Having a formal model of neural networks can greatly help in understanding and verifying their properties,behavior,and response to external factors such as disease and medicine.In this paper,we adopt a formal model to...Having a formal model of neural networks can greatly help in understanding and verifying their properties,behavior,and response to external factors such as disease and medicine.In this paper,we adopt a formal model to represent neurons,some neuronal graphs,and their composition.Some specific neuronal graphs are known for having biologically relevant structures and behaviors and we call them archetypes.These archetypes are supposed to be the basis of typical instances of neuronal information processing.In this paper we study six fundamental archetypes(simple series,series with multiple outputs,parallel composition,negative loop,inhibition of a behavior,and contralateral inhibition),and we consider two ways to couple two archetypes:(i)connecting the output(s)of the first archetype to the input(s)of the second archetype and(ii)nesting the first archetype within the second one.We report and compare two key approaches to the formal modeling and verification of the proposed neuronal archetypes and some selected couplings.The first approach exploits the synchronous programming language Lustre to encode archetypes and their couplings,and to express properties concerning their dynamic behavior.These properties are verified thanks to the use of model checkers.The second approach relies on a theorem prover,the Coq Proof Assistant,to prove dynamic properties of neurons and archetypes.展开更多
为了提高表情识别率并降低表情识别的功耗,提出一种基于改进深度残差网络的表情识别方法。残差学习在解决深度卷积神经网络退化问题、使网络层次大幅加深的同时,进一步增加了网络的功耗。为此,引入具有生物真实性的激活函数来代替已有...为了提高表情识别率并降低表情识别的功耗,提出一种基于改进深度残差网络的表情识别方法。残差学习在解决深度卷积神经网络退化问题、使网络层次大幅加深的同时,进一步增加了网络的功耗。为此,引入具有生物真实性的激活函数来代替已有的整流线性单元(Rectified Linear Units,ReLU)函数,并将其作为卷积层激活函数对深度残差网络进行改进。该方法不仅提高了残差网络的精度,而且训练出的网络权重可直接作为与该深度残差网络具有相同结构的深度脉冲神经网络的权重。将该深度脉冲神经网络部署在类脑硬件上时,其能够以较高的识别率和较低的能耗进行表情识别。展开更多
We study the stochastic resonance （SR） in Hodgkin-Huxley （HH） neural systems with small-world （SW） connections under the noise synaptic current and periodic stimulus, focusing on the dependence of properties of ...We study the stochastic resonance （SR） in Hodgkin-Huxley （HH） neural systems with small-world （SW） connections under the noise synaptic current and periodic stimulus, focusing on the dependence of properties of SR on coupling strength c. It is found that there exists a critical coupling strength c^＊ such that if c 〈 c^＊, then the SR can appear on the SW neural network. Especially, dependence of the critical coupling strength c^＊ on the number of neurons N shows the monotonic even almost linear increase of c^＊ as N increases and c^＊ on the SW network is smaller than that on the random network. For the effect of the SW network on the phenomenon of SR, we show that decreasing the connection-rewiring probability p of the network topology leads to an enhancement of SR. This indicates that the SR on the SW network is more prominent than that on the random network （p = 1.0）. In addition, it is noted that the effect becomes remarkable as coupling strength increases. Moreover, it is found that the SR weakens but resonance range becomes wider with the increase of c on the SW neural network.展开更多
基金This work was supported by the French government through the UCA-Jedi project managed by the National Research Agency(ANR-15-IDEX-01)in particular,by the interdisciplinary Institute for Modeling in Neuroscience and Cognition(NeuroMod)of the UniversitéCôte d'Azur.It was also supported by the Natural Sciences and Engineering Research Council of Canada.
文摘Having a formal model of neural networks can greatly help in understanding and verifying their properties,behavior,and response to external factors such as disease and medicine.In this paper,we adopt a formal model to represent neurons,some neuronal graphs,and their composition.Some specific neuronal graphs are known for having biologically relevant structures and behaviors and we call them archetypes.These archetypes are supposed to be the basis of typical instances of neuronal information processing.In this paper we study six fundamental archetypes(simple series,series with multiple outputs,parallel composition,negative loop,inhibition of a behavior,and contralateral inhibition),and we consider two ways to couple two archetypes:(i)connecting the output(s)of the first archetype to the input(s)of the second archetype and(ii)nesting the first archetype within the second one.We report and compare two key approaches to the formal modeling and verification of the proposed neuronal archetypes and some selected couplings.The first approach exploits the synchronous programming language Lustre to encode archetypes and their couplings,and to express properties concerning their dynamic behavior.These properties are verified thanks to the use of model checkers.The second approach relies on a theorem prover,the Coq Proof Assistant,to prove dynamic properties of neurons and archetypes.
文摘为了提高表情识别率并降低表情识别的功耗,提出一种基于改进深度残差网络的表情识别方法。残差学习在解决深度卷积神经网络退化问题、使网络层次大幅加深的同时,进一步增加了网络的功耗。为此,引入具有生物真实性的激活函数来代替已有的整流线性单元(Rectified Linear Units,ReLU)函数,并将其作为卷积层激活函数对深度残差网络进行改进。该方法不仅提高了残差网络的精度,而且训练出的网络权重可直接作为与该深度残差网络具有相同结构的深度脉冲神经网络的权重。将该深度脉冲神经网络部署在类脑硬件上时,其能够以较高的识别率和较低的能耗进行表情识别。
基金Supported by the National Natural Science Foundation of China under Grant Nos 70571017 and 10247005, the Innovation Project of Guangxi Graduate Education under Grant No 2006106020809M36, and Key Project of the National Natural Science Foundation of China under Grant No 70431002.
文摘We study the stochastic resonance （SR） in Hodgkin-Huxley （HH） neural systems with small-world （SW） connections under the noise synaptic current and periodic stimulus, focusing on the dependence of properties of SR on coupling strength c. It is found that there exists a critical coupling strength c^＊ such that if c 〈 c^＊, then the SR can appear on the SW neural network. Especially, dependence of the critical coupling strength c^＊ on the number of neurons N shows the monotonic even almost linear increase of c^＊ as N increases and c^＊ on the SW network is smaller than that on the random network. For the effect of the SW network on the phenomenon of SR, we show that decreasing the connection-rewiring probability p of the network topology leads to an enhancement of SR. This indicates that the SR on the SW network is more prominent than that on the random network （p = 1.0）. In addition, it is noted that the effect becomes remarkable as coupling strength increases. Moreover, it is found that the SR weakens but resonance range becomes wider with the increase of c on the SW neural network.
