Abstract

This work describes a segmented attractor network that records memories across different sets of information. Unlike typical attractor networks that can associate any given inputs with one another, the attractor network presented here tracks information across multiple sets where each set possesses one or more features. Hyperparameter analysis is performed on the network's recall capability by altering factors such as the network's size, the number of memories within the network, the amount of information required to perform recall, and the amount of feedback a synapse can provide. The analysis done shows an improved hit rate when increasing the features per set and allows the network to obtain a memory capacity higher than other standard attractor networks.

Date

July 23, 2019

Authors

Alexander Jones, Rashmi Jha, Ajey P Jacob, Cory Merkel

Book

Proceedings of the International Conference on Neuromorphic Systems

Pages

1-8