Learning by Choice of Internal Representations
Tal Grossman
Ronny Meir
Eytan Domany
Department of Electronics, Weizmann Institute of Science
Rehovot 76100 Israel
Abstract
We introduce a learning algorithm for two-layer neural networks composed of binary linear threshold elements. Whereas existing algorithms reduce the learning process to minimizing a cost function over the weights, our method treats the internal representations as the fundamental entities to be determined. We perform an efficient search in the space of internal representations. When a correct set of internal representations is arrived at, the weights can be found by the local and biologically plausible Perception Learning Rule. No minimization of any cost function is involved. We tested our method on three problems: contiguity, symmetry, and parity. Our results compare favorably with those obtained using the backpropagation learning algorithm.
