Holographic Reduced Representations

Source: IEEE / DOI Author: Tony A. Plate Date: 1995

Summary

Plate introduces holographic reduced representations as fixed-width distributed representations that can encode compositional symbolic structure using circular convolution and approximate unbinding. The key move is to keep role–filler binding inside a single vector space rather than switching out to a separate symbolic substrate.

Why it matters here

HRR is one of the cleanest classical precedents for a neural-native program core that is compact, compositional, and noise-tolerant. If NNPL wants binding and unbinding operations with graceful degradation, this is the obvious ancestor.