A Multi-functional Memristive Pavlov Associative Memory Circuit Based on Neural Mechanisms.
Yutong Zhang,Zhigang Zeng
DOI: https://doi.org/10.1109/tbcas.2021.3108354
2021-01-01
IEEE Transactions on Biomedical Circuits and Systems
Abstract:Pavlov conditioning is a typical associative memory, which involves associative learning between the gustatory and auditory cortex, known as Pavlov associative memory. Inspired by neural mechanisms and biological phenomena of Pavlov associative memory, this paper proposes a multi-functional memristive Pavlov associative memory circuit. In addition to learning and forgetting, whose rates change with the number of associative learning times, the circuit also achieves other innovative functions. First, consolidation learning, which refers to the continued learning process after acquiring associative memory, changes the rates of learning and forgetting. Secondly, the natural forgetting rate tends to zero when the associative memory has been acquired several times, which means the formation of long-term memory. Thirdly, the generalization and differentiation of associative memory caused by similar stimuli are realized through a simplified memristive feedforward neural network. Besides, this circuit implements the associative learning function of interval stimuli through a simpler structure, which refers to “the longer the stimuli interval, the slower the learning rate”. The above functions are realized by the time interval module, variable rates module, and generalization and differentiation module. It has been shown that the proposed circuit has good robustness, and can reduce the influence of parasitic capacitance, memristive conductance drift, and input noise on circuit functions. Through further research, this circuit is expected to be used in robot platforms to realize human-like perception and associative cognitive functions.