Tutorials
There will be seven tutorials held at the 23rd International Conference on Machine Learning (ICML-2006), on Sunday, June 25, 2006 at Carnegie Mellon University, Pittsburgh, Pennsylvania, U.S.A.
All tutorials have a minimum of 20 registrants to take place.
Tutorials
Sunday AM (~4 hours)
T1:
Grammar Induction:
Techniques and Theory
T2:
Online
Learning: Way Beyond Binary Classification [Cancelled as the
presenter is unable to attend for personal reasons]
T3: Learning
Automata as a Basis for Multiagent Reinforcement Learning
T4:
Learning
Representation and Behavior: Manifold and Spectral Methods for Reinforcement
Learning and Markov Decision Processes
Sunday PM (~4 hours)
T5:
Automatic
Inductive Programming
T6:
Reinforcement Learning
Theory
T7:
Information
Extraction, Theory and Practice
Descriptions
Sunday AM (~4 hours)
T1:
Grammar Induction:
Techniques and Theory
Colin de la Higuera and Tim Oates
The field of grammar induction is concerned with inferring grammatical models of the generative structure underlying sequential data. Both the theory and practice of grammar induction are well-developed, and have found applications in a wide variety of domains. This tutorial will provide an introduction to the theory, practice, and open problems in grammar induction. Among the topics that will be covered are models of learnability; inference of finite state automata and context-free grammars; learning from trees, graphs and other structured data; stochastic finite automata and grammars; inference techniques; positive and negative learnability results; and applications. Attendees will get a broad overview of both the theory and practice of grammar induction, with numerous examples of how both impact real-world applications.
T2:
Online
Learning: Way Beyond Binary Classification
Koby Crammer
Cancelled as the presenter is unable to attend for personal reasons
T3:
Learning Automata as a Basis for
Multiagent Reinforcement Learning
Ann Nowé, Katja Verbeeck, and Karl Tuyls
Learning Automata (LA) are adaptive decision making devices suited for operation in unknown environments. Originally they were developed in the area of mathematical psychology and used for modeling observed behavior. In its current form, LA are closely related to Reinforcement Learning (RL) approaches and are most popular in the area of engineering. Since LA combine fast and accurate convergence with low computational complexity, they are applied to a broad range of modeling and control problems. However, the intuitive, yet analytically tractable concept of a learning automaton is also very suited as a theoretical basis for Multiagent Reinforcement Learning (MARL). LA are updated strictly on the basis of the response of the environment, and not on the basis of any knowledge regarding other automata, i.e. nor their strategies, nor their feedback. As such LA agents are very simple. This tutorial will provide to the student the fundamentals of (Multiagent) Reinforcement Learning, according to the viewpoint of LA theory. We will discus Learning Automata, from a single automaton model acting in a simple stationary random environment to a distributed automata model interacting in a complex environment. The latter being the setting most relevant to the field of Multi-Agent Learning. Moreover, the link with the solution concepts proposed in (evolutionary) game theory will also be covered.
T4:
Learning
Representation and Behavior: Manifold and Spectral Methods for Reinforcement
Learning and Markov Decision Processes
Sridhar Mahadevan and Mauro Maggioni
This tutorial surveys new emerging connections between research in manifold and spectral learning, and novel approximation and decomposition techniques for solving Markov decision processes and reinforcement learning. The tutorial will also provide a detailed introduction to diffusion wavelets, a recently developed class of multiresolution manifold learning methods which are not well known in the machine learning community. Manifold and spectral learning methods hold the promise of a new generation of powerful tools for solving MDPs and RL, including ways of approximating value functions that have piecewise smoothness properties with respect to the geodesic distances on the underlying manifold; faster methods of policy evaluation and novel variants of policy iteration where both representation and optimal policy can be simultaneously learned; algorithms for hierarchical reinforcement learning where the underlying hierarchy is automatically learned; novel approaches to transfer learning by transferring shared representations; and enabling reinforcement learning methods in the absence of (task-specific) rewards. The tutorial will be accesssible to researchers and graduate students working in any area of machine learning or related areas (robotics, statistics etc.). The tutorial will include a detailed introduction to the underlying mathematics, and will also cover a variety of algorithms and state of the art applications. The main ideas will be illustrated throughout using hands-on MATLAB demonstrations.
Sunday PM (~4 hours)
T5:
Automatic
Inductive Programming
Ricardo Aler Mur
Computers which can program themselves is an old dream of Artificial Intelligence, but only nowadays there is some progress of remark. From the point of view of Machine Learning, a computer program is the most complex structure that can be learned, pushing the final goal beyond decision trees or neural networks.There are currently many separate areas, working independently, related to automatic programming, both deductive and inductive. The main goal of this tutorial is to give the attendants a comprehensive view of the main areas related to the automatic induction of programs, like Genetic Programming, Induction of Functional Programs, ILP, and probabilistic methods. The tutorial will focus on the most promising approaches.
T6:
Reinforcement Learning
Theory
John Langford and Satinder Singh
There have been substantial contributions to the theory of reinforcement learning from the areas of machine learning, operations research, artificial intelligence and control theory. These advances are not yet collected and organized in any one text. This tutorial will present a comprehensive look at the theory of RL from the early results on the foundational algorithms such as temporal differences (TD) and Q-learning to the more recent results on generalization in RL. Topics covered will include results on exploration-exploitation, function approximation and multi-agent RL. In addition, some open questions and important new directions will be defined.
T7:
Information
Extraction, Theory and Practice
Ronen Feldman
Information Extraction (IE), is one of the most prominent techniques currently used in Text Mining. In particular, by combining Natural Language Processing tools, lexical resources and semantic constraints, it can provide effective modules for mining the biomedical literature, or to help in preventing terrorism. Complementary visualization tools enable the user to explore, check (and correct if required) the results of the Text Mining process effectively. As a first step in tagging documents, each document is processed to find (extract) Entities and Relationships that are likely to be meaningful and content-bearing. In "Relationships" we refer to Facts or Events involving certain Entities. A possible "Event" may be that a company has entered into a joint venture to develop a new drug. A "Fact" may be that a gene causes a certain disease. In this tutorial we will present the general theory of Information Extraction and will demonstrate several systems that use these principles to enable interactive exploration of large textual collections. We will present a general architecture for information extraction and will outline the algorithms and data structures behind the systems. The Tutorial will cover the state of the art in this rapidly growing area of research. Several real world applications of information extraction will be presented in the areas of business intelligence, competitive intelligence, bio information, and military intelligence.
Call for tutorial proposals is here