CFP

Machine Learning as Experimental Philosophy of Science

Machine learning studies inductive strategies in
algorithms. The philosophy of science investigates
inductive strategies as they appear in scientific
practice. Although the two disciplines have developed
largely independently, they share many of the same
issues. This is slowly coming to be recognized, as
evidenced in the annual Uncertainty in AI and AI and
Statistics conferences. This workshop will explore the
extent to which the methods and resources of philosophy
of science and machine learning can inform one another.

In Computational Philosophy of Science (1988) Paul
Thagard presented a challenge to the philosophical
community: philosophical theories of scientific method,
if they are worth their salt, should be implementable
as computer programs. In this workshop we will address
this challenge and also the inverse challenge to
machine learning researchers: both machine learning
algorithms and methods for evaluating machine learning
algorithms should be implementations of sensible
approaches to philosophy of science. Machine learning
researchers have only recently discovered the relevance
of statistics and philosophical views on the
foundations of statistics to evaluating the performance
of their systems; we hope this workshop will carry that
discussion further.

The workshop will therefore focus on such questions as:

How machine learning experiments and results can inform
our knowledge of scientific inductive discovery?
What theoretical results in computational learning can
be useful to understand scientific methods?
How accounts of confirmation, explanation, discovery
and theoretical unification developed in the philosophy
of science area can be used to develop automatic
learning systems?
How induction is to be assessed: is empirical adequacy
(predictive accuracy) enough both to account for
scientific dynamics and to evaluate automated induction
performance?
Is there a substantial difference between scientific
reasoning as conceived in the philosophy of science and
in artificial intelligence?
Is scientific method mechanisable? Are scientific
practices algorithmic?

Venue

This workshop is one of a number of workshops jointly
sponsored by the 12th European Conference on Machine
Learning (ECML'01) 5th European Conference on
Principles and Practice of Knowledge Discovery in
Databases (PKDD'01). Have a look at their workshop
program.

Invited Speakers
Professor Kevin Kelly (CMU, Philosophy), author of "The
Logic of Reliable Inquiry (Oxford, 1996). His recent
work concerns reliable belief revision, the solution of
methodological regresses, and efficient convergence.
Dr Peter Flach (Bristol, Computer Science), co-editor
of Abduction and Induction: essays on their relation
and integration (Kluwer, 2000) and co-organiser of
workshops on Abductive and Inductive Reasoning in AI at
ECAI'96, IJCAI'97 and ECAI'98.

Publication
Accepted papers will be published in the first instance
as workshop notes and on the web. Authors are invited
to revise their articles in the light of the
discussions at the workshop and submit them to a
special issue we have arranged with the Journal for
Experimental and Theoretical Artificial Intelligence.
Important Dates

Papers due:  8 June 2001
Notification:  25 June 2001
Camera-ready due:  13 July 2001
Workshop:  3 Sept 2001

Submission Instructions: contact hilanb@unb.br and
korb@csse.monash.edu.au

MLEPS Workshop
c/o Kevin B. Korb
School of Computer Science
Monash University
Clayton, VIC 3800
AUSTRALIA

Fax: +61 (03) 9905-5146
Workshop Organisers
Kevin Korb (Monash University, Australia)
Hilan Bensusan (Bristol University, UK)