Intrinsic Plasticity Based Inference Acceleration for Spiking Multi-Layer Perceptron

IEEE Access

Anguo Zhang¹  Yupeng Ma²  Shuxun Zhang³  Wei Zhu³ 

[1] Chemistry, Chinese Academy of Sciences, Urumqi, China;College of Physics and Information Engineering, Fuzhou University, Fuzhou, China;The Xinjiang Technical Institute of Physics &x0026;

关键词: Intrinsic plasticity;    multi-layer perceptrons;    spiking neuron model;   

DOI  :  10.1109/ACCESS.2019.2914424

来源: DOAJ

【 摘 要 】

Intrinsic plasticity (IP) mechanism was originally found in the biological neuron as a membrane potential adaptive tuning scheme, which was used to change the connection strength between neurons, so that animal brain had the ability to learn or store memory. Recently, in the field of artificial neural networks, the bio-inspired IP mechanism attracts increasingly research attention due to its ability of regulating neuron activity in a relative homeostatic level even if the external input of a neuron is extremely low or extremely high and tuning the probability density of a neuron's output toward an exponential distribution, thereby realizing information maximization. In this paper, the IP mechanism was applied to the spiking neuron model-based multi-layer perceptrons (Spiking MLPs). The experiment results showed that compared with the networks without IP, both the convergence speed and the robustness of computation accuracy were effectively improved.

【 授权许可】

Unknown