Generalized Key-Value Memory to Flexibly Adjust Redundancy in Memory-Augmented Networks

TL;DR

This paper introduces a flexible key-value memory system for neural networks that can adapt redundancy to balance robustness and resource use, especially suited for in-memory hardware with non-idealities.

Contribution

It proposes a generalized key-value memory that decouples memory size from support vectors, allowing dynamic redundancy control without retraining.

Findings

Mitigates up to 44% nonidealities in hardware

Maintains accuracy with fewer resources

Enables adaptive robustness in memory-augmented networks

Abstract

Memory-augmented neural networks enhance a neural network with an external key-value memory whose complexity is typically dominated by the number of support vectors in the key memory. We propose a generalized key-value memory that decouples its dimension from the number of support vectors by introducing a free parameter that can arbitrarily add or remove redundancy to the key memory representation. In effect, it provides an additional degree of freedom to flexibly control the trade-off between robustness and the resources required to store and compute the generalized key-value memory. This is particularly useful for realizing the key memory on in-memory computing hardware where it exploits nonideal, but extremely efficient non-volatile memory devices for dense storage and computation. Experimental results show that adapting this parameter on demand effectively mitigates up to 44% nonidealities, at equal accuracy and number of devices, without any need for neural network retraining.