A New Training Algorithm for Kanerva's Sparse Distributed Memory
This work addresses a limitation in memory models for AI systems, but it is incremental as it builds on an existing architecture.
The authors tackled the inefficiency of Kanerva's Sparse Distributed Memory (SDM) with non-random data by introducing a new training algorithm that enables efficient handling of such data and recognition of inverted patterns, achieving improved performance for structured inputs.
The Sparse Distributed Memory proposed by Pentii Kanerva (SDM in short) was thought to be a model of human long term memory. The architecture of the SDM permits to store binary patterns and to retrieve them using partially matching patterns. However Kanerva's model is especially efficient only in handling random data. The purpose of this article is to introduce a new approach of training Kanerva's SDM that can handle efficiently non-random data, and to provide it the capability to recognize inverted patterns. This approach uses a signal model which is different from the one proposed for different purposes by Hely, Willshaw and Hayes in [4]. This article additionally suggests a different way of creating hard locations in the memory despite the Kanerva's static model.