Realizability: Who Needs It
Friday, September 17th, 1999 from 12-1 pm in WeH 4623.
Presented by Andrej Bauer,
In this talk I will attempt to explain to a general computer science audience the main concepts behind the mathematical idea of ``realizability''. This subject will be of particular interest to those who are concerned with various aspects of scientific computation.
In a realizability model we take as true those statements which can be demonstrated by programs. A program which demonstrates a statement is called a `realizer' for the statement. We can think of realizers as evidence of a statement's truth. For example, an implication ``if A then B'' is considered true if there is a program that transforms evidence of A into evidence of B. Depending on what programs we take into account (do we allow parallel execution, exceptions, and timeouts?) we get different realizability models, each having its own set of valid rules of logic.
If we want to perform computations on mathematical objects, such as real numbers, differential equations, and topological spaces, we construct them by using the logic of a realizability model, and look at the realizers so obtained. They correspond to algorithms and data structures that best capture the nature of the desired mathematical objects.
As an example, I will show how the construction of real numbers in a realizability model suggests an implementation of arbitrary precision arithmetic. What we get is almost the IEEE floating point arithmetic. From this we draw the conclusion that IEEE almost got its floating point standard right.
Interactive Control of Rigid Body Motion
Friday, October 1st, 1999 from 12-1 pm in WeH 4623.
Presented by Jovan Popic,
The flow of water, the twirl of a dress and the breaking of glass are just some examples of complex physical phenomena computer animation attempts to recreate. Creating motions of this complexity with traditional hand animation techniques is too time-consuming, but physics provides a framework for generating them automatically. And recently, physical simulations have become the method of choice for creating these motions. Unfortunately, it is difficult to predict how the many physical parameters will affect the resulting motion. In place of the time-consuming hand drawing required by traditional techniques, artists must spend time finding the exact parameters for which the simulation produces the desired effect. Methods that provide direct control over simulated motion are still missing.
I will present an interactive method for intuitive, hands-on control of rigid body motion. This technique relies on a simple extension of the traditional rigid body dynamics formulation. To change the motion, the artist freely manipulates the bodies at any point in time and the system computes the required physical parameters for the new motion. Although this method shows promise, it is still incomplete. I will address in detail the most difficult unresolved problem, which will be of special interest to those interested in machine learning, robot path planning and optimization over discrete domains.
Cyber-Cartography: Drawing the Internet
Friday, October 15th, 1999 from 12-1 pm in WeH 4623.
Presented by Hal Burch,
As the Internet continues to grow and develop, we understand less and less its topology. One method to better grasp the Internet's connectivity is to be able to draw a picture of it. I will present one attempt to visually depict what the Internet looks like, starting from discovering the Internet's connectivity and going through possible uses of visual representations of the Internet. I will discuss some of the difficulties in determining the topological information and possible uses of this information directly. Additionally, I will cover the algorithm utilized to layout the nodes of the resulting graph and methodologies for representing additional data using the calculated layout.
Constructing Software Connectors
Friday, October 29th, 1999 from 12-1 pm in WeH 4623.
Presented by Bridget Spitznagel,
Increasingly, complex software systems are being constructed as compositions of reusable software components; these components are often written independently, then glued together with interaction mechanisms, or connectors, that allow the components to communicate. When existing connectors are not sufficient to meet the needs of a given system, the chief alternative is to write an ideal connector more-or-less from scratch (a nontrivial task). My research addresses the need for a middle ground, in which adequate connectors may be constructed more easily. I will discuss a compositional approach to constructing connectors and possible generation techniques for producing the connector implementations. This will be an informal talk (questions and suggestions welcome).
Multi-Fidelity Support for Interactive Mobile Applications
Friday, November 5th, 1999 from 12-1 pm in WeH 4623.
Presented by Dushyanth Narayanan,
In the decade to come, mobile, wearable computers will become increasingly powerful, cheaper, more usable, and widely prevalent. As they become more widely used, their main use will be to interact with the user, to provide some sort of functionality. These interactive applications are constrained by the need for low latency (for good interactive response). In many cases, we also want a low "user attention requirement" --- i.e., the application does its job with a minimum of distraction to the user, who may be engaged in some real-world task.
Mobility introduces some significant constraints of its own --- mobile machines are severely constrained by the need to conserve battery power. Wireless network connectivity is often low-bandwidth, unreliable, and expensive in terms of both battery power and money.
In this talk, I will introduce the notion of multi-fidelity computations, and how they can help applications to operate within these resource constraints. I will then discuss my approach to providing operating system support for multi-fidelity applications.
Machine Learning vs. The Efficient Market Hypothesis
Friday, November 12th, 1999 from 12-1 pm in WeH 4623.
Presented by James Thomas,
One of the foundations of modern economics is the idea that financial markets are efficient -- that they accurately process and distill all information about a security down to a single price. This idea is called the efficient market hypothesis, and it implies that extracting patterns based on past data that can consistently outperform the market is impossible.
But this cuts both ways: convincingly demonstrating that extracting such patterns is possible counts for empirical evidence against the efficient markets hypothesis. This has led to a line of economics research where machine learning techniques are used to find trading rules that produce beat the market consistently in out-of-sample data.
I will discuss the use of genetic programming to automatically discover these trading rules. Unfortunately, financial data is much noisier than datasets machine learning has traditionally been applied to. and this causes an shift of emphasis in the algorithm: instead of concentrating on search, understanding the representation and the data itself in relation to issues of overfitting is the key to success. Finally, I will discuss the idea of using text data in to supplement price data, and present some preliminary results using data collected from online stock chat boards.
Imagined Electronic Community: Representations of Virtual Community in Contemporary Business Discourse
Friday, December 3rd, 1999 from 12-1 pm in WeH 8220.
Presented by Christoper Werry,
In contemporary business texts corporate sponsored on-line communities are described as central to the commercial development of the Internet, and to the imagined future of narrowcasting and mass customization in the wider world of marketing and advertising.
My talk provides a brief history of how on-line community has been represented within models of Internet commerce. My research critically examines the arguments, narratives and rhetorical strategies drawn on within contemporary business texts to represent on-line community. It examines how the 'discourse of community' employed in these texts represents the possibilities of assembly, access, public use, and the control and ownership of knowledge produced by community members.
My talk provides a critique of some of the ways in which contemporary business models seek to commodify and privatize on-line community, and to police, order and regulate social interaction, cultural practices, community formation and knowledge production on-line. It also discusses why academics have an interest in involving themselves in helping organize alternative models of on-line community formation and knowledge production in the context of moves to corporatize and commodify higher education.
Link to related paper.
Web contact: sss+www@cs