Learning from Sequential and Temporal Data

In conjunction with the
Seventeenth International Conference on Machine Learning
Stanford University, June 29 - July 2, 2000

Description

Learning from sequential and temporal data provides a challenge for researchers working in a variety of application areas. This workshop will provide a forum for discussing original research results on this topic, and will establish an environment in which researchers from diverse application areas can share methods and identify common issues.

For example, the inputs to many systems that learn about natural language are cast as variable-length sequences of discrete symbols, and similar encodings are common in bioinformatics and intrusion detection. Other domains, including motor control, speech recognition, and financial prediction, typically involve measurements of continuous attributes that vary over time.

Researchers in both discrete and continuous domains have used quite different representational frameworks to make learning tractable. These include formalisms such as:

fixed-length and variable-length N-grams
finite-state machines and Markov models
context-free and context-sensitive grammars
attribute values computed from moving windows
other transformations into a fixed attribute set

The goals of this workshop are to present recent research results on learning from sequential and temporal data, to discuss issues relevant to researchers working in this area, and to generate new ideas and collaborations among participants.

More specifically, it will examine relations among these different representational frameworks and encourage exchange of expertise about the best ways to use each approach in particular application areas. Also, since sequential and temporal domains raise novel issues in evaluation, it will also explore new techniques, whether experimental or theoretical, designed to evaluate learning systems of this sort.

Submission Information

We invite potential participants to submit a one-page abstract that describes their research on the workshop theme. Topics of interest include learning for purposes of:

parsing and generation of natural language
recognition and execution of motor skills
prediction of biological structure and function
annotation of biological sequences
detection of anomalies and intrusions
prediction of financial time series
recognition and generation of speech

However, we also welcome researchers who focus on other topics that also involve learning from sequential and temporal data.

Prospective speakers should email their abstract in ascii format to Stefan Schroedl by the submission deadline. This should include the title of the talk, as well as the speaker's name, affiliation, postal address, email address, telephone number, and fax number.

This one-day workshop will consist of research presentations on different application areas and distinct representational frameworks, but will highlight issues relevant to all researchers focusing on learning from sequential and temporal data. It will also include a panel discussion be designed to stimulate discussion on core issues and generate ideas for continued work in this area.

Important Dates

Submission deadline: June 5, 2000
Notification of acceptance: June 9, 2000
Workshop date: July 2, 2000

Preliminary schedule and abstracts

Organizing Committee

Arun Jagota, University of California, Santa Cruz (jagota@cse.ucsc.edu)
Vasant Honavar, Iowa State University (honavar@cs.iastate.edu)
Stefan Schroedl, DaimlerChrysler (schroedl@rtna.daimlerchrysler.com)

Call for Participation