In recent years, spiking neural networks(SNNs) have received increasing attention of research in the field of artificial intelligence due to their high biological plausibility, low energy consumption, and abundant spa...In recent years, spiking neural networks(SNNs) have received increasing attention of research in the field of artificial intelligence due to their high biological plausibility, low energy consumption, and abundant spatio-temporal information.However, the non-differential spike activity makes SNNs more difficult to train in supervised training. Most existing methods focusing on introducing an approximated derivative to replace it, while they are often based on static surrogate functions. In this paper, we propose a progressive surrogate gradient learning for backpropagation of SNNs, which is able to approximate the step function gradually and to reduce information loss. Furthermore, memristor cross arrays are used for speeding up calculation and reducing system energy consumption for their hardware advantage. The proposed algorithm is evaluated on both static and neuromorphic datasets using fully connected and convolutional network architecture, and the experimental results indicate that our approach has a high performance compared with previous research.展开更多
Convolutional neur al netw orks(CNNs)ar e widel y used in the field of fault diagnosis due to their strong feature-extraction capability.How ever,in eac h timeste p,CNNs onl y consider the curr ent input and ignor e a...Convolutional neur al netw orks(CNNs)ar e widel y used in the field of fault diagnosis due to their strong feature-extraction capability.How ever,in eac h timeste p,CNNs onl y consider the curr ent input and ignor e any cyclicity in time,ther efor e pr oducing difficulties in mining temporal features from the data.In this w ork,the third-gener ation neur al netw ork-the spiking neur al netw ork(SNN)-is utilized in bearing fault diagnosis.SNNs incorpor ate tempor al concepts and utilize discrete spike sequences in communication,making them more biolo gically e xplanatory.Inspired by the classic CNN LeNet-5 fr amew ork,a bearing fault diagnosis method based on a convolutional SNN is proposed.In this method,the spiking convolutional network and the spiking classifier network are constructed by using the inte gr ate-and-fire(IF)and leaky-inte gr ate-and-fire(LIF)model,respectively,and end-to-end training is conducted on the overall model using a surrogate gradient method.The signals are adaptively encoded into spikes in the spiking neuron layer.In addition,the network utilizes max-pooling,which is consistent with the spatial-temporal characteristics of SNNs.Combined with the spiking con volutional la y ers,the netw ork fully extracts the spatial-temporal featur es fr om the bearing vibration signals.Experimental validations and comparisons are conducted on bearings.The results show that the proposed method achieves high accuracy and takes fewer time steps.展开更多
基金Project supported by the Natural Science Foundation of Chongqing(Grant No.cstc2021jcyj-msxmX0565)the Fundamental Research Funds for the Central Universities(Grant No.SWU021002)the Graduate Research Innovation Project of Chongqing(Grant No.CYS22242)。
文摘In recent years, spiking neural networks(SNNs) have received increasing attention of research in the field of artificial intelligence due to their high biological plausibility, low energy consumption, and abundant spatio-temporal information.However, the non-differential spike activity makes SNNs more difficult to train in supervised training. Most existing methods focusing on introducing an approximated derivative to replace it, while they are often based on static surrogate functions. In this paper, we propose a progressive surrogate gradient learning for backpropagation of SNNs, which is able to approximate the step function gradually and to reduce information loss. Furthermore, memristor cross arrays are used for speeding up calculation and reducing system energy consumption for their hardware advantage. The proposed algorithm is evaluated on both static and neuromorphic datasets using fully connected and convolutional network architecture, and the experimental results indicate that our approach has a high performance compared with previous research.
基金supported by the Natural Science Foundation of China (Grants No.U1934219,62173137 and 52272347)the Hunan Pr ovincial Natur al Science Foundation of China (Grant No.2021JJ50001).
文摘Convolutional neur al netw orks(CNNs)ar e widel y used in the field of fault diagnosis due to their strong feature-extraction capability.How ever,in eac h timeste p,CNNs onl y consider the curr ent input and ignor e any cyclicity in time,ther efor e pr oducing difficulties in mining temporal features from the data.In this w ork,the third-gener ation neur al netw ork-the spiking neur al netw ork(SNN)-is utilized in bearing fault diagnosis.SNNs incorpor ate tempor al concepts and utilize discrete spike sequences in communication,making them more biolo gically e xplanatory.Inspired by the classic CNN LeNet-5 fr amew ork,a bearing fault diagnosis method based on a convolutional SNN is proposed.In this method,the spiking convolutional network and the spiking classifier network are constructed by using the inte gr ate-and-fire(IF)and leaky-inte gr ate-and-fire(LIF)model,respectively,and end-to-end training is conducted on the overall model using a surrogate gradient method.The signals are adaptively encoded into spikes in the spiking neuron layer.In addition,the network utilizes max-pooling,which is consistent with the spatial-temporal characteristics of SNNs.Combined with the spiking con volutional la y ers,the netw ork fully extracts the spatial-temporal featur es fr om the bearing vibration signals.Experimental validations and comparisons are conducted on bearings.The results show that the proposed method achieves high accuracy and takes fewer time steps.
