The Forum for Artificial Intelligence meets every other week (or so) to discuss scientific, philosophical, and cultural issues in artificial intelligence. Both technical research topics and broader inter-disciplinary aspects of AI are covered, and all are welcome to attend!

If you would like to be added to the FAI mailing list, or have any questions or comments, please send email to Prem Melville, Misha Bilenko, or Nick Jong.

Using Machine Learning to Improve Information Finding on the Web

Dr. Mehran Sahami   [homepage]

Google Inc.

Web search is one of the most important applications used on the Internet, and it also poses many interesting opportunities to apply machine learning. In order to better help people find relevant information in a growing sea of data, we discuss various machine learning techniques that can be harnessed to sift, organize, and present relevant information to users. In this talk, we provide a brief background on information retrieval, and then look at some of the challenges faced in searching the Web. We specifically examine applications of machine learning to improve information retrieval, image classification, topical inference of queries, and record linkage. We show how these tasks are directly related to the overarching goal of improving various aspects of search on the web.

About the speaker:

Mehran Sahami is a Senior Research Scientist at Google and is also a Lecturer in the Computer Science Department at Stanford University. At Google, Mehran conducts research in machine learning and information retrieval technologies to help improve information access. Prior to joining Google, Mehran was involved in a number of commercial and research machine learning projects at E.piphany, Xerox PARC, SRI International, and Microsoft Research. He received his BS, MS, and PhD in Computer Science from Stanford, and evidently still has separation anxiety with respect to completely leaving there.

Autonomic Computing Research Challenges for AI

Dr. Jeff Kephart   
IBM Thomas J. Watson Research Center

[Sign-up schedule for individual meetings]

The increasing complexity of computing systems is beginning to overwhelm the capabilities of software developers and system administrators to design, evaluate, integrate, and manage these systems. Major software and system vendors such as IBM, HP and Microsoft have concluded that the only viable long-term solution is to create computer systems that manage themselves. This realization spurred IBM to launch its autonomic computing initiative over three years ago. While good progress has been made since then, it is clear that a worldwide collaboration among academia, IBM, and other industry partners will be required to fully realize the vision of autonomic computing. The AI community has much to contribute to this endeavor. I will discuss several fundamental challenges in the areas of artificial intelligence and agents, particularly in the areas of machine learning, policy, and architecture, and describe initial steps that IBM Research has taken to address some of these challenges.

About the speaker:

Jeffrey O. Kephart manages the Agents and Emergent Phenomena group at the IBM Thomas J. Watson Research Center, and shares responsibility for developing IBM's Autonomic Computing research strategy. He and his group focus on the application of analogies from biology and economics to massively distributed computing systems, particularly in the domains of autonomic computing, e-commerce, antivirus, and antispam technology. Kephart's research efforts on digital immune systems and economic software agents have been publicized in publications such as The Wall Street Journal, The New York Times, Forbes, Wired, Harvard Business Review, IEEE Spectrum, and Scientific American. In 2004, he co-founded the International Conference on Autonomic Computing. Kephart received a BS from Princeton University and a PhD from Stanford University, both in electrical engineering.

EUROPA - Planning and Scheduling Technology for Human-Robotic Space Exploration

Dr. Conor McGann   
Principal Investigator and Software Architect for EUROPA platform
NASA Ames Research Center

[Sign-up schedule for individual meetings]

NASA is committed to a vision of safe, effective and affordable space exploration over the long-haul. As evidenced by the on-going adventures of Spirit and Opportunity on the Martian surface, robots will play a central role in making that vision a reality. Increasingly ambitious missions require more sophisticated autonomous and collaborative operation of robotic systems. Planning is an integral part of such sophisticated systems. In recognition of this fact, NASA Ames Research Center has developed EUROPA, a planning and scheduling technology designed to infuse advanced planning capabilities into practical NASA applications. EUROPA also supports continued research and development to advance the state of the art in planning and plan execution. In this talk, I outline the key technologies integrated in EUROPA and discuss their theoretical foundations. I further describe applications of this technology in NASA mission-oriented research and mission deployment.

About the speaker:

Dr. Conor McGann is the Principal Investigator and Software Architect for EUROPA, a constraint-based planning and scheduling platform. He has been a research engineer at the NASA Ames Research Center in the Autonomy and Robotics Area since February, 2002. He has worked directly and indirectly on the Mars Exploration Rover (MER) mission and in 2004 was the recipient of the NASA Administrator's Turning Goals into Reality award as part of the MER infusion team. Prior to his work at Ames, Dr. McGann was the Chief Architect for the Customer Management Group for i2 Technologies. There, his focus was the opportunistic integration of customer facing pricing, configuration and quotation systems to the supply chain. Previously, Dr. McGann was the founder and CEO of Cunav Technologies, a Dublin-based software company applying practical problem solving technologies to solve hard business problems. Dr. McGann received his PhD in computer science from Trinity College Dublin in 1995 and his BA in computer engineering from the same institution in 1990.

Adaptive Control for Risk Management in eCommerce

Dr. Joseph Sirosh   
Vice President
Amazon.com

[Sign-up schedule for individual meetings]

Electronic marketplaces operate in a virtual, global eCommerce environment without boundaries, where authentication is difficult, anonymity is easy, and law enforcement is very limited. Traditional methods to minimize transaction risks - guards, security cameras, photo ID verification, human judgments based on appearances, reputation, conversations, location and environment - often taken for granted in the physical world - no longer apply, and new automated mechanisms based on digital transactional behavior are required. This talk will introduce a data driven adaptive control framework for managing risks in such an environment. Adaptive control systems encourage desirable behavior and discourage undesirable activity using sound statistical models developed through machine learning and data mining. As internet fraud, identity theft and information security risks grow in sophistication, such control systems provide for adaptation, and effective risk management for eCommerce.

About the speaker:

Dr. Joseph Sirosh is currently Vice President of Transaction Risk Management at Amazon.com, where he is focused on developing advanced risk management systems for eCommerce. Prior to joining Amazon.com he worked at Fair Isaac Corporation as VP of the Advanced Technology R&D group, exploring advanced analytic and data mining applications. At Fair Isaac and at HNC Software prior to that, he has led several significant R&D projects on security and fraud detection, predictive modeling, information retrieval, content management, intelligent agents and bioinformatics. He has made significant contributions in the field of neural network algorithms and in understanding the fundamental principles by which information is organized and processed in the brain. Dr. Sirosh has published over 20 technical papers and one book, and has been a lead investigator of various research grants from DARPA and other Government agencies.

Education: Ph.D. Computer Science, University of Texas Austin, 1995, M.S. Computer Science, University of Texas at Austin, 1992, B.Tech Computer Science and Engineering, Indian Institute of Technology, Madras, 1990

Hierarchical Search

Dr. Robert C. Holte   [homepage]
Department of Computing Science
University of Alberta

[Sign-up schedule for individual meetings]

Pattern databases are large and time-consuming to build, but enable large search problems to be solved very quickly. They are therefore ideally suited to situations where many different instances of the problem are to be solved, but poorly suited to situations where only a few problem instances are to be solved. This paper examines a technique especially designed for the latter situation - hierarchical search. The key idea is to compute, on-demand, only those pattern database entries that are needed to solve a given problem instance. Our experiments show that Hierarchical IDA* can solve individual problems very quickly, roughly an order of magnitude faster than the time required to build an entire high-performance pattern database.
The only background assumed by this talk is a basic knowledge of heuristic search and the IDA* algorithm. No familiarity with pattern databases is required.

About the speaker:

Professor Robert Holte is a well-known member of the international machine learning research community, former editor-in-chief of the leading international journal in this field (Machine Learning), and current director of the Alberta Ingenuity Centre for Machine Learning. His main scientific contributions are his seminal works on the problem of small disjuncts and the performance of very simple classification rules. His current machine learning research investigates cost-sensitive learning and learning in game-playing (for example: opponent modelling in poker, and the use of learning for gameplay analysis of commercial computer games). In addition to machine learning he undertakes research in single-agent search (pathfinding): in particular, the use of automatic abstraction techniques to speed up search. He has over 55 scientific papers to his credit, covering both pure and applied research, and has served on the steering committee or program committee of numerous major international AI conferences.

Random Walks on Text Structures

Dr. Rada Mihalcea   [homepage]
Department of Computer Science and Engineering
University of North Texas

[Sign-up schedule for individual meetings]

In this talk, I will present a new framework for the application of graph-based ranking algorithms implementing random-walk models (e.g. PageRank or HITS) to structures derived from text, and show how the synergy between graph-theoretical algorithms and graph-based text representations can result in efficient unsupervised methods for several natural language processing tasks. I will illustrate this framework with several text processing applications, including word sense disambiguation, extractive summarization, and keyphrase extraction. I will also outline a number of other applications that can find successful solutions within this framework, and conclude with a discussion of opportunities and challenges for future research.

About the speaker:

Rada Mihalcea is an Assistant Professor of Computer Science at University of North Texas. Her research interests are in lexical semantics, minimally supervised natural language learning, and multilingual natural language processing. She is currently involved in a number of research projects, including word sense disambiguation, shallow semantic parsing, (non-traditional) methods for building annotated corpora with volunteer contributions over the Web, and graph-based ranking algorithms for language processing. She is the president of the ACL special interest group on the lexicon (SIGLEX), a board member for the ACL special interest group on natural language learning (SIGNLL), and serves on the editorial board of Computational Linguistics. Her research is supported by NSF and UNT.

Unsupervised Learning of Human Language Structure

Dr. Christopher Manning   [homepage]
Departments of Computer Science and Linguistics
Stanford University

[Sign-up schedule for individual meetings]

While there is certainly debate about how much inbuilt linguistic bias ("Universal Grammar") human language learners possess and as to whether they receive useful feedback during learning, children nevertheless definitely acquire language in a primarily unsupervised fashion. In contrast, most current computational approaches to language processing are almost exclusively supervised, relying on hand-labeled corpora for training. This reflects the fact that despite the promising rhetoric of machine learning, attempts at unsupervised grammar induction have been seen as largely unsuccessful, and supervised training data remains the practical route to high performing systems.

In this talk I will present work that comes close to solving the problem of inducing tree structure or surface dependencies over language - that is, providing the primary descriptive structures of modern syntax. While this work uses modern learning techniques, the primary innovation is not in learning methods but in finding appropriate representations over which learning can be done. Overly complex models are easily distracted by non-syntactic correlations and local maxima, while overly simple models aren't rich enough to capture important first-order properties of language (such as directionality, adjacency, and valence). We describe several syntactic representations which are designed to capture the basic character of natural language syntax as directly as possible. With these representations, high-quality parses can be learned from surprisingly little text, with no labeled examples and no language-specific biases.

(This talk covers work done with Dan Klein, now at UC Berkeley.)

About the speaker:

Chris Manning is an Assistant Professor of Computer Science and Linguistics at Stanford University. He received his Ph.D. from Stanford University in 1995, and held faculty positions in the Computational Linguistics Program at Carnegie Mellon University (1994-1996) and in the University of Sydney Linguistics Department (1996-1999) before returning to Stanford. He is a Terman Fellow and recipient of an IBM Faculty Award. His recent work has concentrated on statistical parsing, grammar induction, and probabilistic approaches to problems such as word sense disambiguation, part-of-speech tagging, and named entity recognition, with an emphasis on complementing leading machine learning methods with use of rich linguistic features. Manning coauthored the leading textbook on statistical approaches to NLP (Manning and Schuetze 1999) and (with Dan Klein) received the best paper award at the Association for Computational Linguistics 2003 meeting for the paper Accurate Unlexicalized Parsing.

Online Learning by Projecting -- from Theory to Large Scale Web-spam filtering    [Talk slides]

Dr. Yoram Singer   [homepage]
School of Computer Science and Engineering
Hebrew University

A unified algorithmic framework and corresponding analysis for numerous problems in online learning is presented. The basic algorithm works by projecting an instantaneous hypothesis onto a single hyperplane which forms the basis for the next instantaneous hypothesis. In particular we discuss classification, regression, and uniclass problems. The analysis is based on simple convexity properties combined with mistake bound techniques. After describing the basic algorithmic setup we discuss a few extensions to more complex problems. Specifically, we describe online learning algorithms for multiclass problems, hierarchical classification, rank-ordering learning and pseudo-metric learning. Finally, we discuss large scale implementation of the algorithms for the purpose of web-spam filtering.

Based on joint works with Koby Crammer (UPenn), Ofer Dekel and Vineet Gupta (Google), Shai Shwartz and Jospeh Keshet (HUJI), and Andrew Ng (Stanford).

About the speaker:

Yoram Singer is an associate professor of computer science at the Hebrew University, Jerusalem, Israel. He is currently on leave of absence at Google Inc. He got his Ph.D. in 1995 in computer science. From 1995 through 1999 he was a member of the technical staff at AT&T Research. His work focuses on the design, analysis, and implementation of machine learning algorithms.

Exploring Trading Strategy Spaces     [Talk slides]

Dr. Michael Wellman  [homepage]
Department of Computer Science and Engineering
University of Michigan

Given that the complexity of many market games precludes analytic characterization of equilibrium, we require alternative means of evaluating strategic alternatives. My group has been applying an empirical game-theoretic methodology to the study of several interesting market games, yielding insights into key strategic issues as well as evidence bearing on particular strategies. Examples include simultaneous auctions as well as two scenarios from the annual Trading Agent Competition: one in travel shopping and the other in supply chain management.

About the speaker:

Michael P. Wellman received a PhD from the Massachusetts Institute of Technology in 1988 for his work in qualitative probabilistic reasoning and decision-theoretic planning. From 1988 to 1992, Wellman conducted research in these areas at the USAF^Òs Wright Laboratory. For the past dozen+ years, his research has focused on computational market mechanisms for distributed decision making and electronic commerce. As Chief Market Technologist for TradingDynamics, Inc. (now part of Ariba), he designed configurable auction technology for dynamic business-to-business commerce. Wellman is Chair of the ACM Special Interest Group on Electronic Commerce (SIGecom), and previously served as Executive Editor of the Journal of Artificial Intelligence Research. He has been elected Councilor and Fellow of the American Association for Artificial Intelligence. In 2000 he initiated an annual series of international Trading Agent Competitions, and recently founded the Association for Trading Agent Research to organize that ongoing activity.