NICK@AI.AI.MIT.EDU (Nick Papadakis) (06/03/88)
Date: Thu, 2 Jun 88 04:34 EDT From: lugowski@resbld.csc.ti.com To: ailist@mc.lcs.mit.edu Subject: current connectionist literature, etcetera Responding to John Nagle's CURRENT connectionist literature inquiry: In my opinion, there is no good comprehensive book on current connectionist thinking. For one thing, folks are too busy going to conferences. For another, everyone has their own little garden to tend. Recent book content of interest includes "Neural Darwinism" (to judge from preprints) as well as the commentaries by Jim Anderson in "Neuroscience", a recent compendium of not-so-recent papers. You don't want to miss a not-yet-out MIT/Bradford Books book, Pentti Kanerva's 1984 thesis (CSLI 84-7), if you haven't read it yet. Other than that, I'd repeat the obligatory advice: monitor technical reports, the journals "Nature" and "Neural Networks" and the two connectionist mailing lists. To apply for subscription to those lists, send to: connectionists-request@q.cs.cmu.edu (sparsely firing connectionists) neuron-digest-request@csc.ti.com (everyone, sparse and otherwise) As for categorizing work, anything small-grained, bottom up and parallel probably can pass for connectionist. It's not the formalism, it's the claim, really: One must make massively parallel claims pertaining to massive parallelism. (Smirk, lest I get crucified.) Simulated annealing is "very connectionism". Some of the nicest connectionist work of late (Durbin & Willsaw, Cambridge, also stuff out of Los Alamos) has at least references to simulated annealing as benchmark. The trick one would like to see done is casting simulated annelaing as a localized computation *without* the closed-form cost function or globally computed energy -- everything strictly "grassroots". As for tensor calculus, the very idea appears contra connectionism, I hold with those who would like to see discrete, adpative and local formalisms take over the domains historically ceded to 19th century's closed-form mathematical analysis and its applications. Tensor calculus? Sure, but check them determinants at the bar, pardner... [Above opinions are strictly mine.] -- Marek Lugowski lugowski@resbld.csc.ti.com lugowski@ngstl1.ti.com marek@iuvax.cs.indiana.edu