|
|
|
Patrick C. Kyllonen
|
Pat Langley
|
 Patrick Kyllonen is
the Research Center Director for the New Constructs Center at
Educational Testing Service
(ETS) in Princeton, N.J. Before joining ETS in 1999, Dr. Kyllonen
received his B.A. from St. John's University and his Ph.D. from
Stanford University (1984). Dr. Kyllonen's research has focused on the
measurement of human abilities and on learning, cognition, and skill
acquisition. More recently,
he and New Constructs Center colleagues
have been investigating affective (e.g., personality, metacognitive) as
well as cognitive mediators of educational success and job performance,
along with associated new assessments (e.g., communication skills,
situational judgment) and delivery modes (e.g., multimedia, internet).
Dr. Kyllonen is author (with S. Irvine) of Generating Items for
Cognitive Tests: Theory and Practice, published in 2001 by Lawrence
Erlbaum Associates, Learning and Individual Differences: Process,
Trait, and Content Determinants (with P. L. Ackerman & R.D.
Roberts), published in 1999 by the American Psychological Association,
Extending Intelligence: Enhancement and New Constructs (with R. Roberts
and L. Stankov), which will be published in 2006 by Lawrence Erlbaum
Associates, and the forthcoming The Science of Item Generation:
Psychology, Psychometrics, & Practices (with S. Irvine), to be
published by Hogrefe. He is a Fellow of the American Psychological
Association, and has served on the editorial board of Intelligence: A
Multidisciplinary Journal, and Human Factors: The Journal of the Human
Factors and Ergonomics Society.
|
 Dr. Pat Langley serves as Director of the Institute for the
Study of Learning and Expertise, Consulting Professor of Symbolic
Systems at Stanford University, and Head of the Computational Learning
Laboratory at Stanford's Center for the Study of Language and
Information. He has contributed to the fields of artificial
intelligence and cognitive science for over 25 years, having published
200 papers and five books on these topics, including the text Elements
of Machine Learning. Professor Langley is considered a co-founder of
the field of machine learning, where he championed both experimental
studies of learning algorithms and their application to real-world
problems before either were popular and before the phrase `data mining'
became widespread.
Dr. Langley is a AAAI Fellow, he was founding Executive Editor of the
journal Machine Learning, and he was Program Chair for the Seventeenth
International Conference on Machine Learning. His research has dealt
with learning in planning, reasoning, language, vision, robotics, and
scientific knowledge discovery, and he has contributed novel methods to
a variety of paradigms, including logical, probabilistic,
and case-based learning. Dr. Langley's work on adaptive user interfaces
introduced unobtrusive ways to collect data about preferences and use
them to offer personalized services. His current research focuses on
methods for constructing explanatory process models in scientific
domains and on cognitive architectures for intelligent agents.
|
Evaluating
Transfer Learning in Physics
(and
other knowledge-rich problem-solving domains)
14:05 - 14:50
Transfer learning can be said to occur
whenever target (test) performance benefits from prior experience with
related source (training/study) material. A useful division is between
near and far transfer. In knowledge-rich problem-solving domains such
as those reflected by standardized tests, near transfer is invoked when
the target and source can be characterized by a common template. Far
transfer is invoked when target and source do not share a template, but
are similar at a deeper level-e.g., they share a solution equation,
which itself may have to be expressed in a very general way. In
this talk I will discuss a strategy we are currently implementing (in
collaboration with Cycorp, Inc. and Northwestern University) for
evaluating the efficacy of a transfer learning system in knowledge-rich
problem-solving domains, focusing on physics, as it is assessed in
standardized college-level tests (e.g., AP Physics, GRE Physics). The
evaluation strategy involves setting up an experimental design and
modeling the learning system's responses to physics
problems-quantifying the benefits of transfer with a statistical model
of learning curves-so as to be able to assess claims system designers
would like to make about the transfer learning capabilities of their
system. Some of the key theoretical issues are differentiating
knowledge from learning, item-specific from more general learning, and
distinguishing different kinds of near transfer. I will discuss our
system-the Math Test Creation Assistant-for automatically generating
physics problems from templates so as to be able to test near transfer.
I will also discuss our ideas for a non-template, model-based system
being developed by Paul Deane and Michael Flor capable of automatically
generating deep-level problem variants to test far transfer.
|
Transfer of
Knowledge in Cognitive Systems
09:05 - 09:50
In this talk, I review the notion of
transfer from cognitive psychology and consider its relevance to
artificial cognitive systems. I claim that transfer involves the
sequential reuse of knowledge, so that it assumes online learning, and
that transfer is primarily a structural phenomenon, so that only
systems which store knowledge in modular elements can exhibit it. I
also examine some dimensions of transfer from source to target problems
that influence the nature of learning curves in the target setting. I
illustrate these ideas using Icarus, a cognitive architecture for
physical agents, on transfer scenarios taken from Urban Combat and
General Game Playing, two testbeds that support the transfer of learned
knowledge. I close with some challenges in this area that deserve more
attention from the research community.
|
|
