Title :
The perceptron strikes back
Author :
Beigel, Richard ; Reingold, Nick ; Spielman, Daniel
Author_Institution :
Yale Univ., New Haven, CT, USA
fDate :
30 Jun-3 Jul 1991
Abstract :
It is shown that every AC0 predicate is computed by a low-degree probabilistic polynomial over the reals. It is demonstrated that circuits composed of a symmetric gate at the root with AND-OR subcircuits of constant depth can be simulated by probabilistic depth-2 circuits with essentially the same symmetric gate at the root and AND gates of small fanin at the bottom. In particular, every language recognized by a depth-d AC0 circuit is decidable by a probabilistic perceptron of size 2 to the power O(log4d n) and of order O(log4d n) that uses O(log3 n) probabilistic bits. As a corollary, the authors present a new proof that depth-d AND-OR circuits computing the parity of n binary inputs require size 2 to the power nΩ(1/d)
Keywords :
computational complexity; decidability; logic circuits; threshold logic; AC0 predicate; AND gates; AND-OR subcircuits; binary inputs; constant depth; decidable; fanin; language; parity; probabilistic bits; probabilistic depth-2 circuits; probabilistic perceptron; probabilistic polynomial; reals; root; size 2; symmetric gate; Circuit simulation; Computational modeling; Computer errors; Computer science; Educational institutions; Electronic mail; Neural networks; Polynomials; Upper bound; Writing;
Conference_Titel :
Structure in Complexity Theory Conference, 1991., Proceedings of the Sixth Annual
Conference_Location :
Chicago, IL
Print_ISBN :
0-8186-2255-5
DOI :
10.1109/SCT.1991.160270