ASR. These approaches extend the standard hidden Markov model (HMM) based approach by assuming that the different (frequency) channels representing the speech signal are processed by different (independent) ``experts'', each expert focusing on a different characteristic of the signal, and that the different stream likelihoods (or posteriors) are combined at some (temporal) stage to yield a global recognition output. As a further extension to multi-stream ASR, we will finally introduce a new approach, referred to as HMM2, where the HMM emission probabilities are estimated via state specific feature based HMM responsible for merging the stream information and modeling their possible correlation.

Algorithmic Stability and Generalization Performance -- Olivier Bousquet, Andr\'e Elisseeff
We present a novel way of obtaining PAC-style bounds on the generalization error of learning algorithms, explicitly using their stability properties. A 'stable' learner is one for which the learned solution does not change much with small changes in the training set. The bounds we obtain do not depend on any measure of the complexity of the hypothesis space (e.g. VC dimension) but rather depend on how the learning algorithm searches this space, and can thus be applied even when the VC dimension is infinite. We demonstrate that regularization networks possess the required stability property and apply our method to obtain new bounds on their generalization performance.

Exact Solutions to Time-Dependent MDPs -- Justin A. Boyan, Michael L. Littman
We describe an extension of the Markov decision process model in which a continuous time dimension is included in the state space. This allows for the representation and exact solution of a wide range of problems in which transitions or rewards vary over time. We examine problems based on route planning with public transportation and telescope observation scheduling.

Direct Classification with Indirect Data -- Timothy X Brown
This paper studies the following problem: Given input/output examples from a real-valued function, classify the input space according to when the function output is in a given subset. Obvious approaches to the problem are inconsistent or overly complex when the output examples have skewed noise or when the examples are indirect measurements of the function outputs. We contribute a general classifier that is both consistent and avoids the unnecessary complexity of estimating the function. The approach is demonstrated on a problem from telecommunications.

Model Complexity, Goodness of Fit and Diminishing Returns -- Igor V. Cadez, Padhraic Smyth
We investigate a general characteristic of the trade-off in learning problems between goodness-of-fit and model complexity. Specifically we characterize a general class of learning problems where the goodness-of-fit function can be shown to be convex within first-order as a function of model complexity. This general property of ``diminishing returns'' is illustrated on a number of real data sets and learning problems, including finite mixture modeling and multivariate linear regression.

A Linear Programming Approach to Novelty Detection -- Colin Campbell, Kristin P. Bennett
Novelty detection involves modeling the normal behaviour of a system hence enabling detection of any divergence from normality. It has potential applications in many areas such as detection of machine damage or highlighting abnormal features in medical data. One approach is to build a hypothesis estimating the support of the normal data i.e. constructing a function which is positive in the region where the data is located and negative elsewhere. Recently kernel methods have been proposed for estimating the support of a distribution and they have performed well in practice training involves solution of a quadratic programming problem. In this paper we propose a simpler kernel method for estimating the support based on linear programming. The method is easy to implement and can learn large datasets rapidly. We demonstrate the method on medical and fault detection datasets.

Decomposition of Reinforcement Learning for Admission Control of Self-Similar Call Arrival Processes -- Jakob Carlstr{\"o}m
This paper presents predictive gain scheduling, a technique for decomposition of reinforcement learning problems. Admission control of self-similar call traffic in a communications network is used as an example. The control problem is decomposed into on-line prediction of near-future call arrival rates, and precomputation of policies for Poisson call arrival processes. At decision time, predicted arrival rates are used to select a policy. Simulations show that predictive gain scheduling results in significantly faster learning without any performance loss, compared to a reinforcement learning controller that does not decompose the problem.

Overfitting in Neural Nets: Backpropagation, Conjugate Gradient, and Early Stopping -- Rich Caruana, Steve Lawrence, Lee Giles
The conventional wisdom is that backprop nets with excess hidden units generalize poorly: they have low bias, but high variance. Empirically, however, large nets often generalize better. We show that nets with excess capacity generalize well if trained with backprop and early stopping. Experiments suggest two reasons for this: 1) overfitting can vary significantly in different regions of the model. Excess capacity allows better fit to regions of high non-linearity, and backprop often avoids overfitting the regions of low non-linearity. 2) regardless of size, nets learn task subcomponents in similar sequence. Big nets pass through stages similar to those learned by smaller nets. Early stopping can stop training the large net when it generalizes comparably to a smaller net. We also show that conjugate gradient can yield worse generalization because it overfits regions of low non-linearity when learning to fit regions of high non-linearity.

Incremental and Decremental Support Vector Machine Learning -- Gert Cauwenberghs, Tomaso Poggio
An on-line recursive algorithm for training a soft-margin classification support vector machine, one vector at a time, is presented. Adiabatic increments retain the Kuhn-Tucker conditions on all previously seen training data, in a number of steps each computed analytically. The incremental procedure is reversible, and decremental ``unlearning'' offers an efficient method to exactly evaluate leave-one-out generalization performance. Interpretation of decremental unlearning in feature space sheds new light on the relationship between generalization and geometry of the training data.

Vicinal Risk Minimization -- Olivier Chapelle, Jason Weston, L\'eon Bottou, Vladimir Vapnik
The Vicinal Risk Minimization principle establishes a bridge between generative models and methods derived from the Structural Risk Minimization Principle such as Support Vector Machines or Statistical Regularization. We explain how VRM provides a framework which integrates a number of existing algorithms, such as Parzen windows, Support Vector Machines, Ridge Regression, Constrained Logistic Classifiers and Tangent-Prop. We then show how the approach implies new algorithms for solving problems usually associated with generative models. New algorithms are described for dealing with pattern recognition problems with very different pattern distributions and dealing with unlabeled data. Preliminary empirical results are presented.

Temporally Dependent Plasticity: An Information Theoretic Account -- Gal Chechik, Naftali Tishby
The fundamental paradigm of Hebbian learning has recently received a novel interpretation with the discovery of synaptic plasticity that depends on the relative timing of pre and post synaptic spikes. [e.g. Markram et al. 1997; Zhang et al. 1998]. While this type of plasticity has already been studied in various computational frameworks, a generic computational motivation or its derivation from first principles is still missing. In this paper we derive temporally dependent learning from the basic principle of mutual information maximization and study its relation to the experimentally observed plasticity. The comparison shows that not only is the biological learning rule similar in form to the analytically derived one, but it also increases mutual information up to a local optimum. The analysis yields new insights into the temporal characteristics of the observed learning rule and its dependency on neuronal biophysical parameters.

Gaussianization -- Scott Saobing Chen, Ramesh A. Gopinath
High dimensional data modeling is difficult because of the so-called ``curse of dimensionality''. We propose a technique called ``Gaussianization'' for high dimensional density estimation, which alleviates the curse of dimensionality by exploiting the independence structures in the data. Gaussianization is motivated from recent developments in the statistics literature: projection pursuit, independent component analysis and Gaussian mixture models with semi-tied covariances. We propose an iterative Gaussianization procedure which converges weakly: at each iteration, the data is first transformed to the least dependent coordinates and then each coordinate is marginally Gaussianized by univariate techniques. Gaussianization offers density estimation sharper than traditional kernel methods and radial basis function methods. Gaussianization can be viewed as efficient solution of nonlinear independent component analysis and high dimensional projection pursuit.

The Missing Link - A Probabilistic Model of Document Content and Hypertext Connectivity -- David Cohn, Thomas Hofmann
In this paper, we describe a joint probabilistic model for modeling the contents and inter-connectivity of document collections such as sets of web pages or research paper archives. The model is based on a probabilistic factor decomposition and allows identifying principal topics of the collection as well as authoritative documents within those topics. Further, it allows mapping out the relationships of those topics to build a predictive model of link content. Among the many applications of this approach are information retrieval and search, topic identification, query disambiguation, focused web crawling, web authoring, and bibliometric analysis.

The Manhattan World Assumption: Regularities in Scene Statistics which Enable Bayesian Inference -- James M. Coughlan, Allen L. Yuille
Preliminary work by the authors made use of the so-called ``Manhattan world'' assumption about the scene statistics of city and indoor scenes. This assumption stated that such scenes were built on a cartesian grid which led to regularities in the image edge gradient statistics. In this paper we explore the general applicability of this assumption and show that, surprisingly, it holds in a large variety of less structured environments including rural scenes. This enables us, from a single image, to determine the orientation of the viewer relative to the scene structure and also to detect target objects which are not aligned with the grid. These inferences are performed using a Bayesian model with probability distributions (e.g. on the image gradient statistics) learned from real data.

Improved Output Coding for Classification Using Continuous Relaxation -- Koby Crammer, Yoram Singer
Output coding is a general method for solving multiclass problems by reducing them to multiple binary classification problems. Previous research on output coding has employed, almost solely, predefined discrete codes. We describe an algorithm that improves the performance of output codes by relaxing them to continuous codes. The relaxation procedure is cast as an optimization problem and is reminiscent of the quadratic program for support vector machines. We describe experiments with the proposed algorithm, comparing it to standard discrete output codes. The experimental results indicate that continuous relaxations of output codes often improve the generalization performance, especially for short codes.

Sparse Representation for Gaussian Process Models -- Lehel Csat\'o, Manfred Opper
We develop an approach for a sparse representation for Gaussian Process (GP) models in order to overcome the limitations of GPs caused by large data sets. The method is based on a combination of a Bayesian online algorithm together with a sequential construction of a relevant subsample of the data which fully specifies the prediction of the model. Experimental results on toy examples and large real-world datasets indicate the efficiency of the approach. {\bf Lehel Csato received the NIPS Best Student Paper Award sponsored by Athene Software}

Competition and Arbors in Ocular Dominance -- Peter Dayan
Hebbian and competitive Hebbian algorithms are almost ubiquitous in modeling pattern formation in cortical development. We analyse in theoretical detail a particular model (adapted from Piepenbrock \& Obermayer, 1999) for the development of 1d stripe-like patterns, which places competitive and interactive cortical influences, and free and restricted initial arborisation onto a common footing.

Explaining Away in Weight Space -- Peter Dayan, Sham Kakade
Explaining away has mostly been considered in terms of inference of states in belief networks. We show how it can also arise in a Bayesian context in inference about the weights governing relationships such as those between stimuli and reinforcers in conditioning experiments such as backward blocking. We show how explaining away in weight space can be accounted for using an extension of a Kalman filter model; provide a new approximate way of looking at the Kalman gain matrix as a whitener for the correlation matrix of the observation process; suggest a network implementation of this whitener using an architecture due to Goodall; and show that the resulting model exhibits backward blocking.

Feature Correspondence: A Markov Chain Monte Carlo Approach -- Frank Dellaert, Steven M. Seitz, Sebastian Thrun, Charles Thorpe
When trying to recover 3D structure from a set of images, the most difficult problem is establishing the correspondence between the measurements. Most existing approaches assume that features can be tracked across frames, whereas methods that exploit rigidity constraints to facilitate matching do so only under restricted camera motion. In this paper we propose a Bayesian approach that avoids the brittleness associated with singling out one ``best'' correspondence, and instead consider the distribution over all possible correspondences. We treat both a fully Bayesian approach that yields a posterior distribution, and a MAP approach that makes use of EM to maximize this posterior. We show how Markov chain Monte Carlo methods can be used to implement these techniques in practice, and present experimental results on real data.

A New Model of Spatial Representation in Multimodal Brain Areas -- Sophie Deneve, Jean-Rene Duhamel, Alexandre Pouget
Most model of spatial representation in the cortex assume cells with limited receptive fields that are defined in a particular egocentric frame of reference. However, cells outside of primary sensory cortex are either gain modulated by postural input, or partially shifting. We show that solving classical spatial tasks, like sensory prediction, multi-sensory integration, sensory-motor transformation and motor control require more complicated intermediate representation that are not invariant in one frame of reference. We present an iterative basis function map that perform these spatial tasks optimally with gain modulated and partially shifting units, and tests it against neurophysiological and neuropsychological data.

An Adaptive Metric Machine for Pattern Classification -- Carlotta Domeniconi, Jing Peng, Dimitrios Gunopulos
Nearest neighbor classification assumes locally constant class conditional probabilities. This assumption becomes invalid in high dimensions with finite samples due to the curse of dimensionality. Severe bias can be introduced under these conditions when using the nearest neighbor rule. We propose a locally adaptive nearest neighbor classification method to try to minimize bias. We use a {\em Chi-squared} distance analysis to compute a flexible metric for producing neighborhoods that are elongated along less relevant feature dimensions and constricted along most influential ones. As a result, the class conditional probabilities tend to be smoother in the modified neighborhoods, whereby better classification performance can be achieved. The efficacy of our method is validated and compared against other techniques using a variety of real world data.

High-temperature Expansions for Learning Models of Nonnegative Data -- Oliver B. Downs
Recent work has exploited boundedness of data for learning new types of generative model. For nonnegative data it has been shown that the maximum-entropy generative model is a Nonnegative Boltzmann Distribution not a Gaussian distribution, when the model is constrained to match the first and second order statistics of the data. Learning for practical sized problems is made difficult by computing expectations under this distribution. Here I present a second-order approximation for the model, obtained using a ``high-temperature'' expansion. The result is analogous to the TAP-Onsager equations for the Ising model. The theory is tested on learning a bimodal 2-dimensional model, and in learning a high-dimensional translationally-invariant distribution.

Incorporating Second-Order Functional Knowledge for Better Option Pricing -- Charles Dugas, Yoshua Bengio, Fran\c{c}ois B\'elisle, Claude Nadeau, Ren\'e Garcia
Incorporating prior knowledge of a particular task into the architecture of a learning algorithm can greatly improve generalization performance. We study here a case where we know that the function to be learned is non-decreasing in two of its arguments and convex in one of them. For this purpose we propose a class of functions similar to multi-layer neural networks but (1) that has those properties, (2) is a universal approximator of continuous functions with these properties. We apply this new class of functions to the task of modeling the price of call options. Experiments show improvements on regressing the price of call options using the new types of function classes that incorporate the a-priori constraints.

A Productive, Systematic Framework for the Representation of Visual Structure -- Shimon Edelman, Nathan Intrator
We describe a unified framework, based on the MDL principle, for the understanding of structure representation in primate vision. A model derived from this framework is shown to be both productive and effectively systematic. Specifically, it has the ability to associate objects that are related through a rearrangement of ``middle-scale'' parts, represented as image fragments. The model addresses the same concerns as previous work on compositional representation through the use of {\it what+where} receptive fields and attentional gain modulation. It does not require prior exposure to the individual parts, and avoids the need for abstract symbolic binding.

Discovering Hidden Variables: A Structure-Based Approach -- Gal Elidan, Noam Lotner, Nir Friedman, Daphne Koller
A serious problem in learning probabilistic models is the presence of hidden variables. These variables are not observed, yet interact with several of the observed variables. As such, they induce seemingly complex dependencies among the latter. In recent years, much attention has been devoted to the development of algorithms for learning parameters, and in some cases structure, in the presence of hidden variables. In this paper, we address the related problem of detecting hidden variables that interact with the observed variables. This problem is of interest both for improving our understanding of the domain and as a preliminary step that guides the learning procedure towards promising models. A very natural approach is to search for ``structural signatures'' of hidden variables --- substructures in the learned network that tend to suggest the presence of a hidden variable. We make this basic idea concrete, and show how to integrate it with structure-search algorithms. We evaluate this method on several synthetic and real-life datasets, and show that it performs surprisingly well.

Multiple Timescales of Adaptation in a Neural Code -- Adrienne L. Fairhall, Geoffrey D. Lewen, William Bialek, Robert R. de Ruyter van Steveninck
Many neural systems extend their dynamic range by adaptation. We examine the timescales of adaptation in the context of dynamically modulated rapidly-varying stimuli, and demonstrate in the fly visual system that adaptation to the statistical ensemble of the stimulus dynamically maximizes information transmission about the time-dependent stimulus. Further, while the rate response has long transients, the adaptation takes place on timescales consistent with optimal variance estimation.

Learning Joint Statistical Models for Audio-Visual Fusion and Segregation -- John W. Fisher III, Trevor Darrell, William T. Freeman, Paul Viola
People can understand complex auditory and visual information, often using one to disambiguate the other. Automated analysis, even at a low-level, faces severe challenges, including the lack of accurate statistical models for the signals, and their high-dimensionality and varied sampling rates. Previous approaches [6] assumed simple parametric models for the joint distribution, which, while tractable, cannot capture the complex signal relationships. We learn the joint distribution of the visual and auditory signals using a non-parametric approach. First, we project the data into a maximally informative, low-dimensional subspace, suitable for density estimation. Then we model the complicated stochastic relationships between the signals using a nonparametric density estimator. These learned densities allow processing across signal modalities. We demonstrate, on synthetic and real signals, localization in video of the face that is speaking in audio, and, conversely, audio enhancement of a particular speaker selected from the video.

Accumulator Networks: Suitors of Local Probability Propagation -- Brendan J. Frey, Anitha Kannan
One way to approximate inference in richly-connected graphical models is to apply the sum-product algorithm (a.k.a. probability propagation algorithm), while ignoring the fact that the graph has cycles. The sum-product algorithm can be directly applied in Gaussian networks and in graphs for coding, but for many conditional probability functions -- including the sigmoid function -- direct application of the sum-product algorithm is not possible. We introduce ``accumulator networks'' that have low local complexity (but exponential global complexity) so the sum-product algorithm can be directly applied. In an accumulator network, the probability of a child given its parents is computed by accumulating the inputs from the parents in a Markov chain or more generally a tree. After giving expressions for inference and learning in accumulator networks, we give results on the ``bars problem'' and on the problem of extracting images of translated, overlapping faces from an visual scene.

Sequentially Fitting ``Inclusive'' Trees for Inference in Noisy-OR Networks -- Brendan J. Frey, Relu Patrascu, Tommi S. Jaakkola, Jodi Moran
An important class of problems can be cast as inference in noisy-OR Bayesian networks, where the binary state of each variable is a logical OR of noisy versions of the states of the variable's parents. For example, in medical diagnosis, the presence of a symptom can be expressed as a noisy-OR of the diseases that may cause the symptom -- on some occasions, a disease may fail to activate the symptom. Inference in richly-connected noisy-OR networks is intractable, but approximate methods (e.g., variational techniques) are showing increasing promise as practical solutions. One problem with most approximations is that they tend to concentrate on a relatively small number of modes in the true posterior, ignoring other plausible configurations of the hidden variables. We introduce a new sequential variational method for bipartite noisy-OR networks, that favors including all modes of the true posterior and models the posterior distribution as a tree. We compare this method with other approximations using an ensemble of networks with network statistics that are comparable to the QMR-DT medical diagnostic network.

Factored Semi-Tied Covariance Matrices -- Mark J.F. Gales
A new form of covariance modelling for Gaussian mixture models and hidden Markov models is presented. This is an extension to an efficient form of covariance modelling used in speech recognition, semi-tied covariance matrices. The transform associated with each Gaussian component is now factored, reducing the number of free parameters. Maximum likelihood estimation schemes for all the model parameters are estimated including the component/transform assignment, transform and component parameters. This new model form is evaluated on a large vocabulary speech recognition task.

A New Approximate Maximal Margin Classification Algorithm -- Claudio Gentile
A new incremental learning algorithm is described which is shown to approximate the maximal margin hyperplane w.r.t. norm $p \geq 2$ for a set of linearly separable data. Our algorithm, called ALMA$_p$ (Approximate Large Margin algorithm w.r.t. norm $p$), takes $O(\frac{(p-1)\,X^2}{\alpha^2\,\gamma^2})$ corrections to separate the data with $p$-norm margin larger than $(1-\alpha)\,\gamma$, where $\gamma$ is the $p$-norm margin of the data and $X$ is a bound on the $p$-norm of the instances. ALMA$_p$ avoids quadratic (or higher-order) programming methods. It is very easy to implement and is as fast as on-line algorithms, such as Rosenblatt's perceptron. We report on some experiments comparing a variant of ALMA$_p$ to two incremental algorithms: Perceptron and Li and Long's ROMMA. These experiments show that our algorithm performs quite better than both.

Propagation Algorithms for Variational Bayesian Learning -- Zoubin Ghahramani, Matthew J. Beal
Variational approximations are becoming a widespread tool for Bayesian learning of graphical models. We provide some theoretical results for the variational updates in a very general family of conjugate-exponential graphical models. We show how the belief propagation and the junction tree algorithms can be used in the inference step of variational Bayesian learning. Applying these results to the Bayesian analysis of linear-Gaussian state-space models we obtain a learning procedure that exploits the Kalman smoothing propagation, while integrating over all model parameters. We demonstrate how this can be used to infer the hidden state dimensionality of the state-space model in a variety of synthetic problems and one real high-dimensional data set.

Reinforcement Learning with Function Approximation Converges to a Region -- Geoffrey J. Gordon
Many algorithms for approximate reinforcement learning are not known to converge. In fact, there are counterexamples showing that the adjustable weights in some algorithms may oscillate within a region rather than converging to a point. This paper shows that, for two popular algorithms, such oscillation is the worst that can happen: the weights cannot diverge, but instead converge with probability 1 to a bounded region. The algorithms are SARSA(0) and V(0); the latter is the algorithm used in the well-known TD-Gammon program.

The Kernel Gibbs Sampler -- Thore Graepel, Ralf Herbrich
We present an algorithm that samples the hypothesis space of kernel classifiers. Given a uniform prior over normalised weight vectors and a likelihood based on a model of label noise leads to a piecewise constant posterior that can be sampled by the kernel Gibbs sampler (KGS). The KGS is a Markov Chain Monte Carlo method that chooses a random direction in parameter space and samples from the resulting piecewise constant density along the line chosen. The KGS can be used as an analytical tool for the exploration of Bayesian transduction, Bayes point machines, active learning, and evidence-based model selection on small data sets that are contaminated with label noise. For a simple toy example we demonstrate experimentally how a Bayes point machine based on the KGS outperforms an SVM that is incapable of taking into account label noise.

From Margin to Sparsity -- Thore Graepel, Ralf Herbrich, Robert C. Williamson
We present an improvement of Novikoff's perceptron convergence theorem. Reinterpreting this mistake bound as a margin dependent sparsity guarantee allows us to give a PAC-style generalisation error bound for the classifier learned by the dual perceptron learning algorithm. The bound value crucially depends on the margin a support vector machine would achieve on the same data set using the same kernel. Ironically, the bound yields better guarantees than are currently available for the support vector solution itself.

`N-Body' Problems in Statistical Learning -- Alexander G. Gray, Andrew W. Moore
We present very fast algorithms for a large class of statistical problems, which we call all-point-pairs problems, or `$N$-body'-like problems. These are problems which abstractly require a comparison of each of the $N$ points in a dataset with each other point and would naively be solved using $N^2$ computations. Such algorithms cannot be applied to practical problems where $N$ is large. In this paper we focus on several examples of all-point-pairs problems within nonparametric statistical learning: nearest-neighbor classification, kernel density estimation, outlier detection, the two-point correlation, multiple two-point correlation, and the n-point correlation. We give an asymptotic analysis and we show empirically that this algorithmic framework leads to several orders of magnitude in speedup over the naive computation, even for small datasets. We are aware of no exact algorithms for these problems which are faster either empirically or theoretically. The methodology presented in this paper is also applicable in principle to fast, large-scale parametric statistics such as RBF neural networks, mixtures of Gaussians, and Gaussian processes.

A Comparison of Image Processing Techniques for Visual Speech Recognition Applications -- Michael S. Gray, Terrence J. Sejnowski, Javier R. Movellan
We examine eight different techniques for developing visual representations in machine vision tasks. In particular we compare different versions of principal component and independent component analysis in combination with stepwise regression methods for variable selection. We found that local methods, based on the statistics of image patches, consistently outperformed global methods based on the statistics of entire images. This result is consistent with previous work on emotion and facial expression recognition. In addition, the use of a stepwise regression technique for selecting variables and regions of interest substantially boosted performance.

The Interplay of Symbolic and Subsymbolic Processes in Anagram Problem Solving -- David B. Grimes, Michael C. Mozer
Although connectionist models have provided insights into the nature of perception and motor control, connectionist accounts of higher cognition seldom go beyond an implementation of traditional symbol-processing theories. We describe a connectionist constraint satisfaction model of how people solve anagram problems. The model exploits statistics of English orthography, but also addresses the interplay of subsymbolic and symbolic computation by a mechanism that extracts approximate symbolic representations (partial orderings of letters) from subsymbolic structures and injects the extracted representation back into the model to assist in the solution of the anagram. We show the computational benefit of this extraction-injection process and discuss its relationship to conscious mental processes and working memory. We also account for experimental data concerning the difficulty of anagram solution based on the orthographic structure of the anagram string and the target word.

Permitted and Forbidden Sets in Symmetric Threshold-Linear Networks -- Richard H.R. Hahnloser, H. Sebastian Seung
Today, Lyapunov theory is still one of the most widely used concepts for describing emergent phenomena in recurrent neural networks. As a new approach to ascribing computational principles to feedback circuits, we have studied threshold-linear networks. By applying linear analysis to subnetworks composed of coactive neurons, we determine the stability of their potential steady states. We find that stability depends on two types of eigenmodes. One type determines global stability and the other type determines whether or not multistability is possible. We can prove the equivalence of our stability criteria with criteria taken from quadratic programming. Our results go beyond Lyapunov-theory. We show that there are permitted sets of neurons that can be coactive at a steady state and forbidden sets that cannot. Permitted sets are clustered in the sense that subsets of permitted sets are permitted and supersets of forbidden sets are forbidden. By viewing permitted sets as memories stored in the synaptic connections, we can provide a formulation of long-term memory that is more general than the traditional perspective of attractor networks.

Support Vector Novelty Detection Applied to Jet Engine Vibration Spectra -- Paul Hayton, Bernhard Sch{\"o}lkopf, Lionel Tarassenko, Paul Anuzis
A system has been developed to extract diagnostic information from jet engine carcass vibration data. Support Vector Machines applied to novelty detection provide a measure of how unusual the shape of a vibration signature is, by learning a representation of normality. We describe a novel method for Support Vector Machines of including information from a second class for novelty detection and give results from the application to Jet Engine vibration analysis.

Large Scale Bayes Point Machines -- Ralf Herbrich, Thore Graepel
The concept of averaging over classifiers is fundamental to the Bayesian analysis of learning. Based on this viewpoint, it has recently been demonstrated for linear classifiers that the centre of mass of version space (the set of all classifiers consistent with the training set) --- also known as the Bayes point --- exhibits excellent generalisation abilities. However, the billiard algorithm as presented in [Herbrich et al., 2000] is restricted to small sample size because it requires $\mathcal{O}( m^{2})$ of memory and $\mathcal{O}( N\cdot m^2)$ computational steps where $m$ is the number of training patterns and $N$ is the number of random draws from the posterior distribution. In this paper we present a method based on the simple perceptron learning algorithm which allows to overcome this algorithmic drawback. The method is algorithmically simple and is easily extended to the multi-class case. We present experimental results on the MNIST data set of handwritten digits which show that Bayes Point Machines are competitive with the current world champion, the support vector machine. In addition, the computational complexity of BPMs can be tuned by varying the number of samples from the posterior. Finally, rejecting test points on the basis of their (approximative) posterior probability leads to a rapid decrease in generalisation error, e.g. 0.1\% generalisation error for a given rejection rate of 10\%.

A PAC-Bayesian Margin Bound for Linear Classifiers: Why SVMs work -- Ralf Herbrich, Thore Graepel
We present a bound on the generalisation error of linear classifiers in terms of a refined margin quantity on the training set. The result is obtained in a PAC-Bayesian framework and is based on geometrical arguments in the space of linear classifiers. The new bound constitutes an exponential improvement of the so far tightest margin bound by Shawe-Taylor et al. and scales logarithmically in the inverse margin. Even in the case of less training examples than input dimensions sufficiently large margins lead to non-trivial bound values and --- for maximum margins --- to a vanishing complexity term. Furthermore, the classical margin is too coarse a measure for the essential quantity that controls the generalisation error: the volume ratio between the whole hypothesis space and the subset of consistent hypotheses. The practical relevance of the result lies in the fact that the well-known support vector machine is optimal w.r.t. the new bound only if the feature vectors are all of the same length. As a consequence we recommend to use SVMs on normalised feature vectors only --- a recommendation that is well supported by our numerical experiments on two benchmark data sets.

Hierarchical Memory-Based Reinforcement Learning -- Natalia Hernandez-Gardiol, Sridhar Mahadevan
A key challenge for reinforcement learning is how to scale up to large partially observable domains. In this paper, we show how a hierarchy of behaviors can be used to create and select among variable length short-term memories appropriate for a task. At higher levels in the hierarchy, the agent abstracts over lower-level details and looks back over a variable number of high-level decisions in time. We formalize this idea in a framework for solving partially observable, sequential decision tasks called Hierarchical Suffix Memory (HSM). HSM uses a memory-based SMDP Q-learning method to rapidly propagate delayed reward across long decision sequences. We show that the HSM framework outperforms several related reinforcement learning techniques on a realistic corridor navigation task.

Beyond Maximum Likelihood and Density Estimation: A Sample-Based Criterion for Unsupervised Learning of Complex Models -- Sepp Hochreiter, Michael C. Mozer
Many unsupervised learning procedures can be viewed as trying to bring two probability distributions into alignment, including generative models such as Gaussian mixtures and Boltzmann machines, and recoding models such as ICA and projection pursuit. We propose a novel sample-based error measure for these procedures, which works even for situations in which maximum likelihood and probability density estimation-based formulations cannot be applied, e.g., models that are nonlinear or have intractable posterior. Furthermore, our sample-based error measure avoids the difficulties of approximating a density function. We prove that with an unconstrained model, (1) our approach converges on the correct solution as the number of samples goes to infinity, and (2) the expected solution of our approach in the generative framework is the maximum likelihood solution. Finally, we evaluate our sample-based error measure via simulations of linear and nonlinear models on mixture-of-Gaussians and ICA problems. The experiments show the wide applicability and generality of our approach.

Ensemble Learning and Linear Response Theory for ICA -- Pedro A.d.F.R. H{\o}jen-S{\o}rensen, Ole Winther, Lars Kai Hansen
We propose a general framework for performing independent component analysis (ICA) which relies on ensemble learning and linear response theory known from statistical physics. We apply it to both discrete and continuous sources. For the continuous source the underdetermined (overcomplete) case is studied. The naive mean-field approach fails in this case whereas linear response theory--which gives an improved estimate of covariances--is very efficient. The examples given are for sources without temporal correlations. However, this derivation can easily be extended to treat temporal correlations. Finally, the framework offers a simple way of generating new ICA algorithms without needing to define the prior distribution of the sources explicitly.

A Silicon Primitive for Competitive Learning -- David Hsu, Miguel Figueroa, Chris Diorio
Competitive learning is a technique for training classification and clustering networks. We have designed and fabricated an 11-transistor primitive, that we term an automaximizing bump circuit, that implements competitive learning dynamics. The circuit per-forms a similarity computation, affords nonvolatile storage, and implements simultaneous local adaptation and computation. We show that our primitive is suitable for implementing competitive learning in VLSI, and demonstrate its effectiveness in a standard clustering task.

On Reversing Jensen's Inequality -- Tony Jebara, Alex Pentland
Jensen's inequality is a powerful mathematical tool and one of the workhorses in statistical learning. Its applications therein include the EM algorithm, Bayesian estimation and Bayesian inference. Jensen computes simple lower bounds on otherwise intractable quantities such as products of sums and latent log-likelihoods. This simplification then permits operations like integration and maximization. Quite often, though, upper bounds are needed as well, especially in discriminative learning. We derive and prove an efficient analytic inequality that provides such variational upper bounds. This inequality holds for latent variable mixtures of exponential family distributions and thus spans a wide range of contemporary statistical models. We also discuss applications of the upper bounds including maximum conditional likelihood, large margin discriminative models and conditional Bayesian inference. Convergence, efficiency and prediction results are shown.

Automated State Abstraction for Options using the U-Tree Algorithm -- Anders Jonsson, Andrew G. Barto
The learning of a complex task can be significantly facilitated by defining a hierarchy of subtasks. An agent can learn to choose between various temporally abstract actions, each solving an assigned subtask, to achieve the overall task. In this paper, we study hierarchical learning in the framework of options. We argue that to take full advantage of the hierarchical structure, one should perform option-specific state abstraction, and that if this is to scale to larger tasks, state abstraction should be automated. We apply the U-Tree algorithm, which automatically builds option-specific representations of the state feature space, to a simple hierarchical task. Results show that automatic state abstraction for options has a promising future.

Dopamine Bonuses -- Sham Kakade, Peter Dayan
Substantial data support a temporal difference (TD) model of dopamine (DA) neuron activity in which the cells provide a global error signal for reinforcement learning. However, in certain circumstances, DA activity seems anomalous under the TD model, responding to non-rewarding stimuli. We address these anomalies by suggesting that DA cells multiplex information about reward bonuses, including Sutton's exploration bonuses and Ng {\it et al\/}'s non-distorting shaping bonuses. We interpret this additional role for DA in terms of the unconditional attentional and psychomotor effects of dopamine, having the computational role of guiding exploration.

Hippocampally-Dependent Consolidation in a Hierarchical Model of Neocortex -- Szabolcs K{\'a}li, Peter Dayan
In memory consolidation, declarative memories which initially require the hippocampus for their recall, ultimately become independent of it. Consolidation has been the focus of numerous experimental and qualitative modeling studies, but only little quantitative exploration. We present a consolidation model in which hierarchical connections in the cortex, that initially instantiate purely semantic information acquired through probabilistic unsupervised learning, come to instantiate episodic information as well. The hippocampus is responsible for helping complete partial input patterns before consolidation is complete, while also training the cortex to perform appropriate completion by itself.

Second Order Approximations for Probability Models -- Hilbert J. Kappen, Wim Wiegerinck
In this paper, we derive a second order mean field theory for directed graphical probability models. By using an information theoretic argument it is shown how this can be done in the absence of a partition function. This method is the direct generalisation of the well-known TAP approximation for Boltzmann Machines. In a numerical example, it is shown that the method greatly improves the first order mean field approximation. The computational complexity of the method is comparable to the first order mean field method for a restricted class of graphical models.

Generalizable Singular Value Decomposition for Ill-posed Datasets -- Ulrik Kjems, Lars Kai Hansen, Stephen C. Strother
We demonstrate that statistical analysis of ill-posed data sets is subject to a bias, which can be observed when projecting independent test set examples onto a basis defined by the training examples. Because the training examples in an ill-posed data set do not fully span the signal space the observed training set variances in each basis vector will be too high compared to the average variance of the test set projections onto the same basis vectors. On basis of this understanding we introduce the Generalizable Singular Value Decomposition (GSVD) as a means to reduce this bias by re-estimation of the singular values obtained in a conventional Singular Value Decomposition, allowing for a generalization performance increase of a subsequent statistical model. We demonstrate that the algorithm succesfully corrects bias in a data set from a functional PET activation study of the human brain.

Some New Bounds on the Generalization Error of Combined Classifiers -- Vladimir Koltchinskii, Dmitriy Panchenko, Fernando Lozano
We develop a method of bounding the generalization error of a complex classifier that is a combination of simpler classifiers from a base class in terms of its empirical margin distribution. The bounds we have got generalize and improve previously known results for neural networks (Bartlett (1998)) and for convex combinations of classifiers (Schapire, Freund, Bartlett and Lee (1998)). The methods of Gaussian and empirical processes (including concentration and deviation inequalities) allowed us to obtain substantial improvement of the previous bounds in some of the problems and to understand much better the probabilistic nature of the margin type bounds.

Sparsity of Data Representation of Optimal Kernel Machine and Leave-one-out Estimator -- Adam Kowalczyk
Vapnik's result that the expectation of the generalisation error of the optimal hyperplane is bounded by the expectation of the ratio of the number of support vectors to the number of training examples is extended to a broad class of kernel machines. The class includes Support Vector Machines for soft margin classification and regression, and Regularization Networks with a variety of kernels and cost functions. We show that key inequalities in Vapnik's result and its proof become equalities once ``the classification error'' is replaced by ``the margin error'', with the latter defined as an instance with positive cost. In particular we show that expectations of the true margin error and the empirical margin error are equal, and that the sparse solutions for kernel machines are possible only if the cost function is ``partially'' in-sensitive.

Keeping Flexible Active Contours on Track using Metropolis Updates -- Trausti T. Kristjansson, Brendan J. Frey
Condensation, a form of likelihood-weighted particle filtering, has been successfully used to infer the shapes of highly constrained ``active contours'' in video sequences. However, when the contours are highly flexible (e.g. for tracking fingers of a hand), a computationally burdensome number of particles is needed to successfully approximate the contour distribution over contours at each frame in a video sequence. We compare this method to condensation using a video sequence that requires highly flexible contours, and show that the new algorithm performs dramatically better than the condensation algorithm. We discuss the incorporation of this method into the ``active contour'' framework where a shape-subspace is used to constrain shape variation.

Smart Vision Chip Fabricated Using Three Dimensional Integration Technology -- Hiroyuki Kurino, M. Nakagawa, Kang Wook Lee, Tomonori Nakamura, Yuusuke Yamada, Ki Tae Park, Mitsumasa Koyanagi
Smart vision chips have several great advantages compared to conventional image processing systems. However it also has several hindrances such as low resolution, low fill factor and large chip size. In order to overcome such drawbacks, we propose a new smart vision chip fabricated using three dimensional (3D) integration technology. In three dimensional technology, several LSI chips are stacked vertically and electrically connected with each other through vertical interconnect. The high speed, high density interconnection and high density packing can be realized by a novel 3D integration technology. 3D LSI chips has vertical layered structures and is similar to the cortex or the retina. Therefore it is a suitable structure for neuromorphic or neural network circuit. A prototype of a 3D test chip was fabricated and showed good electrical characteristics. In this paper we describe the fabrication process and fabrication results of the 3D test chips. We also present our design of 3D neuromorphic device.

Algorithms for Non-negative Matrix Factorization -- Daniel D. Lee, H. Sebastian Seung
Non-negative matrix factorization (NMF) has previously been shown to be a useful decomposition for multivariate data. Two different multiplicative algorithms for NMF are analyzed. They differ only slightly in the multiplicative factor used in the update rules. One algorithm can be shown to minimize the conventional least squares error while the other minimizes the generalized Kullback-Leibler divergence. The monotonic convergence of both algorithms can be proven using an auxiliary function analogous to that used for proving convergence of the Expectation-Maximization algorithm. The algorithms can also be interpreted as diagonally rescaled gradient descent, where the rescaling factor is optimally chosen to ensure convergence.

Color Opponency Constitutes a Sparse Representation for the Chromatic Structure of Natural Scenes -- Te-Won Lee, Thomas Wachtler, Terrence J. Sejnowski
The human visual system encodes the chromatic signals conveyed by the three types of retinal cone photoreceptors in an opponent fashion. This color opponency has been shown to constitute an efficient encoding by spectral decorrelation of the receptor signals. We analyze the spatial and chromatic structure of natural scenes by decomposing the spectral images into a set of linear basis functions such that they constitute a representation with minimal redundancy. Independent component analysis finds the basis functions that transforms the spatiochromatic data such that the outputs (activations) are statistically as independent as possible, i.e. least redundant. The resulting basis functions show strong opponency along an achromatic direction (luminance edges), along a blue-yellow direction, and along a red-blue direction. Furthermore, the resulting activations have very sparse distributions, suggesting that the use of color opponency in the human visual system achieves a highly efficient representation of colors. Our findings suggest that color opponency is a result of the properties of natural spectra and not solely a consequence of the overlapping cone spectral sensitivities.

Foundations for a Circuit Complexity Theory of Sensory Processing -- Robert A. Legenstein, Wolfgang Maass
We introduce total wire length as a salient complexity measure for judging the biological feasibility of various proposed circuit designs for sensory processing tasks, such as translation-invariant pattern recognition. Biological data show that the total wire length of neural circuits in the cortex is in fact not very large (if one puts it in relation to the number of inputs that these circuits receive). Our analysis shows that most commonly proposed circuit architectures are in fact biologically unrealistic because they would require too much total wire length. In this paper we exhibit a number of alternative circuit design techniques for computational tasks from translation- and scale-invariant sensory processing, that can be implemented within biologically realistic bounds for their total wire length. We also discuss the relationship between total wire length and other complexity measures that had previously been considered in the context of VLSI-design.

A Tighter Bound for Graphical Models -- Martijn A. R. Leisink, Hilbert J. Kappen
We present a method to bound the partition function of a Boltzmann machine neural network with any odd order polynomial. This is a direct extension of the mean field bound, which is first order. We show that the third order bound is strictly better than mean field. Additionally we show the rough outline how this bound is applicable to sigmoid belief networks. Numerical experiments indicate that an error reduction of a factor two is easily reached in the region where expansion based approximations are useful. {\bf Martijn Leisink received an Honorable Mention for NIPS Best Student Paper Award sponsored by Athene Software}

Position Variance, Recurrence and Perceptual Learning -- Zhaoping Li, Peter Dayan
Visual stimulus arrays are inevitably presented at different positions on the retina in perceptual learning tasks, even tasks that nominally require fixation. We show that inference in the face of positional variance has a structurally different quality from inference about fixed position stimuli, involving a particular, quadratic, non-linearity rather than a purely linear discrimination. We show the advantage this non-linearity has for discrimination, and suggest it as a role for recurrent connections in area V1.

Homeostasis in a Silicon Integrate and Fire Neuron -- Shih-Chii Liu, Bradley A. Minch
In this work, we explore homeostasis in a silicon integrate-and-fire neuron. The neuron adapts its firing rate over long time periods on the order of seconds or minutes so that it returns to its spontaneous firing rate after a lasting perturbation. Homeostasis is implemented via two schemes. One scheme looks at the presynaptic activity and adapts the synaptic weight depending on the presynaptic spiking rate. The second scheme adapts the synaptic ``threshold'' depending on the neuron's activity. The threshold is lowered if the neuron's activity decreases over a long time and is increased for prolonged increase in postsynaptic activity. Both these mechanisms for adaptation use floating-gate technology. The results shown here are measured from a chip fabricated in a 2-$\mu$m CMOS process.

Text Classification using String Kernels -- Huma Lodhi, John Shawe-Taylor, Nello Cristianini, Chris Watkins
We introduce a novel kernel for comparing two text documents. The kernel is an inner product in the feature space consisting of all subsequences of length $k$. A subsequence is any ordered sequence of $k$ characters occurring in the text though not necessarily contiguously. The subsequences are weighted by an exponentially decaying factor of their full length in the text, hence emphasising those occurrences which are close to contiguous. A direct computation of this feature vector would involve a prohibitive amount of computation even for modest values of k, since the dimension of the feature space grows exponentially with $k$. The paper describes how despite this fact the inner product can be efficiently evaluated by a dynamic programming technique. A preliminary experimental comparison of the performance of the kernel compared with a standard word feature space kernel is made showing encouraging results.

Constrained Independent Component Analysis -- Wei Lu, Jagath C. Rajapakse
The paper presents a novel technique of constrained independent component analysis (CICA) to eliminate indeterminacies existing in present ICA. CICA introduces constraints into ICA problem and solves the constrained optimization by using Lagrange multiplier methods. This paper shows that CICA can systematically eliminate the ICA's indeterminacies on permutation and dilation by ordering the resulted independent components in a specific manner and normalizing the demixing matrix, respectively. The experiments demonstrate ordering of components in complete as well as undercomplete ICA while providing normalized demixing processes.

Learning Curves for Gaussian Processes Regression: A Framework for Good Approximations -- D\"orthe Malzahn, Manfred Opper
Based on a statistical mechanics approach, we develop a method for approximately computing average case learning curves for Gaussian process regression models. The approximation works well in the large sample size limit and for arbitrary dimensionality of the input space. We explain how the approximation can be systematically improved and argue that similar techniques can be applied to general likelihood models.

Active Support Vector Machine Classification -- Olvi L. Mangasarian, David R. Musicant
An active set strategy is applied to the dual of a simple reformulation of the standard quadratic program of a linear support vector machine. This application generates a fast new dual algorithm that consists of solving a finite number of linear equations, with a typically large dimensionality equal to the number of points to be classified. However, by making novel use of the Sherman-Morrison-Woodbury formula, a much smaller matrix of the order of the original input space is inverted at each step. Thus, a problem with a 32-dimensional input space and 7 million points required inverting positive definite symmetric matrices of size 33-by-33 with a total running time of 96 minutes on a 400 MHz Pentium II. The algorithm requires no specialized quadratic or linear programming code, but merely a linear equation solver which is publicly available.

Weak Learners and Improved Rates of Convergence in Boosting -- Shie Mannor, Ron Meir
The problem of constructing weak classifiers for boosting algorithms is studied. Roughly speaking, an effective weak learner is one that is guaranteed to achieve, for any distribution on the data, classification performance that is sufficiently better than random guessing. We show that linear weak learners always exist; however, they are not always useful for learning purposes. Simple arguments are used to establish conditions for the existence of effective weak learners for one-dimensional problems. We then discuss extensions of the results to any dimension using methods from the theory of combinatorial discrepancy. In particular, precise conditions are established for the existence of effective linear weak learners in any dimension. Additionally, we provide improved convergence rate bounds for the generalization error in situations where the empirical error can be made small, which is exactly the situation that occurs if weak learners with guaranteed performance that are better than random guessing can be established.

Recognizing Hand-written Digits Using Hierarchical Products of Experts -- Guy Mayraz, Geoffrey E. Hinton
The product of experts learning procedure (Hinton, 2000) can discover a set of stochastic binary features that constitute a non-linear generative model of handwritten images of digits. The quality of generative models learned in this way can be assessed by learning a separate model for each class of digit and then comparing the unnormalized probabilities of test images under the 10 different class-specific models. To improve discriminative performance, each of the $10$ digit models can be given more layers of feature detectors. The layers are trained sequentially and each layer learns a generative model of the patterns of feature activities in the preceding layer. After training, each layer of feature detectors produces a separate, unnormalized log probabilty score. With three layers of feature detectors in each of the $10$ digit models, a test image produces $30$ scores which can be used as inputs to a supervised, logistic classification network that is trained on separate data. On the MNIST database, our system is comparable with current state-of-the-art discriminative methods, demonstrating that the product of experts learning procedure can produce effective generative models of high-dimensional data.

Learning Segmentation by Random Walks -- Marina Meil\u{a}, Jianbo Shi
We present a new view of image segmentation by pairwise similarities. We interpret the similarities as edge flows in a Markov random walk and study the eigenvalues and eigenvectors of the walk's transition matrix. This interpretation shows that spectral methods for clustering and segmentation have a probabilistic foundation. In particular, we prove that the Normalized Cut method arises naturally from our framework. Finally, the framework provides a principled method for learning the similarity function as a combination of features.

The Unscented Particle Filter -- Rudolph van der Merwe, Arnaud Doucet, Nando de Freitas, Eric Wan
In this paper, we propose a new particle filter based on sequential importance sampling. The algorithm uses a bank of unscented filters to obtain the importance proposal distribution. This proposal has two very ``nice'' properties. Firstly, it makes efficient use of the latest available information and, secondly, it can have heavy tails. As a result, we find that the algorithm outperforms standard particle filtering and other nonlinear filtering methods by almost an order of magnitude. This experimental finding is in agreement with the theoretical convergence proof for the algorithm. The algorithm also includes resampling and (possibly) Markov chain Monte Carlo (MCMC) steps.

A Mathematical Programming Approach to the Kernel Fisher Algorithm -- Sebastian Mika, Gunnar R{\"a}tsch, Klaus-Robert M{\"u}ller
We investigate a new kernel--based classifier: the Kernel Fisher Discriminant (KFD). A mathematical programming formulation based on the observation that KFD maximizes the {\em average margin} permits an interesting modification of the original KFD algorithm yielding the sparse KFD. We find that both, KFD and the proposed sparse KFD, can be understood in an unifying probabilistic context. Furthermore, we show connections to Support Vector Machines and Relevance Vector Machines. From this understanding, we are able to outline a very intuitive kernel--regression technique based upon the KFD algorithm. Simulations support the usefulness of our approach.

Automatic Choice of Dimensionality for PCA -- Thomas P. Minka
A central issue in principal component analysis (PCA) is choosing the number of principal components to be retained. By interpreting PCA as density estimation, this paper shows how to use Bayesian model selection to determine the true dimensionality of the data. The resulting estimate is simple to compute yet guaranteed to pick the correct dimensionality, given enough data. In simulations, it is convincingly better than cross-validation and other proposed algorithms, plus it runs much faster.

On Iterative Krylov-Dogleg Trust-Region Steps for Solving Neural Networks Nonlinear Least Squares Problems -- Eiji Mizutani, James W. Demmel
This paper describes a method of dogleg trust-region steps, or restricted Levenberg-Marquardt steps, based on a projection process onto a Krylov subspace for neural networks nonlinear least squares problems. In particular, the linear conjugate gradient (CG) method works as the inner iterative algorithm for solving the linearized Gauss-Newton normal equation, whereas the outer nonlinear algorithm repeatedly takes so-called ``Krylov-dogleg'' steps, relying only on matrix-vector multiplication without explicitly forming the Jacobian matrix or the Gauss-Newton model Hessian. That is, our iterative dogleg algorithm can reduce both operational counts and memory space by a factor of O(n) (the number of parameters) in comparison with a direct linear-equation solver. This memory-less property is useful for large-scale problems.

Sex with Support Vector Machines -- Baback Moghaddam, Ming-Hsuan Yang
Nonlinear Support Vector Machines (SVMs) are investigated for visual sex classification with low resolution ``thumbnail'' faces processed from 1,755 images from the FERET face database. The performance of SVMs is shown to be superior to traditional pattern classifiers (Linear, Quadratic, Fisher Linear Discriminant, Nearest-Neighbor) as well as more modern techniques such as Radial Basis Function (RBF) classifiers and large ensemble-RBF networks. Furthermore, the SVM performance (3.4\% error) is currently the best reported in the open literature.

Robust Reinforcement Learning -- Jun Morimoto, Kenji Doya
This paper proposes a new reinforcement learning (RL) paradigm that explicitly takes into account input disturbance as well as modeling errors. The use of environmental models in RL is quite popular for both off-line learning by simulations and for on-line action planning. However, the difference between the model and the real environment can lead to unpredictable, often unwanted results. Based on the theory of H-infinity control, we consider a differential game in which a `disturbing' agent tries to make the worst possible disturbance while a `control' agent tries to make the best control input. The problem is formulated as finding a min-max solution of a value function that takes into account the norm of the output deviation and the norm of the disturbance. We derive on-line learning algorithms for estimating the value function and for calculating the worst disturbance and the best control in reference to the value function. We tested the paradigm, which we call ``Robust Reinforcement Learning (RRL),'' in the task of inverted pendulum. In the linear domain, the policy and the value function learned by the on-line algorithms coincided with those derived analytically by the linear H-infinity control theory. For a fully nonlinear swing-up task, the control by RRL achieved robust performance against two-fold changes in the pendulum length while a standard RL control could not deal with such environmental changes.

Partially Observable SDE Models for Image Sequence Recognition Tasks -- Javier R. Movellan, Paul Mineiro, Ruth J. Williams
This paper explores a framework for recognition of image sequences using partially observable stochastic differential equation (SDE) models. Monte-Carlo importance sampling techniques are used for efficient estimation of sequence likelihoods and sequence likelihood gradients. Once the network dynamics are learned, we apply the SDE models to sequence recognition tasks in a manner similar to the way Hidden Markov models (HMMs) are commonly applied. The potential advantage of SDEs over HMMS is the use of continuous state dynamics. We present encouraging results for a video sequence recognition task in which SDE models provided excellent performance when compared to hidden Markov models.

The Use of MDL to Select among Computational Models of Cognition -- In J. Myung, Mark A. Pitt, Shaobo Zhang, Vijay Balasubramanian
How should we decide among competing explanations of a cognitive process given limited observations? The problem of model selection is at the heart of progress in cognitive science. In this paper, Minimum Description Length (MDL) is introduced as a method for selecting among computational models of cognition. We also show that differential geometry provides an intuitive understanding of what drives model selection in MDL. Finally, adequacy of MDL is demonstrated in two areas of cognitive modeling.

Probabilistic Semantic Video Indexing -- Milind R. Naphade, Igor Kozintsev, Thomas Huang
We propose a novel probabilistic framework for semantic video indexing. We define probabilistic multimedia objects (multijects) to map low-level media features to high-level semantic labels. A graphical network of such multijects (multinet) captures scene context by discovering intra-frame as well as inter-frame dependency relations between the concepts. The main contribution is a novel application of a factor graph framework to model this network. We model relations between semantic concepts in terms of their co-occurrence as well as the temporal dependencies between these concepts within video shots. Using the sum-product algorithm for approximate or exact inference in these factor graph multinets, we attempt to correct errors made during isolated concept detection by forcing high-level constraints. This results in a significant improvement in the overall detection performance.

Finding the Key to a Synapse -- Thomas Natschl{\"a}ger, Wolfgang Maass
Experimental data have shown that synapses are heterogeneous: different synapses respond with different sequences of amplitudes of postsynaptic responses to the same spike train. Neither the role of synaptic dynamics itself nor the role of the heterogeneity of synaptic dynamics for computations in neural circuits is well understood. We present in this article two computational methods that make it feasible to compute for a given synapse with known synaptic parameters the spike train that is optimally fitted to the synapse in a certain sense. One of these methods is based on dynamic programming (similar as in reinforcement learning), the other one on sequential quadratic programming. With the help of these methods one can compute for example the temporal pattern of a spike train (with a given number of spikes) that produces the largest sum of postsynaptic responses for a specific synapse. Several other applications are also discussed. To our surprise we find that most of these optimally fitted spike trains match common firing patterns of specific types of neurons that are discussed in the literature. Furthermore optimally fitted spike trains are rather specific to a certain synapse (``the key to this synapse'') in the sense that they exhibit a substantially smaller postsynaptic response on any other of the mayor types of synapses reported in the literature. This observation provides the first glimpse at a possible functional role of the specific combinations of synapse types and neuron types that was recently found in (Gupta, Wang, Markram, Science, 2000). Our computational analysis provides the platform for a better understanding of the specific role of different parameters that control synaptic dynamics, because with the help of the computational techniques that we have introduced one can now see directly how the temporal structure of the optimal spike train for a synapse depends on the individual synaptic parameters. We believe that this inverse analysis is essential for understanding the computational role of neural circuits.

Processing of Time Series by Neural Circuits with Biologically Realistic Synaptic Dynamics -- Thomas Natschl{\"a}ger, Wolfgang Maass, Eduardo D. Sontag, Anthony Zador
Experimental data show that biological synapses behave quite differently from the symbolic synapses in common artificial neural network models. Biological synapses are dynamic, i.e., their ``weight'' changes on a short time scale by several hundred percent in dependence of the past input to the synapse. In this article we explore the consequences that this synaptic dynamics entails for the computational power and adaptive capability of feedforward neural networks. Our analytical results show that the class of nonlinear filters that can be approximated by neural networks with dynamic synapses, even with just a single hidden layer of sigmoidal neurons, is remarkably rich. It contains every time invariant filter with fading memory, hence arguable every filter that is potentially useful for a biological organism. This result is robust with regard to various changes in the model for synaptic dynamics. Furthermore we show that simple gradient descent suffices to approximate a given quadratic filter by a rather small neural network with dynamic synapses. The computer simulations we performed show that in fact their performance is slightly better than that of previously considered artificial neural networks that were designed for the purpose of yielding efficient processing of temporal signals, without aiming at biological realism. We have tested dynamic networks on tasks such as the learning of a randomly chosen quadratic filter, as well as on the system identification task used in (Back and Tsoi, 1993), to illustrate the potential of our new architecture. We also address the question which synaptic parameters are essential for a network with dynamic synapses to be able to learn a particular target filter. We found that neither just plasticity in the synaptic dynamics nor just plasticity of the maximal amplitude alone yields satisfactory results. However a simple gradient descent learning algorithm that tunes both types of parameters simultaneously yields good approximation capabilities.

Active Inference in Concept Learning -- Jonathan D. Nelson, Javier R. Movellan
People are active information gatherers, constantly experimenting and seeking information relevant to their goals. A reasonable approach to active information gathering is to ask questions and conduct experiments that maximize the expected information gain, given current beliefs (Lindley, 1956; Good, 1966; MacKay, 1992). In this paper we compare the behavior of human subjects with that of an optimal information-gathering agent (infomax) in a concept induction task (Tenenbaum, 1999, 2000). Results show high consistency between subjects in their choices of numbers to sample. However infomax generally fails to predict subject's behavior when information value is measured with respect to Tenenbaum's model. It is unclear at this time whether the failure of infomax to predict human behavior is due to problems with Tenenbaum's concept induction model, or due to the fact that subjects use suboptimal heuristics (e.g., confirmatory sampling).

Learning Continuous Distributions: Simulations With Field Theoretic Priors -- Ilya Nemenman, William Bialek
Learning of a smooth but nonparametric probability density can be regularized using methods of Quantum Field Theory. We implement a field theoretic prior numerically, test its efficacy, and show that the free parameter of the theory (`smoothness scale') can be determined self consistently by the data; this forms an infinite dimensional generalization of the MDL principle. Finally, we study the implications of one's choice of the prior and the parameterization and conclude that even wrong choices can be advantageous for small data sets.

Interactive Parts Model: An Application to Recognition of On-line Cursive Script -- Predrag Neskovic, Philip C. Davis, Leon N. Cooper
In this work, we introduce an Interactive Parts (IP) model as an alternative to hidden Markov models (HMMs). We tested both models on a database of on-line cursive script. We show that implementations of HMMs and the IP model, in which all letters are assumed to have the same average width, give comparable results. However, in contrast to HMMs, the IP model can handle duration modeling without an increase in computational complexity.

Learning Sparse Image Codes using a Wavelet Pyramid Architecture -- Bruno A. Olshausen, Phil Sallee, Michael S. Lewicki
We show how a wavelet basis may be adapted to best represent natural images in terms of sparse coefficients. The wavelet basis, which may be either complete or overcomplete, is specified by a small number of spatial functions which are repeated across space and combined in a recursive fashion so as to be self-similar across scale. These functions are adapted to minimize the estimated code length under a model that assumes images are composed as a linear superposition of sparse, independent components. When adapted to natural images, the wavelet bases become selective to different spatial orientations, and they achieve a superior degree of sparsity on natural images as compared with traditional wavelet bases.

Kernel-Based Reinforcement Learning in Average-Cost Problems: An Application to Optimal Portfolio Choice -- Dirk Ormoneit, Peter Glynn
Many approaches to reinforcement learning combine neural networks or other parametric function approximators with a form of temporal-difference learning to estimate the value function of a Markov Decision Process. A significant disadvantage of those procedures is that the resulting learning algorithms are frequently unstable. In this work, we present a new, kernel-based approach to reinforcement learning which overcomes this difficulty and provably converges to a unique solution. By contrast to existing algorithms, our method can also be shown to be consistent in the sense that its costs converge to the optimal costs asymptotically. Our focus is on learning in an average-cost framework and on a practical application to the optimal portfolio choice problem.

Learning and Tracking Cyclic Human Motion -- Dirk Ormoneit, Hedvig Sidenbladh, Michael J. Black, Trevor Hastie
We present methods for learning and tracking human motion in video. We estimate a statistical model of typical activities from a large set of 3D periodic human motion data by segmenting these data automatically into a sequence of ``cycles''. Then the mean and the principal components of the cycles are computed using a new algorithm that accounts for missing information and enforces smooth transitions between cycles. The learned temporal model provides a prior probability distribution over human motions that can be used in a Bayesian framework for tracking human subjects in complex monocular video sequences and recovering their 3D motion.

Higher-Order Statistical Properties Arising from the Non-Stationarity of Natural Signals -- Lucas Parra, Clay Spence, Paul Sajda
We present evidence that several higher-order statistical properties of natural images and signals can be explained by a stochastic model which simply varies scale of an otherwise stationary Gaussian process. We discuss two interesting consequences. The first is that a variety of natural signals can be related through a common model of spherically invariant random processes, which have the attractive property that the joint densities can be constructed from the one dimensional marginal. The second is that in some cases the non-stationarity assumption and only second order methods can be explicitly exploited to find a linear basis that is equivalent to independent components obtained with higher-order methods. This is demonstrated on spectro-temporal components of speech.

Learning Switching Linear Models of Human Motion -- Vladimir Pavlovi\'c, James M. Rehg, John MacCormick
The human figure exhibits complex and rich dynamic behavior that is both nonlinear and time-varying. Effective models of human dynamics can be learned from motion capture data using switching linear dynamic system (SLDS) models. We present results for human motion synthesis, classification, and visual tracking using learned SLDS models. Since exact inference in SLDS is intractable, we present three approximate inference algorithms and compare their performance. In particular, a new variational inference algorithm is obtained by casting the SLDS model as a Dynamic Bayesian Network. Classification experiments show the superiority of SLDS over conventional HMM's for our problem domain.

Bayes Networks on Ice: Robotic Search for Antarctic Meteorites -- Liam Pedersen, Dimi Apostolopoulos, Red Whittaker
A Bayes network based classifier for distinguishing terrestrial rocks from meteorites is implemented onboard the Nomad robot. Equipped with a camera, spectrometer and eddy current sensor, this robot searched the ice sheets of Antarctica and autonomously made the first robotic identification of a meteorite, in January 2000 at the Elephant Moraine. This paper discusses rock classification from a mobile robotic platform, and describes the system onboard Nomad.

Redundancy and Dimensionality Reduction in Sparse-Distributed Representations of Natural Objects in Terms of Their Local Features -- Penio S. Penev
Low-dimensional representations are key to solving problems in high-level vision, such as face compression and recognition. Factorial coding strategies for reducing the redundancy present in natural images on the basis of their second-order statistics have been successful in accounting for both psychophysical and neurophysiological properties of early vision. Class-specific representations are presumably formed later, at the higher-level stages of cortical processing. Here we show that when retinotopic factorial codes are derived for ensembles of natural objects, such as human faces, not only redundancy, but also dimensionality is reduced. We also show that objects are built from parts in a non-Gaussian fashion which allows these local-feature codes to have dimensionalities that are substantially lower than the respective Nyquist sampling rates.

Fast Training of Support Vector Classifiers -- Fernando P\'erez-Cruz, Pedro L. Alarc\'on-Diana, Angel Navia-V\'azquez, Antonio Art\'es-Rodr\'{\i}guez
In this communication we present a new algorithm for solving Support Vector Classifiers (SVC) with large training data sets. The new algorithm is based on an Iterative Re-Weighted Least Squares procedure in order to optimize the SVC. Moreover, a novel selection strategy for the working set is presented, different from the existing ones. The selection strategy randomly chooses the working set among the training samples that do not fulfill the stopping criteria. The validity of both approaches, the optimization procedure and sample selection strategy, are shown by means of computer experiments using well-known data sets.

The Use of Classifiers in Sequential Inference -- Vasin Punyakanok, Dan Roth
We study the problem of combining the outcomes of several different classifiers in a way that provides a coherent inference that satisfies some constraints. In particular, we develop two general approaches for an important subproblem -- identifying phrase structure. The first is a Markovian approach that extends standard HMMs to allow the use of a rich observation structure and of general classifiers to model state-observation dependencies. The second is an extension of constraint satisfaction formalisms. We develop efficient combination algorithms under both models and study them experimentally in the context of shallow parsing of natural language.

Occam's Razor -- Carl Edward Rasmussen, Zoubin Ghahramani
Confounding the prior over data sets with the complexity of the machinery used to implement the model causes an Occam's Razor behaviour in non-parametric models. If instead the two complexities are carefully separated the effects of Occam's Razor is reduced and the limit of complex models may be preferred. A simple example illustrates the idea and the practical and theoretical implications are discussed.

One Microphone Source Separation -- Sam T. Roweis
Source separation, or computational auditory scene analysis, attempts to extract individual acoustic objects from input which contains a mixture of sounds from different sources, altered by the acoustic environment. Unmixing algorithms such as ICA and its extensions recover sources by reweighting multiple observation sequences, and thus cannot operate when only a single observation signal is available. I present a technique called {\em refiltering} which recovers sources by a nonstationary reweighting (``masking'') of frequency sub-bands from a single recording, and argue for the application of statistical algorithms to learning this masking function. I present results of a simple factorial HMM system which learns on recordings of single speakers and can then separate mixtures using only one observation signal by computing the masking during inference and then refiltering.

Using Free Energies to Represent Q-values in a Multiagent Reinforcement Learning Task -- Brian Sallans, Geoffrey E. Hinton
The problem of reinforcement learning in large factored Markov decision processes is explored. The Q-value of a state-action pair is approximated by the free energy of a product of experts network. Network parameters are learned on-line using a modified SARSA algorithm which minimizes the inconsistency of the Q-values of consecutive state-action pairs. Actions are chosen based on the current value estimates by fixing the current state and sampling actions from the network using Gibbs sampling. The algorithm is tested on a co-operative multi-agent task. The product of experts model is found to perform comparably to table-based Q-learning for small instances of the task, and continues to perform well when the problem becomes too large for a table-based representation. {\bf Brian Sallans received an Honorable Mention for NIPS Best Student Paper Award sponsored by Athene Software}

Minimum Bayes Error Feature Selection for Continuous Speech Recognition -- George Saon, Mukund Padmanabhan
We consider the problem of designing a linear transformation which projects the n-dimensional features of a classifier onto p-dimensional features ($p$ less than $n$) such as to achieve minimum Bayes error (or probability of misclassification). Two avenues will be explored: the first is to maximize the average divergence between the class densities and the second is to minimize the union Bhattacharyya bound on the error in the projected space. While both approaches yield similar performance in practice, they outperform standard LDA features and show a 10\% relative improvement in the word error rate over state-of-the-art cepstral features on a large vocabulary telephony speech recognition task.

Periodic Component Analysis: An Eigenvalue Method for Representing Periodic Structure in Speech -- Lawrence K. Saul, Jont B. Allen
An eigenvalue method is developed for analyzing periodic structure in speech. Signals are analyzed by a matrix diagonalization reminiscent of methods for principal component analysis and independent component analysis. Our method---called periodic component analysis---uses constructive interference to enhance periodic components of the frequency spectrum and destructive interference to cancel noise. The front end emulates certain aspects of auditory processing, such as cochlear filtering, nonlinear compression, and insensitivity to phase, with the aim of matching the robustness of human listeners. The method avoids the inefficiencies of autocorrelation at the pitch period: it does not require long delay lines, and it correlates signals at a clock rate on the order of the actual pitch, as opposed to the original sampling rate. We derive its cost function and present some experimental results.

Spike-Timing-Dependent Learning for Oscillatory Networks -- Silvia Scarpetta, Zhaoping Li, John Hertz
We apply to oscillatory networks a class of learning rules in which synaptic weights change proportional to pre- and post-synaptic activity, with a kernel $A(\tau)$ measuring the effect for a postsynaptic spike a time $\tau$ after the presynaptic one. The resulting synaptic matrices have an outer-product form in which the oscillating patterns are represented as complex vectors. In a simple model, the even part of $A(\tau)$ enhances the resonant response to learned stimulus by reducing the effective damping, while the odd part determines the frequency of oscillation. We relate our model to the olfactory cortex and hippocampus and their presumed roles in forming associative memories and input representations.

Universality and Individuality in a Neural Code -- Elad Schneidman, Naama Brenner, Naftali Tishby, Robert R. de Ruyter van Steveninck, William Bialek
The problem of neural coding is to understand how sequences of action potentials (spikes) are related to sensory stimuli, motor outputs, or (ultimately) thoughts and intentions. One clear question is whether the same coding rules are used by different neurons, or by corresponding neurons in different individuals. We present a quantitative formulation of this problem using ideas from information theory, and apply this approach to the analysis of experiments in the fly visual system. We find significant individual differences in the structure of the code, particularly in the way that temporal patterns of spikes are used to convey information beyond that available from variations in spike rate. On the other hand, all the flies in our ensemble exhibit a high coding efficiency, so that every spike carries the same amount of information in all the individuals. Thus the neural code has a quantifiable mixture of individuality and universality.

Machine Learning for Video-Based Rendering -- Arno Sch{\"o}dl, Irfan Essa
We present techniques for rendering and animation of realistic scenes by analyzing and training on short video sequences. This work extends the new paradigm for computer animation, video textures, which uses recorded video to generate novel animations by replaying the video samples in a new order. Here we concentrate on video sprites, which are a special type of video texture. In video sprites, instead of storing whole images, the object of interest is separated from the background and the video samples are stored as a sequence of alpha-matted sprites with associated velocity information. They can be rendered anywhere on the screen to create a novel animation of the object. We present methods to create such animations by finding a sequence of sprite samples that is both visually smooth and follows a desired path. To estimate visual smoothness, we train a linear classifier to estimate visual similarity between video samples. If the motion path is known in advance, we use beam search to find a good sample sequence. We can specify the motion interactively by precomputing the sequence cost function using Q-learning.

The Kernel Trick for Distances -- Bernhard Sch{\"o}lkopf
A method is described which, like the kernel trick in support vector machines (SVMs), lets us generalize distance-based algorithms to operate in feature spaces, usually nonlinearly related to the input space. This is done by identifying a class of kernels which can be represented as norm-based distances in Hilbert spaces. It turns out that common kernel algorithms, such as SVMs and kernel PCA, are actually really distance based algorithms and can be run with that class of kernels, too. As well as providing a useful new insight into how these algorithms work, the present work can form the basis for conceiving new algorithms.

Natural Sound Statistics and Divisive Normalization in the Auditory System -- Odelia Schwartz, Eero P. Simoncelli
We explore the statistical properties of natural sound stimuli pre-processed with a bank of linear filters. The responses of such filters exhibit a striking form of statistical dependency, in which the response variance of each filter grows with the response amplitude of filters tuned for nearby frequencies. These dependencies may be substantially reduced using an operation known as divisive normalization, in which the response of each filter is divided by a weighted sum of the rectified responses of other filters. The weights may be chosen to maximize the independence of the normalized responses for an ensemble of natural sounds. We demonstrate that the resulting model accounts for non-linearities in the response characteristics of the auditory nerve, by comparing model simulations to electrophysiological recordings. In previous work (NIPS*98) we demonstrated that an analogous model derived from the statistics of natural images accounts for non-linear properties of neurons in primary visual cortex. Thus, divisive normalization appears to be a generic mechanism for eliminating a type of statistical dependency that is prevalent in natural signals of different modalities.

Balancing Multiple Sources of Reward in Reinforcement Learning -- Christian R. Shelton
For many problems which would be natural for reinforcement learning, the reward signal is not a single scalar value but has multiple scalar components. Examples of such problems include agents with multiple goals and agents with multiple users. Creating a single reward value by combining the multiple components can throw away vital information and can lead to incorrect solutions. We describe the multiple reward source problem and discuss the problems with applying traditional reinforcement learning. We then present an new algorithm for finding a solution and results on simulated environments.

An Information Maximization Approach to Overcomplete and Recurrent Representations -- Oren Shriki, Haim Sompolinsky, Daniel D. Lee
The principle of maximizing mutual information is applied to learning overcomplete and recurrent representations. The underlying probabilistic model consists of a network of input units driving a larger number of output units with recurrent interactions. In the limit of zero noise, the network is deterministic and the mutual information can be related to the entropy of the output units. Maximizing this entropy with respect to both the feedforward connections as well as the recurrent interactions results in simple learning rules for both sets of parameters. The conventional independent components (ICA) learning algorithm can be recovered as a special case where there is an equal number of output units and no recurrent connections. The application of these new learning rules is illustrated on a simple two-dimensional input example.

Development of Hybrid Systems: Interfacing a Silicon Neuron to a Leech Heart Interneuron -- Mario F. Simoni, Gennady S. Cymbalyuk, Michael Q. Sorensen, Ronald L. Calabrese, Stephen P. DeWeerth
We have developed a silicon neuron that is inspired by a mathematical model of the leech heartbeat (HN) interneuron. The temporal and ionic current behaviors of this silicon neuron are close to that of the living cell. Because of this similarity we were able to interface this silicon neuron to a living HN cell using a dynamic clamp technique. We present data showing dynamic behaviors of the hybrid half-center oscillator.

FaceSync: A Linear Operator for Measuring Synchronization of Video Facial Images and Audio Tracks -- Malcolm Slaney, Michele Covell
FaceSync is an optimal linear algorithm that finds the degree of synchronization between the audio and image recordings of a human speaker. Using canonical correlation, it finds the best direction to combine all the audio and image data, projecting them onto a single axis. FaceSync uses Pearson's correlation to measure the degree of synchronization between the audio and image data. We derive the optimal linear transform to combine the audio and visual information and describe an implementation that avoids the numerical problems caused by computing the correlation matrices.

The Early Word Catches the Weights -- Mark A. Smith, Garrison W. Cottrell, Karen L. Anderson
The strong correlation between the frequency of words and their naming latency has been well documented. However, as early as 1973, the Age of Acquisition (AoA) of a word was alleged to be the actual variable of interest, but these studies seem to have been ignored in most of the literature. Recently, there has been a resurgence of interest in AoA. While some studies have shown that frequency has no effect when AoA is controlled for, more recent studies have found an independent contribution of frequency and AoA. Connectionist models have repeatedly shown strong effects of frequency, but little attention has been paid to whether they can also show AoA effects. Indeed, several researchers have explicitly claimed that they cannot show AoA effects. In this work, we explore these claims using a simple feed forward neural network. We find a very significant contribution of AoA to naming latency, as well as the conditions under which frequency provides an independent contribution.

Sparse Greedy Gaussian Process Regression -- Alex J. Smola, Peter Bartlett
We present a simple sparse greedy technique to approximate the maximum a posteriori estimate of Gaussian Processes with much improved scaling behaviour in the sample size $m$. In particular, computational requirements are $O(n^2 m)$, storage is $O(n m)$, the cost for prediction is $O(n)$ and the cost to compute confidence bounds is $O(n m)$, where $n \ll m$. We show how to compute a stopping criterion, give bounds on the approximation error, and show applications to large scale problems.

Regularization with Dot-Product Kernels -- Alex J. Smola, Zolt\'an L. \'Ov\'ari, Robert C. Williamson
In this paper we give necessary and sufficient conditions under which kernels of dot product type $k(x, y) = k(x \cdot y)$ satisfy Mercer's condition and thus may be used in Support Vector Machines (SVM), Regularization Networks (RN) or Gaussian Processes (GP). In particular, we show that if the kernel is analytic (i.e.\ can be expanded in a Taylor series), all expansion coefficients have to be nonnegative. We give an explicit functional form for the feature map by calculating its eigenfunctions and eigenvalues.

APRICODD: Approximate Policy Construction Using Decision Diagrams -- Robert St-Aubin, Jesse Hoey, Craig Boutilier
We propose a method of approximate dynamic programming for Markov decision processes (MDPs) using algebraic decision diagrams (ADDs). We produce near-optimal value functions and policies with much lower time and space requirements than exact dynamic programming. Our method reduces the sizes of the intermediate value functions generated during value iteration by replacing the values at the terminals of the ADD with ranges of values. Our method is demonstrated on a class of large MDPS (with up to 2 billion states), and we compare the results with the optimal value functions.

Four-legged Walking Gait Control Using a Neuromorphic Chip Interfaced to a Support Vector Learning Algorithm -- Susanne Still, Bernhard Sch{\"o}lkopf, Klaus Hepp, Rodney J. Douglas
To control the walking gaits of a four-legged robot we present a novel neuromorphic VLSI chip that coordinates the relative phasing of the robot's legs similar to how spinal Central Pattern Generators are believed to control vertebrate locomotion. The chip controls the leg movements by driving motors with time varying voltages which are the outputs of a small network of coupled oscillators. The characteristics of the chip's output voltages depend on a set of input parameters. The relationship between input parameters and output voltages can be computed analytically for an idealized system. In practice, however, this idealized relationship is only approximately true. Fine tuning of the chip's input parameters is done autonomously by the robotic system, using an unsupervised Support Vector learning algorithm that was introduced recently. The machine learns from (unlabeled) examples how to set the parameters to the chip in order to obtain a desired motor behavior.

Kernel Expansions with Unlabeled Examples -- Martin Szummer, Tommi S. Jaakkola
Modern classification applications necessitate supplementing the few available labeled examples with unlabeled examples to improve classification performance. We present a new tractable algorithm for exploiting unlabeled examples in discriminative classification. This is achieved essentially by expanding the input vectors into longer feature vectors via the unlabeled examples. The resulting classification method can be interpreted as a discriminative kernel density estimate and is readily trained via the EM algorithm, which in this case is both discriminative and achieves the optimal solution. We provide, in addition, a purely discriminative formulation of the estimation problem by appealing to the maximum entropy framework. We demonstrate that the proposed approach requires very few labeled examples for high classification accuracy.

Analysis of Bit Error Probability of Direct-Sequence CDMA Multiuser Demodulators -- Toshiyuki Tanaka
We analyze the bit error probability of multiuser demodulators for direct-sequence binary phase-shift-keying (DS/BPSK) CDMA channel with additive gaussian noise. The problem of multiuser demodulation is cast into the finite-temperature decoding problem, and replica analysis is applied to evaluate the performance of the resulting MPM (Marginal Posterior Mode) demodulators, which include the optimal demodulator and the MAP demodulator as special cases. An approximate implementation of demodulators is proposed using analog-valued Hopfield model as a naive mean-field approximation to the MPM demodulators, and its performance is also evaluated by the replica analysis. Results of the performance evaluation shows effectiveness of the optimal demodulator and the mean-field demodulator compared with the conventional one, especially in the cases of small information bit rate and low noise level.

Noise Suppression Based on Neurophysiologically-motivated SNR Estimation for Robust Speech Recognition -- J\"urgen Tchorz, Michael Kleinschmidt, Birger Kollmeier
A novel noise suppression scheme is proposed which is based on a neurophysiologically-motivated estimation of the local signal-to-noise ratio (SNR) in different frequency channels. For SNR-estimation, the input signal is transformed into so-called Amplitude Modulation Spectrograms (AMS), which represent both spectral and temporal characteristics of the respective analysis frame, and which imitate the representation of modulation frequencies in higher stages of the mammalian auditory system. A neural network is used to analyse AMS patterns generated from noisy speech and provides estimates of the local SNR. Noise suppression is achieved by attenuating frequency channels according to their SNR. The noise suppression algorithm is evaluated in speaker-independent digit recognition experiments and compared to noise suppression by Spectral Subtraction.

Rate-coded Restricted Boltzmann Machines for Face Recognition -- Yee Whye Teh, Geoffrey E. Hinton
We describe a neurally-inspired, unsupervised learning algorithm that builds a non-linear generative model for pairs of face images from the same individual. Individuals are then recognized by finding the highest relative probability pair among all pairs that consist of a test image and an image whose identity is known. Our method compares favorably with other methods in the literature. The generative model consists of a single layer of rate-coded, non-linear feature detectors and it has the property that, given a data vector, the true posterior probability distribution over the feature detector activities can be inferred rapidly without iteration or approximation. The weights of the feature detectors are learned by comparing the correlations of pixel intensities and feature activations in two phases: When the network is observing real data and when it is observing reconstructions of real data generated from the feature activations.

Structure Learning in Human Causal Induction -- Joshua B. Tenenbaum, Thomas L. Griffiths
We use graphical models to explore the question of how people learn simple causal relationships form data. The two leading psychological theories can both be seen as estimating the parameters of a fixed graph. We argue that a complete account of causal induction should also consider how people learn the underlying causal graph structure, and we propose to model this inductive process as a Bayesian inference. Our argument is supported through the discussion of three data sets.

Sparse Kernel Principal Component Analysis -- Michael E. Tipping
`Kernel' principal component analysis (PCA) is an elegant nonlinear generalisation of the popular linear data analysis method, where a kernel function implicitly defines a nonlinear transformation into a feature space wherein standard PCA is performed. Unfortunately, the technique is not `sparse', since the components thus obtained are expressed in terms of kernels associated with every training vector. This paper shows that by approximating the covariance matrix in feature space by a reduced number of example vectors, using a maximum-likelihood approach, we may obtain a highly sparse form of kernel PCA without loss of effectiveness.

Data Clustering by Markovian Relaxation and the Information Bottleneck Method -- Naftali Tishby, Noam Slonim
We introduce a new, non-parametric and principled, distance based clustering method. This method combines a pairwise based approach with a vector-quantization method which provide a meaningful interpretation to the resulting clusters. The idea is based on turning the distance matrix into a Markov process and then examine the decay of mutual-information during the relaxation of this process. The clusters emerge as quasi-stable structures during this relaxation, and then are extracted using the information bottleneck method. These clusters capture the information about the initial point of the relaxation in the most effective way. The method can cluster data with no geometric or other bias and makes no assumption about the underlying distribution.

Adaptive Object Representation with Hierarchically-Distributed Memory Sites -- Bosco S. Tjan
Most conventional theories of object recognition assume that within a single visual-processing pathway only one form of representation is derived and used to recognize objects. Versatility of the visual system comes from having multiple visual-processing pathways, each supporting a particular class of objects or recognition tasks. We propose a theoretically simpler alternative, capable of explaining the same set of data and more. A single primary visual-processing pathway, loosely modular, is assumed. Memory modules are attached to sites along this pathway. Object-identity decision is made independently at each site. A site's response time is a monotonic-decreasing function of its confidence regarding its decision. The system's response is the first-arriving response from any site. The effective representation of such a system, determined behaviorally, appears to be self-adapting and specialized for different tasks and stimuli. This is merely a reflection of the decisions being made at the appropriate level of abstraction.

Active Learning for Parameter Estimation in Bayesian Networks -- Simon Tong, Daphne Koller
Bayesian networks are graphical representations of probability distributions. In virtually all of the work on learning these networks, the assumption is that we are presented with a data set consisting of randomly generated instances from the underlying distribution. In many situations, however, we also have the option of active learning, where we have the possibility of guiding the sampling process by querying for certain types of samples. This paper addresses the problem of estimating the parameters of Bayesian networks in an active learning setting. We provide a theoretical framework for this problem, and an algorithm that chooses which active learning queries to generate based on the model learned so far. We present experimental results showing that our active learning algorithm can significantly reduce the need for training data in many situations.

Mixtures of Gaussian Processes -- Volker Tresp
We use several Gaussian processes to model the dependencies between discrete or continuous variables of a Bayesian network. We consider the case that training data are complete or contain latent variables. As a special case of the latter we consider the Gaussian process mixtures of experts model and derive EM Fisher-scoring learning rules. This model is useful for applications where the bandwidth of the map is input dependent. We experimentally verify the benefits of the Gaussian process mixtures of experts model.

Bayesian Video Shot Segmentation -- Nuno Vasconcelos, Andrew Lippman
Prior knowledge about video structure can be used both as a means to improve the performance of content analysis and to extract features that allow semantic classification. We introduce statistical models for two important components of this structure, shot duration and activity, and demonstrate the usefulness of these models by introducing a Bayesian formulation for the shot segmentation problem. The new formulations is shown to extend standard thresholding methods in an adaptive and intuitive way, leading to improved segmentation accuracy.

Error-correcting Codes on a Bethe-like Lattice -- Renato Vicente, David Saad, Yoshiyuki Kabashima
We analyze Gallager codes by employing a simple mean-field approximation that distorts the model geometry and preserves important interactions between sites. The method naturally recovers the probability propagation decoding algorithm as a minimization of a proper free-energy. We find a thermodynamic phase transition that coincides with information theoretical upper-bounds and explain the practical code performance in terms of the free-energy landscape.

Whence Sparseness? -- Carl van Vreeswijk
It has been shown that the receptive fields of simple cells in V1 can be explained by assuming optimal encoding, provided that an extra constraint of sparseness is added. This finding suggests that there is a reason, independent of optimal representation, for sparseness. However this work used an ad hoc model for the noise. Here I show that, if a biologically more plausible noise model, describing neurons as Poisson processes, is used sparseness does not have to be added as a constraint. Thus I conclude that sparseness is not a feature that evolution has striven for, but is simply the result of the evolutionary pressure towards an optimal representation.

Tree-Based Modeling and Estimation of Gaussian Processes on Graphs with Cycles -- Martin J. Wainwright, Erik B. Sudderth, Alan S. Willsky
We present the embedded trees algorithm, an iterative technique for estimation of Gaussian processes defined on arbitrary graphs. By exactly solving a series of modified problems on embedded spanning trees, it computes the conditional means with an efficiency comparable to or better than other techniques. Unlike other methods, the ET algorithm also computes exact error covariances. The algorithm is especially efficient in application to ``near-trees'', in which removing a small number of edges reveals an embedded tree. In this context, we demonstrate that near-tree graphs provide a significant increase in modeling power, with only a minor increase in estimation complexity.

Algebraic Information Geometry for Learning Machines with Singularities -- Sumio Watanabe
In the previous paper in NIPS*99, we proved the fundamental relation between algebraic geometry and the asymptotic stochastic complexity of a hierarchical learning machine. In this paper, we show that singularities in the manifold of parameters strongly affect and improve the learning curve, even if singularities are not contained in the neighborhood of the optimal parameter. It is proven based on algebraic geometry that a singular point makes the variance of Bayes learning smaller than regular points. The Bayes a posteriori distribution automatically selects the optimal singular point that minimizes the sum of the bias and the variance, resulting that a hierarchical learning machine attains the best performance even if the number of training samples are not sufficiently large.

Feature Selection for SVMs -- Jason Weston, Sayan Mukherjee, Olivier Chapelle, Massimiliano Pontil, Tomaso Poggio, Vladimir Vapnik
We introduce a method of feature selection for Support Vector Machines. The method is based upon finding those features which minimize bounds on the leave-one-out error, which can be efficiently computed via gradient descent. The resulting algorithms are shown to be superior to some standard feature selection algorithms on both toy data and real-life problems of face recognition, pedestrian detection and analyzing DNA microarray data.

On a Connection between Kernel PCA and Metric Multidimensional Scaling -- Christopher K. I. Williams
In this paper we show that the kernel PCA algorithm of Sch{\"o}lkopf et al.~(1998) can be viewed as a form of metric multidimensional scaling (MDS) when the kernel function $k({\bf x}, {\bf y})$ is isotropic, i.e.\ it depends only on $||{\bf x} - {\bf y}||$. This leads to a metric MDS algorithm where the desired configuration of points is found via the solution of an eigenproblem rather than through the iterative optimization of the stress objective function. The question of kernel choice is also discussed.

Using the Nystr{\"o}m Method to Speed Up Kernel Machines -- Christopher K. I. Williams, Matthias Seeger
A major problem for kernel-based predictors (such as Support Vector Machines and Gaussian processes) is that the computational complexity required to find the solution scales as $O(n^3)$, where $n$ is the number of training examples. We show that an approximation to the eigendecomposition of the Gram matrix can be computed by the Nystr{\"o}m method (which is used for the numerical solution of eigenproblems). This is achieved by carrying out an eigendecomposition on a smaller system of size $m < n$, and then expanding the results back up to $n$ dimensions. The computational complexity of this scheme is $O(m^2 n)$. We report experiments on the USPS data set and show that we can set $m \ll n$ without any significant decrease in the accuracy of the solution.

Computing with Finite and Infinite Networks -- Ole Winther
Using statistical mechanics results, I calculate learning curves (average generalization error) for Gaussian processes (GPs) and Bayesian neural networks (NNs) used for regression. Applying the results to learning a teacher defined by a two-layer network, I can directly compare GP and Bayesian NN learning. I find that a GP in general requires ${\cal O}(d^s)$-training examples to learn input features of order $s$ ($d$ is the input dimension), whereas a NN can learn the task with order the number of adjustable weights training examples. Since a GP can be considered as an infinite NN, the results show that even in the Bayesian approach, it is important to limit the complexity of the learning machine. The theoretical findings are confirmed in simulations with analytical GP learning and a NN mean field algorithm.

Stagewise Processing in Error-correcting Codes and Image Restoration -- K. Y. Michael Wong, Hidetoshi Nishimori
We introduce stagewise processing in error-correcting codes and image restoration, by extracting information from the former stage and using it selectively to improve the performance of the latter one. Both mean-field analysis using the cavity method and simulations show that it has the advantage of being robust against uncertainties in hyperparameter estimation.

Learning Winner-take-all Competition Between Groups of Neurons in Lateral Inhibitory Networks -- Xiaohui Xie, Richard H.R. Hahnloser, H. Sebastian Seung
It has long been known that lateral inhibition in neural networks can lead to a winner-take-all competition, so that only a single neuron is active at a steady state. Here we show how to organize lateral inhibition so that groups of neurons compete to be active. Given a collection of potentially overlapping groups, the inhibitory connectivity is set by a formula that can be interpreted as arising from a simple learning rule. Our analysis demonstrates that such inhibition generally results in winner-take-all competition between the given groups, with the exception of some degenerate cases. In a broader context, the network serves as a particular illustration of the general distinction between permitted and forbidden sets, which was introduced recently. From this viewpoint, the computational function of our network is to store and retrieve memories as permitted sets of coactive neurons.

Generalized Belief Propagation -- Jonathan S. Yedidia, William T. Freeman, Yair Weiss
Belief propagation (BP) was only supposed to work for tree-like networks but works surprisingly well in many applications involving networks with loops, including turbo codes. However, there has been little understanding of the algorithm or the nature of the solutions it finds for general graphs. We show that BP can only converge to a stationary point of an approximate free energy, known as the Bethe free energy in statistical physics. This result characterizes BP fixed-points and makes connections with variational approaches to approximate inference. More importantly, our analysis lets us build on the progress made in statistical physics since Bethe's approximation was introduced in 1935. Kikuchi and others have shown how to construct more accurate free energy approximations, of which Bethe's approximation is the simplest. Exploiting the insights from our analysis, we derive generalized belief propagation (GBP) versions of these Kikuchi approximations. These new message passing algorithms can be significantly more accurate than ordinary BP, at an adjustable increase in complexity. We illustrate such a new GBP algorithm on a grid Markov network and show that it gives much more accurate marginal probabilities than those found using ordinary BP.

A Gradient-Based Boosting Algorithm for Regression Problems -- Richard S. Zemel, Toniann Pitassi
In adaptive boosting, several weak learners trained sequentially are combined to boost the overall algorithm performance. Recently adaptive boosting methods for classification problems have been derived as gradient descent algorithms. This formulation justifies key elements and parameters in the methods, all chosen to optimize a single common objective function. We propose an analogous formulation for adaptive boosting of regression problems, utilizing a novel objective function that leads to a simple boosting algorithm. We prove that this method reduces training error, and compare its performance to other regression methods.

Divisive and Subtractive Mask Effects: Linking Psychophysics and Biophysics -- Barbara Zenger, Christof Koch
We describe an analogy between psychophysically measured effects in contrast masking, and the behavior of a simple integrate-and-fire neuron that receives time-modulated inhibition. In the psychophysical experiments, we tested observers ability to discriminate contrasts of peripheral Gabor patches in the presence of collinear Gabor flankers. The data reveal a complex interaction pattern that we account for by assuming that flankers provide divisive inhibition to the target unit for low target contrasts, but provide subtractive inhibition to the target unit for higher target contrasts. A similar switch from divisive to subtractive inhibition is observed in an integrate-and-fire unit that receives inhibition modulated in time such that the cell spends part of the time in a high-inhibition state and part of the time in a low-inhibition state. The similarity between the effects suggests that one may cause the other. The biophysical model makes testable predictions for physiological single-cell recordings.

Regularized Winnow Methods -- Tong Zhang
In theory, the Winnow multiplicative update has certain advantages over the Perceptron additive update when there are many irrelevant attributes. Recently, there has been much effort on enhancing the Perceptron algorithm by using regularization, leading to a class of linear classification methods called support vector machines. Similarly, it is also possible to apply the regularization idea to the Winnow algorithm, which gives methods we call regularized Winnows. We show that the resulting methods compare with the basic Winnows in a similar way that a support vector machine compares with the Perceptron. We investigate algorithmic issues and learning properties of the derived methods. Some experimental results will also be provided to illustrate different methods.

Convergence of Large Margin Separable Linear Classification -- Tong Zhang
Large margin linear classification methods have been successfully applied to many applications. For a linearly separable problem, it has been shown that under appropriate assumptions, the expected misclassification error of the computed ``optimal hyperplane'' approaches zero at a rate proportional to the inverse training sample size. This rate is usually characterized by the margin and the maximum norm of the input data. In this paper, we argue that another quantity, namely the robustness of the input data distribution, also plays an important role in characterizing the convergence behavior of expected misclassification error. Based on this concept of robustness, we show that the expected misclassification error can converge exponentially in the training sample size.

We have attempted to ensure that all information is correct, but we cannot guarantee it. Please send website-related comments and corrections to: (for webpage content-related issues) Alexander Gray, Carnegie Mellon University, or (for server-related or script-related issues) Xin Wang, Oregon State University.