Uncertainty-based queries

[David Cohn]

Concerning the talk abstract "Uncertainty-Based Queries in Neural
networks" which appeared here a few days ago:

Quite a few people have pointed out that the use of measures of
uncertainty to gather new data is not a new idea. I must apologize if
my abstract made it appear that I thought it was. The idea can be
traced back at least as far as the 40's (A. Wald, Ann. Math. Stat.,
Vol 14, 1943). The field of optimal experiment design, as it has
become known to statisticians, is rather well developed. The most
complete treatment of which I am aware is Fedorov (Theory of Optimal
Experiments, Academic Press, 1972). Thanks also to Grace Wahba for
providing additional references.

It has been said, however, that the best new ideas in neural networks
are coming from the 60's and 70's -- that people in the field are
rediscovering the utility of tools (in statistics, control theory,
etc.) that were developed "ahead of their time" and had only limited
practical application before the advent of today's machine learning
algorithms.

My current research-in-progress, and the subject of the talk whose
abstract appeared here earlier, is a practical examination of how we
can take the techniques of optimal experiment design and apply them to
the problems facing neural network researchers today. (D. MacKay
addressed some aspects of this problem in his dissertation, and in a
paper in Neural Computation 4(4)).

As this *is* work in progress, I do not yet have a paper written on my
results so far, but I will be happy to discuss them with any
interested persons.

Again, thank you all for your comments. Yes, I am aware of the body of
results from optimal experiment design, and the point of my work is to
see how these results can be applied to (excuse the expression)
"practical problems" facing neural net researchers. I am sorry that my
abstract did not make that clear.

 -David Cohn				e-mail: cohn at psyche.mit.edu
  Dept. of Brain & Cognitive Science	phone:  (617) 253-8409
  MIT, E10-243
  Cambridge, MA 02139