Brain-Inspired Model and Neuromorphic Circuit Implementation for Feature-Affective Associative Memory Network

Affective associative memory is one method by which agents acquire knowledge, experience, and skills from natural surroundings or social activities. Using neuromorphic circuits to implement affective associative memory aids in developing brain-inspired intelligence. In this paper, a feature-affective associative memory (FAAM) network model and its memristive circuit are proposed for real-time and mutual associative memory and retrieval between multiple features and emotions. With the context of fear conditioning, FAAM network circuit is verified to enable the acquisition and extinction of associations. Different from other works, the proposed temporal-rate mixed coding circuit encodes stimulus intensity and arousal level as different pulses, allowing the associative learning rate and emotion degree can vary with stimulus intensity and arousal level. Furthermore, the bidirectional and multi-feature-to-multi-emotion association model allows the circuit to be extended to associative memory network containing 10 neurons and 90 synapses, with capabilities such as emotion generation and modulation, associative generalization and differentiation, which are applied to feature binding, situational memory and inference decision. This work enables advanced cognitive functions, and is expected to enable intelligent robot platforms for real-time learning, reasoning decisions, and emotional companionship in dynamic environments.