Noisy Encoders

Author: Paul Munro (University of Pittsburgh)

Abstract: Just as autoencoding networks may offer important new approaches for developing efficient codes, they may also be applicable to developing codes with other useful application areas, such as error correction or encryption. Here, the role of noise in developing robust representations will be discussed (see Judd & Munro, 1993), with some speculative ideas regarding the role of noise in promoting generalization.

Judd and Munro (1993) Nets with unreliable hidden nodes learn error correcting codes. NIPS 5: 89-96.