tal@vor.cs.cornell.edu (Doron Tal) (03/08/91)
Artificial intelligence is split in two vastly different paths: symbolic computation (e.g., prolog, theorem-proving, planning..) and connectionst computation (neural networks, PDP..). I am hoping to start a project which combines the advantages of both approaches. Although I have some specific ideas, at this point I am interested in anything anyone has heard about attempts to combine neural-networks with classic-AI theorem-provers, planners, and the like. The kinds of studies I am looking for are the ones where a neural-network is used to assist some kind of expensive exponential search, such as planning or theorem-proving --not the other way around, where a theorem- prover is used to control different NN modules (this other direction seems less interesting because it still maintains the gap between the two approaches to AI). If you email me references I promise to post a summary soon. Thanks, --Doron. (tal@cs.cornell.edu)