Forum for Artificial Intelligence


[ About FAI   |   Upcoming talks   |   Past talks ]

About FAI

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 Nick Jong, Lily Mihalkova, or Dan Tecuci .

Upcoming talks

Past talks

Thursday, July 13
1:00pm, ACES 2.402
William W. Cohen
CMU
A Framework for Learning to Query Heterogeneous Data
Friday, Aug. 25
11:00am, ACES 6.304
Paul Bennett,
CMU
Building Reliable Metaclassifiers for Text Learning
Thursday, Aug. 31
3:30pm, ACES 2.302
Deb Roy,
MIT
Meaning Machines
Thursday, Sep. 7
3:30pm, ACES 2.302
David Fogel,
Natural Selection, Inc.
Behind the Scenes with Blondie24: Evolving Intelligence in Checkers and Chess
Friday, November 3
3:00pm, ACES 2.402
Matthew Campbell
University of Texas at Austin, Department of Mechanical Engineering
Search Methods for Finding Optimal Graph Topologies
Friday, November 10
11:00am, ACES 2.402
Katrin Erk
University of Texas at Austin, Department of Linguistics
Automatic meaning analysis of free text: small steps towards a big goal
Wednesday, November 29
2:00pm, ACES 2.402
Charles Ofria
Michigan State University
Life in the Machine: The Evolution of Novel Complexity in Digital Organisms
Friday, December 8
11:00am, ACES 2.402
Sugato Basu
SRI International, Menlo Park, CA
Machine Learning in Web 2.0: Analyzing Dynamic Content-driven Social Networks
Friday, February 9
11:00am, ACES 2.402
Rich Caruana
Cornell University
Which Supervised Learning Method Works Best for What?
An Empirical Comparison of Learning Methods and Metrics++
Thursday, February 15
2:00pm, ACES 2.402
Mark Johnson
Brown University
Bayesian Inference of Grammars
Friday, February 23
11:00am, ACES 2.302 (Avaya)
Robert Holte
University of Alberta
Additive Pattern Database Heuristics
Wednesday, March 7
11:00am, ACES 2.402
Mark Steedman
University of Edinburgh
The Computational Problem of Natural Language Acquisition
Friday, March 9
11:00am, ACES 6.304
Hermann Helbig
University of Hagen, Germany
Multilayered Extended Semantic Networks as a Knowledge Representation Paradigm and Interlingua for Meaning Representation
Wednesday, March 14
11:00am, ACES 6.304
Rong Jin
Michigan State University
Generalized Maximum Margin Clustering and Unsupervised Kernel
Friday, March 30
2:00pm, ACES 2.402
Renata Vieira
Department of Informatics
UNISINOS
Sao Leopoldo, Brazil
Have I mentioned it?
Thursday, April 5
1:30pm, ACES 2.402
Katya Scheinberg
Mathematical Sciences Dept.,
T.J. Watson Research Center, IBM
Active set methods for large-scale support vector machines
Friday, April 20
11:00am, ACES 6.304
Michael L. Littman
Rutgers University
Advancing the Theory and Practice of Model-based Reinforcement Learning
Friday, April 27
11:00am, TAY 3.128
Gregory Dudek
McGill University
Vision-Based Behavior Control for Underwater Robotics
Monday, April 30
11:00am, ACES 2.302 (Avaya)
Nicholas Roy
Massachusetts Institute of Technology
Planning in Uncertain Worlds: Exploration and Information Gathering
Monday, August 6
11:00am, ACES 6.304
Kevin Knight
University of Southern California
The Voynich Manuscript

Thursday, July 13, 1:00pm

Coffee at 12:45pm

ACES 2.402

A Framework for Learning to Query Heterogeneous Data

Dr. William W. Cohen   [homepage]

Machine Learning Department
Carnegie Mellon University

[Sign-up schedule for individual meetings]

A long-term goal of research on data integration is to develop data models and query languages that make it easy to answer structured queries using heterogeneous data. In this talk, I will describe a very simple query language, based on typed similarity queries, which are answered based on a graph containing a heterogeneous mixture of textual and non-textual objects. The similarity metric proposed is based on a lazy graph walk, which can be approximated efficiently using methods related to particle filtering. Machine learning techniques can be used to improve this metric for specific tasks, often leading to performance far better than plausible task-specific baseline methods. We experimentally evaluate several classes of similarity queries from the domains of analysis of biomedical text and personal information management: for instance, in one set of experiments, a user's personal information is represented as a graph containing messages, calendar information, social network information, and a timeline, and similarity search is used to find people likely to attend a meeting.

This is joint work with Einat Minkov and Andrew Ng.

About the speaker:

William Cohen received his bachelor's degree in Computer Science from Duke University in 1984, and a PhD in Computer Science from Rutgers University in 1990. From 1990 to 2000 Dr. Cohen worked at AT&T Bell Labs and later AT&T Labs-Research, and from April 2000 to May 2002 Dr. Cohen worked at Whizbang Labs, a company specializing in extracting information from the web. Dr. Cohen is member of the board of the International Machine Learning Society, and has served as an action editor for the Journal of Machine Learning Research, the journal Machine Learning and the Journal of Artificial Intelligence Research. He co-organized the 1994 International Machine Learning Conference, is the co-Program Committee Chair for the 2006 International Machine Learning Conference, and has served on more than 20 program committees or advisory committees.

Dr. Cohen's research interests include information integration and machine learning, particularly information extraction, text categorization and learning from large datasets. He holds seven patents related to learning, discovery, information retrieval, and data integration, and is the author of more than 100 publications.

Friday, Aug. 25, 11:00am

Coffee at 10:45am

ACES 6.304

Building Reliable Metaclassifiers for Text Learning

Dr. Paul Bennett   [homepage]

Computer Science Department
Carnegie Mellon University

[Sign-up schedule for individual meetings]

Appropriately combining information sources is a broad topic that has been researched in many forms. It includes sensor fusion, distributed data-mining, regression combination, classifier combination, and even the basic classification problem. After all, the hypothesis a classifier emits is just a specification of how the information in the basic features should be combined. This talk addresses one subfield of this domain: leveraging locality when combining classifiers for text classification.

After discussing and introducing improved methods for recalibrating classifiers, we define local reliability, dependence, and variance and discuss the roles they play in classifier combination. Using these insights, we motivate a series of reliability-indicator variables which intuitively abstract the input domain to capture the local context related to a classifier's reliability.

We then present our main methodology, STRIVE. STRIVE employs a metaclassification approach to learn an improved model which varies the combination rule by considering the local reliability of the base classifiers via the indicators. The resulting models empirically outperform state-of-the-art metaclassification approaches that do not use locality. Next, we analyze the contributions of the various reliability indicators to the combination model and suggest informative features to consider when redesigning the base classifiers. Finally, we show how inductive transfer methods can be extended to increase the amount of labeled training data for learning a combination model by collapsing data traditionally viewed as coming from different learning tasks.

About the speaker:

Paul Bennett is currently a Postdoctoral Fellow in the Language Technologies Institute at Carnegie Mellon University where he serves as Chief Learning Architect on the RADAR project. Paul's primary research interests are in text classification, information retrieval, ensemble methods, and calibration, with wider interests in statistical learning and applications of artificial intelligence in adaptive systems in general. His published work includes research on classifier combination, action-item detection, calibration, inductive transfer, machine translation, and recommender systems. Paul received his Ph.D. (2006) from the Computer Science Department at Carnegie Mellon University.

Thursday, Aug. 31, 3:30pm

Coffee at 3:15pm

ACES 2.302, Avaya Auditorium

Meaning Machines

Dr. Deb Roy   [homepage]

MIT Media Laboratory
MIT

[Sign-up schedule for individual meetings]

People use words to refer to the world as a means for influencing the beliefs and actions of others. Although many isolated aspects of the structure and use of language have been extensively studied, a unified model of situated language use remains unexplored. Any attempt to explain unconstrained adult language use appears futile due to the overwhelming complexity of the physical, cognitive, and cultural factors at play. A strategy for making progress towards a holistic account of language use is to study simple forms of language (e.g., conversational speech about objects and events in the here-and-now in limited social contexts) and strive for "vertically integrated" computational models. I will present experiments guided by this strategy in building conversational robots and natural language interfaces for video games. An emerging framework suggests a semiotic perspective may be useful for designing systems that process language grounded in social and physical context.

About the speaker:

Deb Roy is Associate Professor of Media Arts and Sciences at the Massachusetts Institute of Technology. He is Director of the Cognitive Machines Group at the MIT Media Laboratory which he founded in 2000. Roy also directs the 10x research program, a lab-wide effort to design new technologies for enhancing human cognitive and physical capabilities. Roy has published numerous peer-reviewed papers in the areas of knowledge representation, speech and language processing, machine perception, robotics, information retrieval, cognitive modeling, and human-machine interaction, and has served as guest editor of the journal Artificial Intelligence. He has lectured widely in academia and industry. His work has been featured in various popular press venues including the New York Times, the Globe and Mail, CNN, BBC, and PBS. In 2003 Roy was appointed AT&T Career Development Professor. He holds a B.A.Sc. in Computer Engineering from University of Waterloo, and a Ph.D. in Media Arts and Sciences from MIT.

Thursday, Sep. 7, 3:30pm

Coffee at 3:15pm

ACES 2.302, Avaya Auditorium

Behind the Scenes with Blondie24: Evolving Intelligence in Checkers and Chess

Dr. David Fogel   [homepage]

Chief Executive Officer
Natural Selection, Inc.

[Sign-up schedule for individual meetings]

Blondie24 is a self-learning checks program that taught itself to play at the level of human experts. Starting with only rudimentary information about the location, number, and types of checkers pieces on the board, Blondie24 learned to play well enough to be ranked in the top 500 of 120,000 checkers players registered at Microsoft's zone.com. The program uses a simple evolutionary algorithm to optimize neural networks as board evaluators. Any sophisticated features used to interpret the positions of pieces were invented within the neural network. Furthermore, the evolving neural networks were not told whether they won, lost, or drew any specific game; instead, the only feedback they received was a point score associated with an overall result of playing a random number of games. In so doing, the line of research addressed two fundamental issues raised by Arthur Samuel and Allen Newell over three decades ago: Can a computer invent features in checkers and can a computer learn how play without receiving explicit credit assignment? A similar process has also been applied to chess (Blondie25). Starting with an open source program rated about 1800 (Class A), the evolved program has demonstrated grandmaster-level performance. The lecture will provide motivation and technical details for this research, as well as offer materials not found in any technical or book treatments of the development. Attendees will be able to challenge Blondie to a game, if they like.

About the speaker:

Dr. David Fogel is chief executive officer of Natural Selection, Inc. in La Jolla, California. Dr. Fogel has over 200 technical publications and 6 books, including Blondie24: Playing at the Edge of AI (Morgan Kaufmann, 2002) and How to Solve It: Modern Heuristics (with Zbigniew Michalewicz, 2nd ed., Springer 2005, translated into Chinese and Polish). Among many leadership roles, Dr. Fogel was the founding editor-in-chief of the IEEE Transactions on Evolutionary Computation (1996-2002), general chairman for the 2002 IEEE World Congress on Computational Intelligence, and will chair the upcoming 2007 IEEE Symposium Series in Computational Intelligence to be held April 1-5, 2007 in Honolulu, Hawaii. He was elected a Fellow of the IEEE in 1999 and received the 2004 IEEE Kiyo Tomiyasu Technical Field Award. He was elected president-elect of the IEEE Computational Intelligence Society for 2007.

Friday, November 3, 3:00pm

Coffee at 2:45pm

ACES 2.402

Search Methods for Finding Optimal Graph Topologies

Dr. Matthew Campbell   [homepage]

Mechanical Engineering
University of Texas at Austin

This research offers a fundamental new view of topology optimization. To date, topological synthesis approaches are simply augmentations of existing stochastic optimization techniques. The generic approach defined here combines aspects of existing optimization techniques, graph theory, mathematical programming, artificial intelligence, and shape and graph grammars. Graph transformation research has existed for nearly 40 years in an esoteric corner of artificial intelligence but only recently has the work been deemed useful in design automation as knowledge and heuristics of a particular problem domain can be encapsulated into rules. In this presentation, various example problems are presented that are in the process of being solved by these newly defined search methods.

About the speaker:

Dr. Campbell joined the Department of Mechanical Engineering at the University of Texas at Austin in 2000, and is currently an Associate Professor in the Manufacturing and Design area. His research focuses on computational methods that aid the engineering designer earlier in the design process than traditional optimization would. To date, he has been awarded $1.57 million in research funding, including the CAREER award for research into a generic graph topology optimization method. This research represents a culmination of past computational synthesis research including the automatic design of sheet metal components, multi-stable MEMS devices, MEMS resonators, function structures, and electro-mechanical configurations. Dr. Campbell is a member of the AAAI, the AIAA, Phi Kappa Phi Honor Society, Pi Tau Sigma Mechanical Engineering Honorary Fraternity, the ASME, the ASEE, and the Design Society and has been acknowledged with best paper awards at conferences sponsored by the latter three.

Friday, November 10, 11:00am

Coffee at 10:45am

ACES 2.402

Automatic meaning analysis of free text: small steps towards a big goal

Dr. Katrin Erk   [homepage]

Department Of Linguistics
University of Texas at Austin

Viewed as a whole, the problem of doing an automatic meaning analysis of free text is huge. But maybe the problem can be carved up in more manageable pieces: Recently several approaches to an automatic predicate-argument structure analysis have been proposed, what has been called a ``who does what to whom'' analysis. This can be seen as a first building block in a modular meaning analysis (where other, very much necessary, building blocks would include negation and modals). It is an important building block, which focuses on lexical semantics and on the link to semantic taxonomies -- and a building block that has recently become much more accessible, with the availability of manually annotated corpora.
In this talk I first take a closer look at the data for predicate-argument structure, from the viewpoint of a manual annotation effort, where we annotated a German corpus with FrameNet-style information. I then present a system for automatic predicate-argument structure analysis, its architecture and the statistical modeling for its subtasks. Lastly, I discuss a study on cross-lingual semantic analysis -- which opens up the possibility of deriving cross-lingual paraphrases.

About the speaker:

Katrin Erk is an assistant professor in the Department of Linguistics at the University of Texas at Austin. She completed her dissertation on tree description languages at Saarland University in 2002, advised by Gert Smolka. From 2002 to 2006, she held a researcher position in Saarbruecken working with Manfred Pinkal. Her current work includes research on machine learning methods for semantic analysis, the acquisition of lexical information from corpora, manual semantic annotation, the detection of multiword expression, and computational models for word sense.

Wednesday, November 29, 2:00pm

Coffee at 1:45pm

ACES 2.402

Life in the Machine: The Evolution of Novel Complexity in Digital Organisms

Dr. Charles Ofria   [homepage]

Computer Science and Engineering
Michigan State University

[Sign-up schedule for individual meetings]

When Darwin first proposed his theory of evolution by natural selection, he realized that it had a problem explaining the origins of traits of "extreme perfection and complication" such as the vertebrate eye. Over the years, critics of Darwin's theory have latched onto this perceived flaw as proof that Darwinian evolution is impossible. In anticipation of this issue, Darwin described the perfect data needed to understanding this process, but lamented that such data are "scarcely ever possible" to obtain. In this talk, I will discuss research where we use digital organisms (populations of self-replicating and evolving computer programs) to elucidate the process by which new, highly-complex traits arise, drawing inspiration directly from Darwin's wistful thinking and hypotheses. I will also explore some of the implications of this research to other aspects of evolutionary biology and new ways that these evolutionary principles can be applied toward solving computational problems.

About the speaker:

Dr. Charles Ofria is an assistant professor at Michigan State University in the Computer Science Department and the Ecology, Evolutionary Biology, and Behavior Program. He has a heavily multidisciplinary background, receiving a PhD from the Computation and Neural Systems department at Caltech under physicist Chris Adami, then doing a postdoc for three years in the Microbial Ecology program at MSU under biologist Richard Lenski. He is now the director of the MSU Digital Evolution Lab, a multidisciplinary group using digital organisms to answer fundamental questions in evolutionary biology and harnessing the results to solve more applied computational problems. Please see http://devolab.cse.msu.edu/ for more information.

Friday, December 8, 11:00am

Coffee at 10:45am

ACES 2.402

Machine Learning in Web 2.0: Analyzing Dynamic Content-driven Social Networks

Dr. Sugato Basu   [homepage]

SRI International
Menlo Park, CA

[ Sign-up schedule for individual meetings]

In the last decade, machine learning and data mining techniques have seen widespread successful application to different Internet technologies, including web search, product recommendation, spam detection, spelling correction, and news clustering. However, the web is fast undergoing a paradigm shift, moving from being a mechanism for delivering static web-content in the existing Web 1.0 model to a platform facilitating dynamic collaborative content creation in the emerging Web 2.0 paradigm. This trend is reflected in the growing popularity of new social web-services, for example, tagging (Flickr) compared to photo editing (Ofoto), and blogging (Blogger) compared to homepage hosting (Geocities).
This talk will outline how this new emphasis on rapid creation and sharing of consumer-generated data (CGM) over large social networks has given rise to dynamic content-driven social networks, and a new set of challenging machine learning problems in this context. Focusing on a project (iLink) that the speaker is currently working on, the talk will discuss research problems like online learning of topic models over streaming text, large-scale topic analysis over social networks, and learning to route messages in a social query model.

About the speaker:

Sugato Basu works on machine learning, data mining, information retrieval, statistical pattern recognition and optimization, with applications to analysis of text data and social networks on the web. He did his PhD from UT Austin in 2005, and is now a research scientist at the AI Center in SRI International. For more information: http://www.ai.sri.com/people/basu/

Friday, February 9, 11:00am

Coffee at 10:45am

ACES 2.402

Which Supervised Learning Method Works Best for What?
An Empirical Comparison of Learning Methods and Metrics++

Dr. Rich Caruana   [homepage]

Cornell University

[Sign-up schedule for individual meetings]

Decision trees are intelligible, but do they perform well enough that you should use them? Have SVMs replaced neural nets, or are neural nets still best for regression, and SVMs best for classification? Boosting maximizes margins similar to SVMs, but can boosting compete with SVMs? And if it does compete, is it better to boost weak models, as theory might suggest, or to boost stronger models? Bagging is simpler than boosting -- how well does bagging stack up against boosting? Breiman said Random Forests are better than bagging and as good as boosting. Was he right? And what about old friends like logistic regression, KNN, and naive bayes? Should they be relegated to the history books, or do they still fill important niches?

In this talk we compare the performance of these supervised learning methods on a number of preformaance criteria: Accuracy, F-score, Lift, Precision/Recall Break-Even Point, Area under the ROC, Average Precision, Squared Error, Cross-Entropy, and Probability Calibration. The results show that no one learning method does it all, but some methods can be "repaired" so that they do very well across all performance metrics. In particular, we show how to obtain the best probabilities from max margin methods such as SVMs and boosting via Platt's Method and isotonic regression. We then describe a new ensemble method that combines select models from these ten learning methods to yield much better performance. Although these ensembles perform extremely well, they are too complex for many applications. We'll describe a model compression method we are developing to fix that. Finally, if time permits, we'll discuss how the performance metrics relate to each other, and which of them you probably should (or shouldn't) use.

About the speaker:

Rich Caruana obtained his Ph.D. in computer science from Carnegie Mellon University in 1998. His current research focus is on ensemble learning, model calibration, inductive transfer, and adaptive clustering, and applications of these methods to problems in medical decision making and bioinformatics. In 2000 Caruana led a team that developed the first automated system for the early detection of bioterrorist releases of anthrax. The system applies data mining to consumer purchases in supermarkets to look for unexplained increases in the sales of products such as cough syrup. Because consumers tend to self-medicate using easily available products such as cough syrup and throat lozenges before consulting physicians, the system can detect the onset of flu-like symptoms 24-48 hours before these can be detected by visits to hospitals and doctors offices. A theme that runs through all of Professor Caruana's work is the importance of developing methods that are effective on real-world problems. He likes to mix algorithm development with applications work to insure that the methods he develops are useful in practice.

Thursday, February 15, 2:00pm

Coffee at 1:45pm

ACES 2.402

Bayesian Inference of Grammars

Dr. Mark Johnson   [homepage]

Brown University

[Sign-up schedule for individual meetings]

Even though Maximum Likelihood Estimation (MLE) of Probabilistic Context-Free Grammars (PCFGs) is well-understood (the Inside-Outside algorithm can do this efficiently from the terminal strings alone) the inferred grammars are usually linguistically inaccurate. In order to better understand why maximum likelihood finds poor grammars, this talk examines two simple natural language induction problems: morphological segmentation and word segmentation. We identify several problems with the MLE PCFG models of these problems and propose Hierarchical Dirichlet Process (HDP) models to overcome them. In order to test these HDP models we develop MCMC algorithms for Bayesian inference of these models from strings alone. Finally, we discuss to what extent the lessons learnt from these examples can be put into a unified framework and applied to the general problem of grammar induction.
Joint work with Sharon Goldwater and Tom Griffiths.

About the speaker:

Mark Johnson is a Professor of Cognitive and Linguistic Science and Computer Science at Brown University and a Visiting Researcher in the Natural Language group at Microsoft Research for 2006--2007. He was awarded a BSc (Hons) in 1979 from the University of Sydney, an MA in 1984 from the University of California, San Diego and a PhD in 1987 from Stanford University. He held a postdoctoral fellowship at MIT from 1987 until 1988, and has been a visiting researcher at the University of Stuttgart, the Xerox Research Centre in Grenoble and CSAIL at MIT. He has worked on a wide range of topics in computational linguistics, but his main research area is parsing and its applications to text and speech processing. He was President of the Association for Computational Linguistics in 2003.

Friday, February 23, 11:00am

Coffee at 10:45am

ACES 2.302 (Avaya)

Additive Pattern Database Heuristics

Dr. Robert Holte   [homepage]

University of Alberta

[Sign-up schedule for individual meetings]

This research studies heuristic functions defined by abstractions, where the distance from a state S to the goal state is estimated with the true distance from the abstract state corresponding to S to the abstract goal state. When precomputed and stored as lookup tables, such heuristics are called pattern databases (PDBs), and are the most powerful heuristics known for many problems.
The question addressed in this presentation is: under what conditions is the sum of several PDB heuristic functions admissible (guaranteed never to overestimate a distance) ?
This question has previously been addressed by Felner, Korf, and Hanan in a 2004 JAIR paper. Our work generalizes their answer, greatly enlarging the types of search spaces for which admissible additive heuristics can be defined. Experimental results on the Pancake puzzle show that well-chosen additive PDBs reduce solution time by three orders of magnitude over the best published results for this puzzle. We also show that additive PDBs are not always superior to taking the maximum over heuristics based on the same abstractions.
The presentation assumes basic knowledge of heuristic search, but does not require any knowledge of pattern databases.

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.

Wednesday, March 7, 11:00am

Coffee at 10:45am

ACES 2.402

The Computational Problem of Natural Language Acquisition

Dr. Mark Steedman   [homepage]

University of Edinburgh

[Sign-up schedule for individual meetings]

The talk reviews work-in-progress on language acquisition in children and robots using combinatory categorial grammar (CCG), building on work by Siskind, Villavicencio, and Zettlemoyer, among others.

CCG is a theory of grammar in which all language-specific grammatical information resides in the lexicon. A small universal set of strictly type-driven, non-structure dependent, syntactic rules (based on Curry's combinators B, S, and T) then "projects" lexical items into sentence-meaning pairs. The task that faces the child in the earliest stages of language acquisition can therefore be seen as learning a lexicon on the basis of exposure to (probably ambiguous, possibly somewhat noisy) sentence-meaning pairs, given this universal combinatory "projection principle", and a mapping from semantic types to the set of all universally available lexical syntactic types.

The talk argues that a very simple statistical model allows children to arrive at a target lexicon without navigation of subset principles, or attention to any attendant notion of trigger other than the notion "reasonably short sentence in a reasonably understandable situation". The model explains the pattern of errors that have been found in elicitation experiments. The linguistic notion of "parameter" appears to be redundant to this process.

The talk goes on to consider some more general implications of the theory, including its application to the phenomenon of "syntactic bootstrapping," touching on the question of the prelinguistic origin of the combinatory projection principle itself.

About the speaker:

Mark Steedman is Professor in the School of Informatics at the University of Edinburgh. He received his PhD from the University of Edinburgh in 1973. He came to Edinburgh in 1998 from the University of Pennsylvania, where he was Professor in the Department of Computer and Information Science. He is a Fellow of the American Association for Artificial Intelligence, a Fellow of the Royal Society of Edinburgh, and a Fellow of the British Academy.

His research interests cover issues in computational linguistics, artificial intelligence, computer science, and cognitive science, and their applications in practical systems, including syntax and semantics of natural language, wide-coverage parsing, comprehension of natural language by humans and by machine, and the role of intonation in spoken language generation and analysis. Some of his research concerns the analysis of music by humans and machines. He has acted as advisor for twenty-four PhDs.

Friday, March 9, 11:00am

Coffee at 10:45am

ACES 6.304

Multilayered Extended Semantic Networks as a Knowledge Representation Paradigm and Interlingua for Meaning Representation

Dr. Hermann Helbig   [homepage]

University of Hagen

[Sign-up schedule for individual meetings]

The talk gives an overview of Multilayered Extended Semantic Networks (abbreviated MultiNet), which is one of the most comprehensively described knowledge representation paradigms used as a semantic interlingua in large-scale NLP applications and for linguistic investigations into the semantics and pragmatics of natural language.

As with other semantic networks, concepts are represented in MultiNet by nodes, and relations between concepts are represented as arcs between these nodes. Additionally to that, every node is classified according to a predefined conceptual ontology forming a hierarchy of sorts, and the nodes are embedded in a multidimensional space of layer attributes and their values.

MultiNet provides a set of about 150 standardized relations and functions which are described in a very concise way including an axiomatic apparatus, where the axioms are classified according to predefined types. The representational means of MultiNet claim to fulfill the criteria of universality, homogeneity, and cognitive adequacy. In the talk, it is also shown, how MultiNet can be used for the semantic representation of different semantic phenomena.

To overcome the quantitative barrier in building large knowledge bases and semantically oriented computational lexica, MultiNet is associated with a set of tools including a semantic interpreter NatLink for automatically translating natural language expressions into MultiNet networks, a workbench LIA for the computer lexicographer, and a workbench MWR for the knowledge engineer for managing and graphically manipulating semantic networks.

The applications of MultiNet as a semantic interlingua range from natural language interfaces to the Internet and to dedicated databases, over question-answering systems, to systems for automatic knowledge acquisition.

About the speaker:

Hermann Helbig is Professor at the University of Hagen, Germany, and head of the chair Intelligent Information and Communication Systems. He received his Dr.rer.nat. (PhD) in 1976 in Automatic Symbolic Formula Manipulation and his Dr.rer.nat.habil. (Habilitation) in 1986 in Knowledge Representation. His experiences in AI research cover a period of more than 30 years. His main contributions lie in the fields of question answering (question answering system FAS-80), natural language interfaces to data bases (NLI-AIDOS), word-class controlled functional analysis (WCFA), knowledge representation (MultiNet paradigm), and computational lexicography (semantically based computational lexicon HaGenLex). He is author of several monographs in AI, his last book relevant to the talk is "Knowledge Representation and the Semantics of Natural Language".

His research interests cover issues in Natural Language Processing, Computational Lexicography, Knowledge Representation and Management, Semantics of NL, and Electronic Distance Teaching in AI.

Wednesday, March 14, 11:00am

Coffee at 10:45am

ACES 6.304

Generalized Maximum Margin Clustering and Unsupervised Kernel

Dr. Rong Jin   [homepage]

Michigan State University

[Sign-up schedule for individual meetings]

Maximum margin clustering extends the theory of support vector machine to unsupervised learning, and has shown promising performance in recent studies. However, it has three major problems that question its application of real-world applications: (1) it is computationally expensive and difficult to scale to large-scale datasets; (2) it requires data preprocessing to ensure the clustering boundary to pass through the origins, which makes it unsuitable for clustering unbalanced dataset; and (3) its performance is sensitive to the choice of kernel functions. In this paper, we propose the "Generalized Maximum Margin Clustering" framework that addresses the above three problems simultaneously.

The new framework generalizes the maximum margin clustering algorithm in that (1) it allows any clustering boundaries including those not passing through the origins; (2) it significantly improves the computational efficiency by reducing the number of parameters; and (3) it automatically determines the appropriate kernel matrix without any labeled data. Our empirical studies demonstrate the efficiency and the effectiveness of the generalized maximum margin clustering algorithm. Furthermore, in this talk, I will show the theoretical connection among the spectral clustering, the maximum margin clustering and the generalized maximum margin clustering.

About the speaker:

Dr. Rong Jin is an Assistant Prof. of the Computer Science and Engineering Dept. of Michigan State University since 2003. He is working in the areas of statistical machine learning and its application to information retrieval. Dr. Jin holds a B.A. in Engineering from Tianjin University, an M.S. in Physics from Beijing University, and an M.S. and Ph.D. from School of Computer Science of Carnegie Mellon University.

Friday, March 30, 2:00pm

Coffee at 1:45pm

ACES 2.402

Have I mentioned it?

Dr. Renata Vieira   [homepage]

Department of Inofrmatics
UNISINOS
Sao Leopoldo, Brazil

Coreference resolution is a well known NLP task, relevant for the more general task of information extraction, among others. A related problem is the problem of identifying if a referring expression is introducing a new entity in the text or if it follows previously mentioned ones. Definite descriptions are a type of referring expressions which are highly ambiguous between these two roles. In fact they sometimes lie in between these two, when mentioning a new entity whose interpretation is anchored in given ones. When developing a classifier that distinguishes between these many roles of definite descriptions we face not only the old AI problem of grasping common sense and world knowledge but also the problem of unbalanced data. In this talk I will present some experiments dealing with these problems and I will mention some NLP tasks that the classifier is useful for.

About the speaker:

Renata Vieira is Professor in the Computer Science Department at Universidade do Vale do Rio dos Sinos, Sao Leopoldo, Brazil. She received her PhD from the University of Edinburgh in 1998. Her research interests cover issues in computational linguistics, artificial intelligence, including natural language understanding, discourse processing, agent communication, knowledge representation, ontologies and the semantic web.

Thursday, April 5, 1:30pm

Coffee at 1:15pm

ACES 2.402

Active set methods for large-scale support vector machines

Katya Scheinberg   [homepage]

Mathematical Sciences Dept.,
T.J. Watson Research Center, IBM

In the past decade one of the most popular approaches to solving classification problem in machine learning is support vector machines (SVM). At the core of the SVM approach lies a convex quadratic programming (QP) problem. This approach has been shown to be very efficient on certain applications in terms of resulting generalization error. There also have been numerous implementations targeting large-scale data sets. Most of these implementations, however, use inferior optimization methods to solve the convex QP, due to inherent computational cost of higher order methods. We will discuss how exploiting the special structure of SVM QPs leads to a competitive implementation of an classical active set method for convex QPs. We will also discuss the advantages on active set methods over first order methods used before. One of the advantages is that the method readily adapts to the parametric mode which computes the entire regularization path for SVM. The method is similar to that described in the paper by Hastie, Rosset, Tibshirani and Zhu. We will compare this method to a dual active set method which computes just one solution on a regularization path. In theory, this is as hard as computing the entrie regularization path, but in practice this is not so. We will describe the challenges of parametric active set method, present computational comparison of the two methods on large-scale classification problems and discuss possible approach of reducing the computational time by computing an approximate regularization path.

About the speaker:

Katya Scheinberg was born in Moscow, Russia. After 4 years of undergraduate study in applied mathematics at Moscow State University, she joined the Ph.D. program in the Department of Industrial Engineering and Operations Research at Columbia University. She received her Ph.D. in 1997. Her thesis work was dedicated to various aspects of semidefinite programming and interior point methods. During the last two years in her Ph.D. program she was awarded an IBM Fellowship for graduate Students and began collaboration on derivative free methods for general nonlinear optimization. She is currently employed as a Research Staff Member at the Mathematical Sciences Department at T.J. Watson Research Center, IBM where she has been since 1997. Her research interests include theoretical and implementational issues of derivative free optimization (she released an open source software called DFO and is currently working on a book on the subject), numerical stability and efficient algorithms for linear algebra in interior points methods, parametric optimization, conic linear optimization, large-scale support vector machines (she is the author of an open source software called SVM-QP.) She has been recently involved in an effort to bring together the optimization and machine learning community and organized several workshops dedicated to the subject.

Friday, April 20, 11:00am

Coffee at 10:45am

ACES 6.304

Advancing the Theory and Practice of Model-based Reinforcement Learning

Dr. Michael L. Littman   [homepage]

Rutgers University

[Sign-up schedule for individual meetings]

Reinforcement learners seek to minimize sample complexity, the amount of experience needed to achieve adequate behavior, and computational complexity, the amount of computation needed per experience. Focusing on these two issues, we have been developing theoretically motivated algorithms that exhibit practical advantages over existing learning algorithms. I will present some of my lab's more recent theoretical accomplishments as well as some video footage of robots learning.

About the speaker:

Michael L. Littman directs the Rutgers Laboratory for Real-Life Reinforcement Learning (RL3) and his research in machine learning examines algorithms for decision making under uncertainty. After earning his Ph.D. from Brown University in 1996, Littman worked as an assistant professor at Duke University, a member of technical staff in AT&T's AI Principles Research Department, and is now an associate professor of computer science at Rutgers. Both Duke and Rutgers awarded him teaching awards and his research has been recognized with three best-paper awards on the topics of computer crossword solving, complexity analysis of planning, and efficient reinforcement learning. He served on the executive council of the American Association for Artificial Intelligence, and is an advisory board member of the Journal of Artificial Intelligence Research and an action editor of the Journal of Machine Learning Research.

Friday, April 27, 11:00am

Coffee at 10:45am

TAY 3.128

Vision-Based Behavior Control for Underwater Robotics

Dr. Gregory Dudek   [homepage]

McGill University

This talk discusses an ongoing research effort regarding the development of autonomous underwater vehicles, with particular emphasis on vision-based sensing. We have been developing an underwater vehicle for several applications, notably the environmental assessment of coral reefs habitats. Semi- autonomous behavior underwater is especially challenging since it combined 6 degree of freedom mobility, restricted communications, hard real-time constraints and unstructured environments. I will describe the system design of a small underwater and amphibious robot that uses computer vision as its principal sensing modality, and some of the ongoing challenges we have encountered. This includes an outline and discussion of how to accomplish operator control of the vehicle using a vision- based human-robot interface. The exploits a combination of a symbol-recognition system with a gestural inference and a special- purpose visual language. We also make use of Markov Random Fields for color correction (and, we hope, for scene reconstruction) I will comment on the use of physical feedback for behavior control and the development of a vision-based user interface. This is join work with doctoral candidates Philippe Giguere and Junaed Sattar, as well as Anqi Xu and out colleagues at York University led by Michael Jenkin.

About the speaker:

Gregory Dudek is a Professor with the School of Computer Science, and an Associate member of the Department of Electrical Engineering at McGill University. He is the Director of McGill's Research Center for Intelligent Machines, a 20 year old inter-faculty research facility. In 2002 he was named a William Dawson Scholar (an honorary chair). He directs the McGill Mobile Robotics Laboratory.

He has recently been on the organizing and/or program committees of Robotics: Systems and Science, the IEEE International Conference on Robotics and Automation (ICRA), the IEEE/RSJ International Conference on Intelligent Robotics and Systems (IROS), the International Joint Conference on Artificial Intelligence (IJCAI), Computer and Robot Vision, IEEE International Conference on Mechatronics (ICM2005) and International Conference on Hands-on Intelligent Mechatronics and Automation (HIMA2005) among other bodies. He is president of CIPPRS, the Canadian Information Processing and Pattern Recognition Society, an ICPR national affiliate.

He was on leave in 2000-2001 as Visiting Associate Professor at the Department of Computer Science at Stanford University and at Xerox Palo Alto Research Center (PARC). He obtained his PhD in computer science (computational vision) from the University of Toronto, his MSc in computer science (systems) at the University of Toronto and his BSc in computer science and physics at Queen's University.

He has published over 150 research papers on subjects including visual object description and recognition, robotic navigation and map construction, distributed system design and biological perception. This includes a book entitled "Computational Principles of Mobile Robotics" co-authored with Michael Jenkin and published by Cambridge University Press. He has chaired and been otherwise involved in numerous national and international conferences and professional activities concerned with Robotics, Machine Sensing and Computer Vision. He research interests include perception for mobile robotics, navigation and position estimation, environment and shape modelling, computational vision and collaborative filtering.

Monday, April 30, 11:00am

Coffee at 10:45am

ACES 2.302 (Avaya)

Planning in Uncertain Worlds: Exploration and Information Gathering

Dr. Nicholas Roy   [homepage]

Massachusetts Institute of Technology

[Sign-up schedule for individual meetings]

Decision making in uncertain and incomplete models is an essential capability of robots operating in natural, dynamic domains. Separating model learning and planning into two distinct processes simplifies both problems, but prevents the planner from deliberately learning more to improve its own performance. In my group we have developed two approaches for planning in the information space of models; these algorithms allow a robot to generate plans that are robust to model errors while planning to learn more about the world. I will present results of our work in the domains of robot navigation and human-robot dialogue management.

About the speaker:

Nicholas Roy is the Boeing Assistant Professor in the Department of Aeronautics & Astronautics at the Massachusetts Institute of Technology. He received his Ph. D. in Robotics from Carnegie Mellon University, Pittsburgh in 2003. He is a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. His research interests include autonomous systems, mobile robotics, human-computer interaction, decision-making under uncertainty and machine learning.

Monday, August 6, 11:00am

Coffee at 10:45am

ACES 6.304

The Voynich Manuscript

Dr. Kevin Knight   [homepage]

University of Southern California

The medieval Voynich Manuscript has been called the most mysterious document in the world. Its pages contain bizarre drawings of strange plants and astrological diagrams, as well as an undeciphered script of 20,000 running words, written in a character set that has never been seen elsewhere. Its origin is also controversial, with many theories abounding. I will describe the document, show samples, explain where it may have come from, and present some properties of the text and experiments with it. This will more of a history/mystery talk than a computer science talk.

About the speaker:

Kevin Knight is a Senior Research Scientist and Fellow at the USC/Information Sciences Institute. He is also a Research Associate Professor in the Computer Science Department at USC. He received his Ph.D. from Carnegie Mellon University in 1991 and his BA from Harvard University in 1986. He is co-author, with Elaine Rich, of the textbook "Artificial Intelligence". His main research interests are statistical natural language processing, machine translation, natural language generation, and decipherment.

Past Schedules

Fall 2005 - Spring 2006

Spring 2005

Fall 2004

Spring 2004

Fall 2003

Spring 2003

Fall 2002

Spring 2002

Fall 2001

Spring 2001

Fall 2000

Spring 2000