Dangerous Ideas Seminar
Speaker: Tom Knight , MIT CSAIL
Date: November 2 2005
Time: 1:00PM to 2:00PM
Location: Stata - Kiva 32-G449
Host: Metin Sezgin, MIT CSAIL
Contact: Tevfik Metin Sezgin, 617-253-2663, mtsezgin@csail.mit.edu
Relevant URL: http://projects.csail.mit.edu/dangerous-ideas/dangerous/www/Building fault-tolerant circuits on highly unreliable substrates: Lessons from neurophysiology
Tom Knight
Leslie Valiant has proposed a representation of neural concepts which relies on sparse m out of n codes in a randomly, but weakly connected network of cells with high fan-in and high fan-out(*). In this talk I will discuss a proto-idea in which a similar representation is used to build logic circuits which are robust against failures of the nodes, interconnections, and even partitioning of the network. I speculate on how these ideas could be applied in the construction of inherently reliable hardware through the addition of another level to the engineering hierarchy: changing the representation of logic signals from being associated with activity of single wires to being represented by the collective activity of ensembles of wires.
* Valiant is not responsible for my bad ideas; the good ones are his.
His paper, which I recommend, is:
Valiant LG (2005) Memorization and association on a realistic neural
model, Neural Computation 17:527-55.
See other events that are part of Dangerous Ideas Seminar Series Fall 2005
See other events happening in November 2005
