Multi-Output Associative Memory for Content Addressable Information Retrieval

EasyChair Preprint 2046

Abstract

This paper introduces a multi-output associative memory model that can efficiently store auto-associative and hetero-associative pattern associations for content addressable information retrieval. This biologically inspired model can not only store multi-output associations for a single input but also permits the traversal of the stored memories for that particular input - both breadth-wise and depth-wise traversal in a highly connected graph of associated memories thus making the model a good candidate for multi-modal information retrieval due to the variety and flexibility in its search. We show that the model gives a high average testing accuracy during recall of lower-dimensional pattern association vectors.

Keyphrases: Associative Memory Networks, Content addressable information retrieval, Information Retrieval, neural networks, pattern recognition

@booklet{EasyChair:2046,
  author    = {Sidharth Anupkrishnan and Arjun Sharma},
  title     = {Multi-Output Associative Memory for Content Addressable Information Retrieval},
  howpublished = {EasyChair Preprint 2046},
  year      = {EasyChair, 2019}}