Click here for the conference pre-proceedings.
Robotic Mine Mapping
I'll be presenting work by CMU and friends on the mapping of underground mines. In particular, I'll address two key components of our research. Firstly, in order to produce accurate large-scale maps of these subterranean spaces, we have extended current mapping approaches to deal more robustly with loops. Secondly, it is of greatest use to us if our robots are able to venture into these spaces without needing tethers. To this end, we have developed a robotic system capable of autonomously traversing mine environments.
Geometric Clustering using the Information Bottleneck method,
Mark V. Albert
I will very briefly summarize the NIPS-presented research results of Still,
Bialek, and Bottou on deriving K-means clustering using the information
bottleneck framework. This work is motivated by concepts which have been
applied to areas as distinct as semantic clustering of English words,
document categorization, neural coding, and spectral analysis. A short
introduction of this framework will be given in the context of lossy source
compression. Results of the derivation of K-means will be mentioned, and
comparitive results will be given between traditional K-means and its
derived counterpart. In addition to conceptual unification of the
algorithms mentioned, the derived iterative approaches appear to be more
robust and efficient than their classical counterparts. In 10 minutes, only
key concepts will be introduced.
Hierarchical Topic Models and the Nested Chinese Restaurant Process
by David M. Blei, Thomas L. Griffiths, Michael I. Jordan, Joshua B. Tenenbaum
We address the problem of learning topic hierarchies from data. The model selection problem in this domain is daunting which of the large collection of possible trees to use? We take a Bayesian approach, gen-erating an appropriate prior via a distribution on partitions that we refer to as the nested Chinese restaurant process. This nonparametric prior al-lows arbitrarily large branching factors and readily accommodates grow-ing data collections. We build a hierarchical topic model by combining this prior with a likelihood that is based on a hierarchical variant of latent Dirichlet allocation. We illustrate our approach on simulated data and with an application to the modeling of NIPS abstracts.
Back to the Main Page
Pradeep Ravikumar Last modified: Thu Jan 29 22:17:09 EST 2004