Special Issue of Computational Intelligence on "Learning to Improve Reasoning"

Seth Rogers (srogers@csli.stanford.edu)
Afzal Upal (afzal@eecs.utoledo.edu)

Machine Learning has made great strides in recent years in maturing as a cohesive research topic and producing real-world applications, but most of the progress has been in the sub-topics of classification and reactive control. However, machine learning also aims to contribute in more complex tasks that involve reasoning, planning and inference. There has been a resurgence of interest in this area, as evidenced by the successful symposium on "Reasoning and Learning in Cognitive Systems" held at Stanford on March 21-22, 2004. The special issue will gather a sampling of recent research on machine learning for complex, multi-step performance tasks and present a comprehensive picture of the field. The issue would contain fresh approaches to a variety of specific tasks that are not already covered in the archival literature. Taken as a whole, this will inspire researchers to renew efforts to study learning on these more complex tasks and extend the capabilities within the current reach of machine learning systems.

Research Areas: We welcome original and previously unpublished papers (previous publication of partial results at a conference/workshop is allowed) that make substantial contributions to the area of learning for multi-step performance tasks (such as reasoning, planning, and inference). These include papers that (a) propose novel algorithms for learning to improve the performance of domain independent reasoning systems, (b) review, compare, and analyze different learning paradigms providing new insights such as relating domain features to solution features, (c) advance evaluation techniques for measuring the performance of learning-for-reasoning systems, and (d) analyze issues involved in deploying the learning for reasoning systems in real world applications. The topics of particular interest include, but are not limited to, the following:

Review Criteria: All papers will be reviewed by at least two experts. An ideal paper will clearly define the learning problem, describe the proposed learning algorithm in enough detail to allow replication by others, specify and motivate the performance measure(s), and detail evidence that supports conclusions drawn by the authors. All submissions should be clearly written and must discuss relationship of the proposed research to previously published work. Editors reserve the right to return a submission without review if it is deemed not to address issues of interest identified in this CFP or adhere to the formatting guidelines.

Formatting Guidelines: Manuscripts should conform to the formatting instructions found at Blackwell Publishing. Due to the short timeline we will not be able to review submissions that are more than 30 pages long.

Proposed Timeline:
August 1, 2004 Letter of intent to submit sent to srogers@csli.stanford.edu
September 1, 2004 full paper submission deadline
March 1, 2005 decision
August 26, 2005 publication

Please contact srogers@csli.stanford.edu or afzal@eecs.utoledo.edu for further information.