DYNAMIC PAGE -- HIGHEST POSSIBLE CLASSIFICATION IS
TOP SECRET // SI / TK // REL TO USA AUS CAN GBR NZL
(U//FOUO) Advanced Analysis Seminar: Computational Neurobiology
FROM:
Office of Tradecraft for Analysis (S2113)
Run Date: 04/05/2004
FROM:
Office of Tradecraft for Analysis (S2113)
(U//FOUO) You are invited to the following presentation:
Topic: Mechanization of Intelligence -- Computational Neurobiology
Speaker: Dr. Hecht-Nielsen
Affiliation: Fair Isaac Corporation
Date/Time: April 6, 2004 [Tuesday] 1000-1200
Location: Ops2B Conference Ctr, 2B4118-3
(U//FOUO) The topic of this lecture is the description of the underlying mathematical basis of
cognition. The main thrust of this talk will cover Confabulation (outlined in pages 2-11 in the
document below).
(U) Here is a summary of confabulation quoted from the paper referenced below:
"The main research program of Artificial Intelligence has been to find some sort of reasoning
process that can account for animal cognition. Confabulation, a symbolic prediction technique
introduced here, is the antithesis of this program. Instead of trying to reason from assumed
facts to conclusions, it uses a strange from of knowledge to rule out unreasonable outcomes;
leaving a core set of not-unreasonable possibilities is called an expectation. Then, employing the
same knowledge in a different way, it identifies the highest quality conclusions in the
expectation. Unlike reasoning, which must employ highly refined and specific knowledge that is
difficult and expensive to obtain, confabulation uses a type of knowledge (conditional
probabilities between pairs of symbols) that is simple and easy to obtained in the required vast
quantities. The recently discovered neuronal Marder/Turigiano/Hebb learning principle produces
this kind of knowledge. Confabulation is proposed as the underlying mechanism of all vertebrate
cognition."
(U) Dr. Hecht-Nielsen proposes that these discoveries will profoundly impact how artificial
intelligence techniques are applied to a wide variety of problems.
(U) The Institute for Neural Computation's external Web site is: http://inc2.ucsd.edu
(U) Dr. Hecht-Nielsen's Paper can be accessed on the external Web at :
http://inc2.ucsd.edu/pdfs/UCSD_INC_TechReport0401.pdf
"(U//FOUO) SIDtoday articles may not be republished or reposted outside NSANet
without the consent of S0121 (DL sid_comms)."
DYNAMIC PAGE -- HIGHEST POSSIBLE CLASSIFICATION IS
TOP SECRET // SI / TK // REL TO USA AUS CAN GBR NZL
DERIVED FROM: NSA/CSSM 1-52, DATED 08 JAN 2007 DECLASSIFY ON: 20320108