close this bookVolume 4: No. 30
View the documentFunding news
View the documentBusiness news
View the documentInternet news
View the documentJob opportunities
View the documentDiscussion groups
View the documentArtificial brains
View the documentScholarly publishing
View the documentComputists' news

Cellular automata (CA) in CA machines (CAMs) can be evolved rapidly using genetic algorithms. Toffoli's group at MIT has a CAM8 machine that can update 200M cells/second, with prospects of 10^22 by the end of the decade. (However, the MIT group is being terminated -- against Minsky's objections.) Hugo de Garis at ATR is working (with NTT) toward brain models with thousands of neural modules in two years, then millions. 2D versions with 11,000 rules have worked well so far. [degaris@hip.atr.co.jp, sci.nanotech, 6/17/94.]

Carver Mead's laboratory has announced reliable analog on-chip learning "that will make possible adaptive mechanisms and learning at all levels of processing, as occurs in biological systems." The NSF-sponsored 1st Annual Telluride Workshop on Neuromorphic Engineering studied this and other steps toward artificial brains. Low-power analog VLSI chips already exist for low-level visual and auditory processing. Sensorimotor integration is being studied by Dana Ballard and Richard Andersen, and Rodney Douglas and Misha Mahowald are working on cortical chips. Workshop results will be made available soon via FTP. [Terry Sejnowski (terry@salk.edu), connectionists, 7/22/94.]

On the other hand, jbower@smaug.bbb.caltech.edu notes that few connectionists attend neuroscience meetings. Connectionists have little understanding of, or interest in, neurobiology -- and we are still far from modeling the brain of a bee. For those who want to know more, Dave Beeman is organizing a new discussion list for computational neurobiology. Contact comp-neuro @smaug.bbb.caltech.edu to sign up. [connectionists, 7/28/94.]