papers in neuroprose archive

[Sebastian Thrun]

Dear Connectionists,

this mail is to announce two new papers in Jordan Pollack's
neuroprose archive:

1) Explanation-Based Neural Network Learning for Robot Control
          by Tom Mitchell and Sebastian Thrun, to appear in: NIPS-5

2) Exploration and Model Building in Mobile Robot Domains
          by Sebastian Thrun, to appear in: Proceedings of the ICNN-93

Enclosed you find both abstracts and the (standard) instructions for
retrieval. Comments are welcome. 

Have fun,
Sebastian Thrun

----------------------------------------------------------------------
				   
     Explanation-Based Neural Network Learning for Robot Control
				   
	      Tom M. Mitchell (CMU, mitchell at cs.cmu.edu)
		 Sebastian B. Thrun (Bonn University,
		  thrun at uran.informatik.uni-bonn.de)


How can artificial neural nets generalize better from fewer examples?
In order to generalize successfully, neural network learning methods
typically require large training data sets.  We introduce a neural
network learning method that generalizes rationally from many fewer
data points, relying instead on prior knowledge encoded in previously
learned neural networks.  For example, in robot control learning tasks
reported here, previously learned networks that model the effects of
robot actions are used to guide subsequent learning of robot control
functions.  For each observed training example of the target function
(e.g. the robot control policy), the learner *explains* the
observed example in terms of its prior knowledge, then *analyzes*
this explanation to infer additional information about the shape, or
slope, of the target function. This shape knowledge is used to bias
generalization in the learned target function.  Results are presented
applying this approach to a simulated robot task based on
reinforcement learning.
                                   (file name: mitchell.ebnn-nips5.ps.Z)



	Exploration and Model Building in Mobile Robot Domains

			  Sebastian B. Thrun
	 (Bonn University, thrun at uran.informatik.uni-bonn.de)

I present first results on COLUMBUS, an autonomous mobile robot.
COLUMBUS operates in initially unknown, structured environments.  Its
task is to explore and model the environment efficiently while
avoiding collisions with obstacles.  COLUMBUS uses an instance-based
learning technique for modeling its environment. Real-world
experiences are generalized via two artificial neural networks that
encode the characteristics of the robot's sensors, as well as the
characteristics of typical environments the robot is assumed to face.
Once trained, these networks allow for the transfer of knowledge
across different environments the robot will face over its lifetime.
COLUMBUS' models represent both the expected reward and the confidence
in these expectations.  Exploration is achieved by navigating to low
confidence regions.  An efficient dynamic programming method is
employed in background to find minimal-cost paths that, executed by
the robot, maximize exploration.  COLUMBUS operates in real-time. It
has been operating successfully in an office building environment for
periods up to hours.
                                    (file name: thrun.robots-icnn93.ps.Z)


----------------------------------------------------------------------


Postscript versions of both papers may be retrieved from Jordan
Pollack's neuroprose archive. If you have a Postscript printer, please
follow the following instructions below. If not, feel free to contact
me (thrun at uran.informatik.uni-bonn.de) for a hardcopy.

	unix>           ftp archive.cis.ohio-state.edu

	ftp login name> anonymous
	ftp password>   xxx at yyy.zzz
	ftp>            cd pub/neuroprose
	ftp>		bin
	ftp>            get mitchell.ebnn-nips5.ps.Z
	ftp>            get thrun.robots-icnn93.ps.Z
	ftp>            bye

	unix>           uncompress mitchell.ebnn-nips5.ps.Z
	unix>           uncompress thrun.robots-icnn93.ps.Z
	unix>           lpr mitchell.ebnn-nips5.ps.Z
	unix>           lpr thrun.robots-icnn93.ps.Z


Note that the second file is rather long. Some printers have
limitations for the document size to be printed. In this case, it might
be necassary to circumvent this limitation by using "lpr" with the "-s"
option at that machine the printer is physically connected to.