keithd@cadovax.UUCP (Keith Doyle) (05/15/85)
[..........] Anyone on the net using John Fogel's "Artificial Intellegence Through Simulated Evolution" as a basis for R&D work? One of our guys here who dosen't have his UNIX account yet is interested in learing more about this approach to AI, and was wondering if there is anyone out there who's interested in communicating on the subject. Keith Doyle # {ucbvax,ihnp4,decvax}!trwrb!cadovax!keithd
holte@brueer.UUCP (Robert Holte) (05/27/85)
> Anyone on the net using John Fogel's "Artificial Intellegence Through > Simulated Evolution" as a basis for R&D work? > > Keith Doyle > {ucbvax,ihnp4,decvax}!trwrb!cadovax!keithd There are two different approaches to simulated evolution: (1) evolution by mutation and selection, and (2) evolution by selective reproduction. Type-1 systems are receiving considerable attention these days, in the guise of simulated annealing and Boltzmann machines. Some mainstream AI learning systems employ variations of Type-1 evolution, the systems of Doug Lenat being the clearest example. I believe Fogel et al. took a type-1 approach to simulated evolution. Type-2 systems are less well known. The principal figure here is John Holland (U. of Michigan at Ann Arbor). He and his students have been studying the theory and application of "Genetic Algorithms" (Holland's term for type-2 systems) since the 60's. Theoretically, genetic algorithms have several significant advantages over type-1 systems. In some applications, practical considerations can favour type-1 systems despite their theoretical shortcomings. Here at Brunel, Michael Fourman (same addresses as me but "michael" instead of "holte") has begun experimenting with Genetic Algorithms for some aspects of VLSI design: we are quite impressed with their behaviour on a simplified symbolic layout task. There will be a conference on Genetic Algorithms and Their Applications in late July at Carnegie-Mellon: for information about this conference contact Stephen F. Smith (ARPA address: sfs@cmu-ri-isl1). I would like to make contact with anyone working with learning systems, evolutionary ot otherwise. - Rob Holte ARPA: holte%brueer@ucl-cs.arpa UUCP: ...!mcvax!ukc!reading!brueer!holte REFERENCES: John H. Holland, "Adaptation in Natural and Artificial Systems," U. of Michigan Press, Ann Arbor, 1975 John H. Holland, pp. 263-293 in "Progress in Theoretical Biology" vol. 4, Robert Rosen & Fred Snell (eds), Academic Press, 1976 Douglas B. Lenat, pp. 243-306 in "Machine Learning", R.S. Michalski, J.G. Carbonell & T.M. Mitchell (eds), Tioga Press, 1983