Memristor-Based Operant Conditioning Neural Network With Blocking and Competition Effects

Operant conditioning is an important learning mechanism for organisms, as well as a basic theory for reinforcement learning in artificial intelligence. Although there are already some memristive neural circuits for operant conditioning, they can only process a single stimulus and cannot handle multiple inputs simultaneously. This article proposes a multi-input operant conditioning neural network that incorporates blocking and competing effects. This network can achieve the blocking and overshadowing effects in the presence of multiple inputs and learn efficiently in complex environments. In addition, it incorporates time differences between signals and excitations, random exploration, feedback learning, experience memory, decision-making based on experience, and adaptive learning in low-reward environments. Finally, the feasibility of the proposed circuit function is verified through PSPICE simulation. This work provides an implementation idea for the hardware implementation of artificial intelligence.