Neuromodulation-inspired gated associative memory networks: extended memory retrieval and emergent multistability

Area of Science:

• Computational neuroscience
• Neuro-inspired computing

Background:

• Classical autoassociative memory models are crucial for understanding neural circuits but often ignore neuromodulation.
• Neuromodulatory agents significantly influence memory capacity and stability in biological systems.

Purpose of the Study:

• To introduce a biophysically motivated associative memory network incorporating neuromodulation-like gating.
• To investigate how activity-dependent gating impacts attractor structure and memory capacity.

Main Methods:

• Development of a minimal network model with self-adaptive, activity-dependent gating.
• Utilizing many-body simulations and dynamical mean-field theory for analysis.

Main Results:

• The gating mechanism reorganizes attractor structure, bypassing the spin-glass transition.
• Robust, high-overlap retrieval is maintained beyond the standard critical capacity.
• Transient pattern remnants are stabilized into multistable attractors, enhancing memory capacity.

Conclusions:

• Neuromodulation-like gating dramatically enhances associative memory capacity in neural networks.
• This mechanism eliminates catastrophic breakdown and reshapes the memory landscape.
• Provides a route to richer memory dynamics for neuromodulated circuits and neuromorphic architectures.