Abstract:The dreaming Hopfield model constitutes a generalization of the Hebbian paradigm for neural networks, that is able to perform on-line learning when "awake" and also to account for off-line "sleeping" mechanisms. The latter have been shown to enhance storing in such a way that, in the long sleep-time limit, this model can reach the maximal storage capacity achievable by networks equipped with symmetric pairwise interactions. In this paper, we inspect the minimal amount of information that must be supplied to such a network to guarantee a successful generalization, and we test it both on random synthetic and on standard structured datasets (i.e., MNIST, Fashion-MNIST and Olivetti). By comparing these minimal thresholds of information with those required by the standard (i.e., always "awake") Hopfield model, we prove that the present network can save up to ∼90% of the dataset size, yet preserving the same performance of the standard counterpart. This suggests that sleep may play a pivotal role in explaining the gap between the large volumes of data required to train artificial neural networks and the relatively small volumes needed by their biological counterparts. Further, we prove that the model Cost function (typically used in statistical mechanics) admits a representation in terms of a standard Loss function (typically used in machine learning) and this allows us to analyze its emergent computational skills both theoretically and computationally: a quantitative picture of its capabilities as a function of its control parameters is achieved and consistency between the two approaches is highlighted. The resulting network is an associative memory for pattern recognition tasks that learns from examples on-line, generalizes correctly (in suitable regions of its control parameters) and optimizes its storage capacity by off-line sleeping: such a reduction of the training cost can be inspiring toward sustainable AI and in situations where data are relatively sparse.

A Feed-Forward Neural Network for Increasing the Hopfield-Network Storage Capacity

Pattern Separation in a Spiking Neural Network of Hippocampus Robust to Imbalanced Excitation/Inhibition

A Brain-Inspired Spiking Neural Network Model with Temporal Encoding and Learning

Scalable Implementation of Hippocampal Network on Digital Neuromorphic System towards Brain-Inspired Intelligence

A hippocampus CA3 spiking neural network model for storage and retrieval of sequential memory

A Hierarchical Organized Memory Model Using Spiking Neurons

Effect of adult hippocampal neurogenesis on pattern separation and its applications

Distinguishing examples while building concepts in hippocampal and artificial networks

Simplicial Hopfield networks

Storage Capacity of the Hopfield Network Associative Memory

Maximizing memory capacity in heterogeneous networks

Dreaming neural networks: Forgetting spurious memories and reinforcing pure ones

Dynamic Capacity Estimation in Hopfield Networks

Pattern Storage and Hopfield Neural Associative Memory with Hidden Structure.

Uniform Memory Retrieval with Larger Capacity for Modern Hopfield Models

Long Sequence Hopfield Memory

Recursive Synaptic Bit Reuse: An Efficient Way to Increase Memory Capacity in Associative Memory

Dynamical analysis and DSP implementation of 3D Hopfield neural network under dual memristive electromagnetic radiation

Extending memory capacity of neural associative memory based on recursive synaptic bit reuse.

Hebbian dreaming for small datasets

Online Training of Hopfield Networks using Predictive Coding