Empirical Analysis of Limits for Memory Distance in Recurrent Neural Networks

TL;DR

This paper empirically investigates the maximum memory distance of RNNs, LSTMs, and GRUs, revealing a hard limit below the information-theoretic maximum for recognizing relations between data points over time.

Contribution

It provides a detailed empirical analysis of the memory limits of common RNN architectures, highlighting the constraints imposed by network type and size.

Findings

Memory distance is highly limited in classical RNNs, LSTMs, and GRUs.

The limits are below the theoretical maximum for information retention.

Memory capacity depends on network architecture and size.

Abstract

Common to all different kinds of recurrent neural networks (RNNs) is the intention to model relations between data points through time. When there is no immediate relationship between subsequent data points (like when the data points are generated at random, e.g.), we show that RNNs are still able to remember a few data points back into the sequence by memorizing them by heart using standard backpropagation. However, we also show that for classical RNNs, LSTM and GRU networks the distance of data points between recurrent calls that can be reproduced this way is highly limited (compared to even a loose connection between data points) and subject to various constraints imposed by the type and size of the RNN in question. This implies the existence of a hard limit (way below the information-theoretic one) for the distance between related data points within which RNNs are still able to recognize said relation.