Abstract
This paper aims to place neural networks in the context of boolean circuit complexity. We define appropriate classes of feedforward neural networks with specified fan-in, accuracy of computation and depth and using techniques of communication complexity proceed to show that the classes fit into a well-studied hierarchy of boolean circuits. Results cover both classes of sigmoid activation function networks and linear threshold networks. This provides a much needed theoretical basis for the study of the computational power of feedforward neural networks.
Original language | Undefined |
---|---|
Pages (from-to) | 971-977 |
Journal | Neural networks |
Volume | 5 |
Issue number | 6 |
DOIs | |
Publication status | Published - 1992 |
Keywords
- IR-57459