About sequential learning (or interference)

[Danny L. Silver]

For me the significance of inteference in neurally inspired learning systems
is the message that an effective learner must not only be capable
of learning a single task from a set of examples but must also be
capable of effectively integrating variant task knowledge at a meta-
level.  This falls in line with McClelland's recent papers on consolidation
of hippcocampal memories into cortical regions; his "interleaved learning".
This is a delicate and complex process which undoubtedly occurs during sleep.
In tune with Sebastian Thrun and Tom Mitchell's efforts on "Life Long
Learning" I feel the next great step in learning theory will be the discovery
of methods which allow our machine learning algorthms to take advantage of
previously acquired task knowledge.

At UWO we have been investigating methods of storing neural net
task knowledge in an interleaved fashion with other, previously learned
tasks, so as to create an "experience database".  This database can then be
used to prime the initial weights of the neural net for a new task.
Thus far, studies on simple boolean logic tasks has shown promise.
Incremental learning is possible (with decreases in learning times by
1 or 2 orders of magnitude)), but dependent upon task order.  Thus one
of the key aspects of consolidation, so as to overcome interference,
appears to be a reordering of learned tasks.

Have others (besides those authors I have mentioned) tried methods of
task consolidation at a meta level?

... Danny
--

=========================================================================
=  Daniel L. Silver    University of Western Ontario, London, Canada    =
=                      N6A 3K7 - Dept. of Comp. Sci. - Office: MC27b    =
=  dsilver at csd.uwo.ca  H: (519)473-6168   O: (519)679-2111 (ext.6903)   =
=========================================================================

REF:

McClelland, J. & McNaughton B. & O'Reiily, R. "Why there are complemetary
learning sysetms in the hipocampus and neocortex: Insights from the successes
and failures of connectionist models of learning and memory".  Technical
Report PDP.CNS.94.1, Carnegie Mellon Univeristy and The University of Arizona,
March, 1994.

Thrun, S. & Mitchell T. "Lifelong Robot Learning". Techincal Report
IAI-TR-93-7, Universitat Bonn, Institut fur Informatik II, Germany,
July, 1993.

Thrun, S. "A Lifelong Learning Perspective for Mobile Robot Control";
Proceedings  of the IEEE Conference on Intelligent Robots and Systems,
Munich, Germnay, Sept, 1994.

Thrun, S. & Mitchell T. "Learning One More Thing". Techincal Report
CMU-CS-94-184, Carnegie Mellon University, Pittsburg, PA, Sept, 1994.