Construction Of Memristor-Based Neural Network Circuits For Associative Learning And Brain-like Emotion

In the era of rapid development of big data technology,artificial intelligence methods based on the simulation of brain structure and function,known as neuromorphic intelligence,have been widely studied and applied in recent years.Synapses and neurons,as the basic units of information transmission and learning in the brain,are the focus of research.As the study of neuromorphic intelligence deepens,there is an explosive growth in data processing and computing needs,but hardware neural network systems face the challenge of the failure of Moore’s Law.To address this issue,scholars have begun to explore new types of nanoscale devices that can integrate large-scale neural network circuits.Memristor is a device that can simulate the behavior of neurons and the plasticity of synaptic connections.With the characteristics of low power consumption,fast response and miniaturization,it is expected to become the core component of integrating large-scale brain-like neural network circuits.In this paper,a memristive neural network circuit for associative learning and brain-like emotion is constructed,and the functions of operant conditioning,multimodal generalization and differentiation and PAD emotional space model are realized.The main contents are as follows:Aiming at the active learning method in associative learning,the various processes and influencing factors of operant conditioning are analyzed.A memristor-based operant conditioning neural network circuit is constructed to refine the active learning process in associative learning.The constructed circuit consists of voltage control modules,facilitation suppression modules and operation modules.The processes of positive reinforcement,positive punishment,negative punishment,avoidance and weakening during operant conditioning can be simulated.The effects of immediacy and satiety on operant conditioning are considered.The operant conditioning neural network circuit based on memristor can effectively simulate the learning behavior of operant conditioning,and provide ideas for building a brain-like neural network that conforms to biological characteristics.Aiming at the generalization phenomenon produced by multiple neurons in Pavlov associative memory,the concept of secondary differentiation is proposed.The Pavlov associative memory neural network circuit with multimodal generalization and differentiation is constructed.The constructed circuit is composed of signal judgment modules,signal regulation modules and synaptic neuron modules.The functions of learning,forgetting,generalization,differentiation,secondary differentiation and multiple generalization differentiation are realized,and the phenomenon of misjudgment of similar neurons in the process of associative memory is solved.The multimodal generalization and differentiation Pavlov associative memory neural network circuit based on memristor can effectively distinguish the misjudgment behavior caused by similar stimuli,and provides a method for integrating large-scale neural network circuits.Aiming at the relationship between memory and emotion,a three-dimensional emotion space model of emotion generation circuit is proposed and applied to emotion consistency.A memory-based model of the affective space of PAD is constructed.The model is based on the theory of emotion of the limbic system of the brain,including modules such as the hypothalamus,sensory cortex,orbitofrontal cortex,cingulate gyrus and amygdala.The circuit can generate brain-like emotions consisting of three dimensions of pleasure,arousal and domination based on visual,speech and text information.The three-dimensional emotional space model of PAD based on memristor can help to better understand the process of brain-like emotion generation.The combination of learning and emotion provides methods and ideas for constructing a brain-like neural network that conforms to biological characteristics.In this paper,using memristor as the core component,an operant conditioning neural network circuit,a multimodal generalization and differentiation neural network circuit and a PAD emotional space model are constructed,which are applied to the neural network circuit of emotional consistency.The functions of learning,memory and emotion in brain-inspired intelligence have been realized,providing reference for the research and development of braininspired intelligence.