Correlation of neuronal activities is widely observed in the central nervous system and is likely to play a key role in its functioning. It is, thus, essential to understand the effects of correlated synaptic input on...
详细信息
Correlation of neuronal activities is widely observed in the central nervous system and is likely to play a key role in its functioning. It is, thus, essential to understand the effects of correlated synaptic input on the response of neurons. Here, we model neuronal input as correlated Poisson processes. and assess their impact on the leaky integrate-and-fire neuron. We found that neuronal output firing rate typically is a non-monotonic function of the input correlation, and propose that the response of neurons is critically dependent on the input ensemble statistics. (C) 2002 Elsevier Science B.V. All rights reserved.
Spiking neural networks (SNNs) are considered to be biologically plausible and can yield high energy efficiency when implemented on neuromorphic hardware due to their highly sparse asynchronous binary event-driven nat...
详细信息
Spiking neural networks (SNNs) are considered to be biologically plausible and can yield high energy efficiency when implemented on neuromorphic hardware due to their highly sparse asynchronous binary event-driven nature. Recently, surrogate gradient (SG) approaches have enabled SNNs to be trained from scratch with backpropagation (BP) algorithms under a deep learning framework. However, a popular SG approach known as straight-through estimator (STE), which only propagates the same gradient information, does not take into account the activation differences between the membrane potentials and output spikes. To address this issue, we propose surrogate gradient scaling (SGS), which scales up or down the gradient information of the membrane potential according to the sign of the gradient of the spiking neuron output and the difference between the membrane potential and the output of the spiking neuron. This SGS approach can also be applied to unimodal functions that propagate different gradient information from the output spikes to the input membrane potential. In addition, SNNs trained directly from scratch suffer from poor generalization performance, and we introduce Lipschitz regularization (LR), which is incorporated into the loss function. It not only improves the generalization performance of SNNs but also makes them more robust to noise. Extensive experimental results on several popular benchmark datasets (CIFAR10, CIFAR100 and CIFAR10-DVS) show that our approach not only outperforms the SOTA but also has lower inference latency. Remarkably, our SNNs can lead to 34x, 29x, and 17x computation energy savings compared to standard Artificial neural networks (ANNs) on above three datasets.
We investigate the representation of visual stimuli and the short-term dynamics of activity within primary visual cortex in a 'free-viewing' scenario with 'saccading eye movements' modeled as a series ...
详细信息
We investigate the representation of visual stimuli and the short-term dynamics of activity within primary visual cortex in a 'free-viewing' scenario with 'saccading eye movements' modeled as a series of visual stimuli that are flashed onto the retina for the duration of a fixation period (200-300 ms). We assume that the entire activity pattern from the beginning of fixation until time t constitutes the neural code. Given a noisy (Poissonian) representation it follows that the signal-to-noise ratio increases with time, because more spikes become available for representation. Here, we show that for archiving an optimal stimulus representation in any increasing time-window beginning with stimulus onset, the processing strategy of the network should be dynamic in the sense that an initially high recurrent cortical competition between orientation selective cells attenuates with time, i.e. mediated by the instrinsic property of spike-frequency adaptation of pyramidal cells. (C) 2001 Elsevier Science B.V. All rights reserved.
We analysed spike trains from the descending contralateral movement detector (DCMD) neuron of locusts. The locusts either performed jumps or did not jump in response to visual looming stimuli. An evolutionary algorith...
详细信息
We analysed spike trains from the descending contralateral movement detector (DCMD) neuron of locusts. The locusts either performed jumps or did not jump in response to visual looming stimuli. An evolutionary algorithm (EA) was employed to sort spike trains into the correct behavioural categories by optimising threshold parameters, so jump behaviour occurred if the spike-train data exceeded the threshold parameters from the EA. A candidate behavioural trigger appeared to be prolonged high-frequency spikes at a relatively early stage in the approach of the stimulus. This technique provides a useful precursor to a full biological analysis of the escape jump mechanism. (c) 2006 Elsevier B.V. All rights reserved.
It has been suggested that tactile intensity perception can be explained by a linear function of spike rate weighted by afferent type. Other than relying on mathematical models, verifying this experimentally is diffic...
详细信息
It has been suggested that tactile intensity perception can be explained by a linear function of spike rate weighted by afferent type. Other than relying on mathematical models, verifying this experimentally is difficult due to the frequency tuning of different afferent types and changes in population recruitment patterns with vibrotactile frequency. To overcome these complexities, we used pulsatile mechanical stimuli which activate the same afferent population regardless of the repetition rate (frequency), generating one action potential per pulse. We used trains of different frequencies (20-200 Hz) to investigate perceived intensity. Subjects' magnitude ratings increased with pulse rate up to similar to 100 Hz and plateaued beyond this frequency. This was true regardless of pulse amplitude, from small pulses that exclusively activated Pacinian (PC) afferents, to pulses large enough to activate other afferents including slowly adapting. Electrical stimulation, which activates afferents indiscriminately, plateaued at a similar frequency, although not in all subjects. As the plateauing did not depend on indentation magnitude and hence on afferent weights, we propose that the contribution of spike count to intensity perception is weighted by a function of frequency. This may explain why fine textures evoking high frequency vibrations of a small magnitude do not feel disproportionally intense.
For the analysis of natural neural responses it is necessary to evaluate and compare the reliability of the produced spike sequences. The same occurs in the development and evaluation of neural models, which should mi...
详细信息
For the analysis of natural neural responses it is necessary to evaluate and compare the reliability of the produced spike sequences. The same occurs in the development and evaluation of neural models, which should mimic the real neural centers that are being modeled. Several neural metrics have been proposed to analyze neural responses, and to tune and evaluate neural models. neural metrics measure different characteristics of the neural code and can be grouped into distinct classes, as they follow a firing rate or time-code perspective. In this paper, several metrics belonging to the firing rate, spike train and firing event classes are reviewed. Using sets of neuronal responses and a set of models, the metrics are analyzed and compared to disclose their advantages and drawbacks. In most cases these metrics depend on a free parameter, that establishes their sensitivity to particular characteristics of the neural code. After showing that the incorrect choice of these parameters can lead to meaningless results, methods are presented in this paper to define a valid range of values for the parameters. These methods are based on a statistical analysis of the inter-trials errors. The application of neural metrics to the tuning and assessment of neural models of distinct classes reveals important results. Some of the analyzed metrics possess pronounced minima, specifically around the origin, which makes the optimization process more difficult;nonetheless, they provide insightful results for the evaluation of models. This paper also discusses the application of the neural metrics to evaluate neural models, providing relevant guidelines for their utilization. (c) 2009 Elsevier B.V. All rights reserved.
An important problem in neuroscience is to obtain quantitative knowledge of how information is represented, or encoded, in the signals that nerve cells process and transmit. Sensory receptors have provided important m...
详细信息
An important problem in neuroscience is to obtain quantitative knowledge of how information is represented, or encoded, in the signals that nerve cells process and transmit. Sensory receptors have provided important models for the study of neural coding because their inputs can often be relatively easily controlled and measured, while the resultant activity is recorded. A variety of engineering concepts have been successfully applied to physiological sciences, particularly those related to control of dynamic systems. Linear systems analysis was one of the earliest methods used to probe sensory coding, and measurements such as step responses and frequency responses have become standard tools for describing sensory functions. Modern systems analysis has evolved to provide accurate and efficient linear identification of encoding in sensory receptors that use either graded potentials or action potentials. It has also led to nonlinear systems analysis, the creation of parametric nonlinear models, and measures of information coding by sensory neurons. These methods promise to provide important new knowledge about sensory systems in the future, especially when complemented with parallel biophysical and molecular studies of sensory neurons. Mechanoreceptors provided some of the earliest preparations for the investigation of neural coding, and both the linear and nonlinear properties of wide variety of vertebrate and invertebrate mechanoreceptors continue to be explored.
作者：
BARAM, YNASA
AMES RES CTR MOFFETT FIELD CA 94035 USA
Networks of ternary neurons, storing random vectors over the set {-1,0,1} by the so-called Hebbian rule, are considered. It is shown that the maximal number of stored patterns that are equilibrium states of the networ...
详细信息
Networks of ternary neurons, storing random vectors over the set {-1,0,1} by the so-called Hebbian rule, are considered. It is shown that the maximal number of stored patterns that are equilibrium states of the network with probability tending to one as N tends to infinity is at least of the order of N2-1/alpha/K, where N is the number of neurons, K is the number of nonzero elements in a pattern and t = alpha-K, 1/2 < alpha < 1, is the threshold in the neuron function. While for small K this bound is similar to that obtained for fully connected binary networks, the number of interneural connections required in the ternary case can be considerably smaller. Similar bounds incorporating error probabilities are shown to guarantee, in the same probabilistic sense, the correction of errors in the nonzero elements and in the location of these elements.
We consider the dependence of information transfer by neurons on the Type I vs. Type II classification of their dynamics. Our computational study is based on Type I and II implementations of the Morris-Lecar model. It...
详细信息
We consider the dependence of information transfer by neurons on the Type I vs. Type II classification of their dynamics. Our computational study is based on Type I and II implementations of the Morris-Lecar model. It mainly concerns neurons, such as those in the auditory or electrosensory system, which encode band-limited amplitude modulations of a periodic carrier signal, and which fire at random cycles yet preferred phases of this carrier. We first show that the Morris-Lecar model with additive broadband noise ("synaptic noise") can exhibit such firing patterns with either Type I or II dynamics, with or without amplitude modulations of the carrier. We then compare the encoding of band-limited random amplitude modulations for both dynamical types. The comparison relies on a parameter calibration that closely matches firing rates for both models across a range of parameters. In the absence of synaptic noise, Type I performs slightly better than Type II, and its performance is optimal for perithreshold signals. However, Type II performs well over a slightly larger range of inputs, and this range lies mostly in the subthreshold region. Further, Type II performs marginally better than Type I when synaptic noise, which yields more realistic baseline firing patterns, is present in both models. These results are discussed in terms of the tuning and phase locking properties of the models with deterministic and stochastic inputs.
We apply the recently developed information distortion method (Comput. neural Systems 12 (4) (2001) 441) to the analysis of coding by single neurons in the cricket cercal sensory system. This technique allows simultan...
详细信息
We apply the recently developed information distortion method (Comput. neural Systems 12 (4) (2001) 441) to the analysis of coding by single neurons in the cricket cercal sensory system. This technique allows simultaneous identification of stimulus features and corresponding neural responses. The best approximation of a coding scheme that we obtained suggests that significant information is encoded in spike patterns, We compare this method to the linear stimulus reconstruction approach. Our coarsest nontrivial reproduction completely recovers the stimulus reconstruction results. Further refinements uncover additional structure, not present in the stimulus reconstruction results. (C) 2002 Elsevier Science B.V. All rights reserved.
暂无评论