About FAI...

The Forum for Artificial Intelligence meets every other Friday at 3pm to discuss topics in artificial intelligence. After the formal talk, we continue our conversation at the Crown and Anchor. All are welcome to attend.

Please send questions or comments to Patrick Beeson or Jeff Provost.

Full Schedule: 9/2010/410/1811/111/812/6

September 20th

ACES 2.302

The Trading Agent Competition:
Two Champion Adaptive Bidding Agents

Prof. Peter Stone
Department of Computer Sciences, UT Austin

The first Trading Agent Competition (TAC-00) was held in Boston, MA on July 8, 2000. Organized by researchers at the University Michigan and North Carolina State University, TAC was designed to provide a benchmark problem in the complex domain of e-marketplaces and to motivate researchers to apply unique approaches to a common task. The follow-up event, TAC-01, was held in Tampa, FL on October 14, 2001.

This talk presents ATTac-2000 and ATTac-2001, top-scoring agents in their respective competitions. TAC agents must bid in several simultaneous auctions for interacting goods. In auctions such as these, a fundamental problem is price prediction, and more broadly, the modeling of uncertainty regarding these prices. ATTac-2000's key design feature is its on-line adaptability that gave it the flexibility to react to a variety of market conditions. ATTac-2001 uses a novel and general boosting-based machine learning algorithm for conditional density estimation problems such as the price prediction problem.

In this talk, I present details of the TAC rules and the ATTac agents. In addition to the competition results, I present controlled experiments isolating the effectiveness of the agents' core features.

October 4th

ACES 2.302

The Skeleton in the Cognitive Map:
A Computational and Empirical Exploration

Dan Tecuci
Department of Computer Sciences, UT Austin

Experts seem to find routes in a complex environment by finding a connection from the starting place to a subset of major paths --- the ``skeleton'' --- then moving within the skeleton to the neighborhood of the destination, making a final connection to the destination.

In this talk we present a computational hypothesis to account for the skeleton as an emergent phenomenon, and test this hypothesis from both a computational and an empirical point of view.

The emergence of the skeleton in the cognitive map is based on the the interaction of three factors. (1) The topological map is represented as a bipartite graph of places and paths, where a path is an extended one-dimensional description of an ordered set of places. (2) Travel through the environment allows the traveler to incrementally accumulate topological relationships, including the relation of a place to a path serving as a dividing boundary separating two regions. (3) A bounding path is often a natural subgoal during way-finding search, meaning that paths rich in boundary relations are likely to appear in routes, which means they are likely to acquire more boundary relations. This positive-feedback loop leads to an oligarchy of paths rich in boundary relations: the skeleton.

October 18th

ACES 2.302

The Artificial Artist

Harold Henry Chaput
President, Austin Museum of Digital Art [web site]
Graduate Student, Department of Computer Sciences, UT Austin

Artists and art theorists have long been developing algorithms and formal theories of art. Some of these theories have even been somewhat generative, in that they allow for the production of a work given a set of rules and guidance from the artist. Notable examples include the music of J. S. Bach and Arnold Schönberg, and the visual art of Wassily Kandinsky. The trend of algorithmic art, though, has not been to increase control but to remove it, allowing the artist to establish a set of initial conditions and let the work generate itself. An essential component of Algorithmic Art (as it is understood today) is the "surprise," the ability of the algorithm to produce something that neither the viewer nor the artists expects. The rules of Algorithmic Art do not define or constrain the artist so that he or she would act like a machine. They liberate the machine, they grant it autonomy to freely produce original works of art.

In my talk, I will discuss the role of the algorithm in art, past and present. I will talk about the role of computer science, particularly artificial intelligence, in art and music today.

November 1st

Taylor 3.128

Qualitative Analysis of Linear Time-Invariant/Variant Dynamic Systems

Juan J. Flores [web]
Universidad Michoacana, Mexico

Linear Time-Invariant Dynamic Systems are the best understood kind of systems. Classical Engineering techniques exist to analyze and synthesize this kind of systems. Still, we can deploy the mathematical basis to develop methods that emphasize on the qualitative properties of the exhibited behaviour of those systems. Knowing that the solution to these kind of systems are composed solely of exponential terms (pairs of which can become sinusoidals), the resulting behaviours can be categorized according to their qualitative properties. A qualitative description language is being developed based on those properties. This tool allows us to predict all possible qualitatively different behaviours a system can exhibit. The predicted behaviors can be expressed in qualitative terms or, using an example generator, can be displayed emphasizing on the qualitative properties of the response. On the other direction we can use this theory to perform structural system identification. The proposed method extracts each component of the response, identifying the order of the system that may have produced that behaviour. Other methods (e.g. genetic algorithms) can yield a good numerical approximation to quantitative identification once we now the system's structure. The qualitative simulation also produces a behaviour graph that shows how the system's response can vary with time. This fact introduces us into the realm of time-varying systems. Properties of time-varying systems can be used for diagnosis Research is being done in that direction and we present preliminary results.

November 8th

ACES 6.304

Higher-order behavior-based systems

Dr. Ian Horswill [web]
Assistant Professor of Computer Sciences
Northwestern University

Classical artificial intelligence systems presuppose that all knowledge is stored in a central database of logical assertions or other symbolic representations and that reasoning consists largely of searching and sequentially updating that database. While this model has been very successful for disembodied reasoning systems, it is problematic for robots. Robots are distributed systems; multiple sensory, reasoning, and motor control processes run in parallel, often on separate processors that are only loosely coupled with one another. Each of these processes necessarily maintains its own separate, limited representation of the world and task; requiring them to constantly synchronize with the central knowledge base is probably unrealistic. I will discuss an alternative class of architectures tagged behavior-based systems that support a large subset of the capabilities of classical AI architectures, including limited quantified inference, forward- and backward-chaining, simple natural language question answering and command following, reification, and computational reflection, while allowing object representations to remain distributed across multiple sensory and representational modalities. Although limited, they also support extremely fast, parallel inference.

December 6th

ACES 2.302

RoboCup-Rescue: A Multi-Agent Environment

Mazda Ahmadi
Sharif University of Technology, Iran

One of the main requisites for multi-agent researchers is a multi-purpose infrastructure that enables them to create, test and evaluate protocols, methods and related algorithms. The RoboCup-Rescue simulation environment is a flexible and configurable environment that can fill this gap. Being configured differently, it turns out to have different features in such a way that enables easy implementation and testing of different designed methods. Most of multi-agent research issues such as cooperation, coordination, agent communication languages, and agent micro-level issues can also be experienced in this way. More importantly, sensible roles and tasks in the system motivate more research and turns this environment to a suitable infrastructure for educational purposes.

Past Schedules

Spring 2002

Fall 2001

Spring 2001

Fall 2000

Spring 2000

fai (fA) n. Archaic. [Middle English]: Faith.