Learning under weight constraints in networks of temporal encoding spiking neurons

Wu, Qingxiang; McGinnity, T.M.; Maguire, Liam; Glackin, Brendan; Belatreche, Ammar

doi:10.1016/j.neucom.2005.11.023

Cited by 46 publications

(18 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Table II compares the convergence accuracy of SWAT against existing algorithms for the WBC dataset [20]. For this dataset, the test data accuracy is comparable to that of the other approaches.…”

Section: B Wbc Datasetmentioning

confidence: 98%

“…To remove this instability, the weights can be capped, which implies that the maximum value of the weight vector is predetermined and therefore bears no relation to the temporal characteristics of the input data. One such learning algorithm [20] that used the capping of weights between a maximum and minimum value was developed. It was tested on several benchmark problems and produced excellent results.…”

Section: A Training Algorithmsmentioning

confidence: 99%

“…In this paper, τ m = 60 ms and R m = 1000 M [24]. The firing threshold of the neuron is implemented using [41] V th New = m × V th Orig exp − t − t i τ decay (20) where V th New is the new value of the firing threshold after the neuron has fired, m is a multiplication factor, V th Orig is the original firing threshold, t is the current time, t i is the time the neuron fired, and τ decay sets the rate at which the threshold decays back to its equilibrium value. In the present case τ decay = 20 ms, V th Orig = 9 mV [42] and m was set to 11.1, this sets the threshold to 100 mV after firing.…”

Section: Neuron Modelmentioning

confidence: 99%

“…Training and testing was achieved in these experiments by the process described in Section III-C. To encode the data while minimizing the number of neurons in each array in the hidden layer, the squared cosine method [20] was used. Using this method, the complete attribute range is multiplied The SNN topology in Fig.…”

Section: A Iris Datasetmentioning

confidence: 99%

“…Also, from the weight distribution it is clear that the mapping process associated with the training neuron causes competition between synapses. Table I compares the convergence accuracy of SWAT against existing algorithms for the Iris dataset [20]. MatlabBP and MatlabLM are built-in functions used by MATLAB that implement the backpropagation and Levenberg-Marquardt training algorithms, respectively.…”

Section: A Iris Datasetmentioning

confidence: 99%

See 4 more Smart Citations

SWAT: A Spiking Neural Network Training Algorithm for Classification Problems

Wade

McDaid

Santos

et al. 2010

IEEE Trans. Neural Netw.

165

128

View full text Add to dashboard Cite

Abstract-This paper presents a synaptic weight association training (SWAT) algorithm for spiking neural networks (SNNs). SWAT merges the Bienenstock-Cooper-Munro (BCM) learning rule with spike timing dependent plasticity (STDP). The STDP/BCM rule yields a unimodal weight distribution where the height of the plasticity window associated with STDP is modulated causing stability after a period of training. The SNN uses a single training neuron in the training phase where data associated with all classes is passed to this neuron. The rule then maps weights to the classifying output neurons to reflect similarities in the data across the classes. The SNN also includes both excitatory and inhibitory facilitating synapses which create a frequency routing capability allowing the information presented to the network to be routed to different hidden layer neurons. A variable neuron threshold level simulates the refractory period. SWAT is initially benchmarked against the nonlinearly separable Iris and Wisconsin Breast Cancer datasets. Results presented show that the proposed training algorithm exhibits a convergence accuracy of 95.5% and 96.2% for the Iris and Wisconsin training sets, respectively, and 95.3% and 96.7% for the testing sets, noise experiments show that SWAT has a good generalization capability. SWAT is also benchmarked using an isolated digit automatic speech recognition (ASR) system where a subset of the TI46 speech corpus is used. Results show that with SWAT as the classifier, the ASR system provides an accuracy of 98.875% for training and 95.25% for testing.Index Terms-Automatic speech recognition, BienenstockCooper-Munro, dynamic synapses, spike timing dependent plasticity, spiking neural networks.

show abstract

Section: B Wbc Datasetmentioning

confidence: 98%

Section: A Training Algorithmsmentioning

confidence: 99%

Section: Neuron Modelmentioning

confidence: 99%

Section: A Iris Datasetmentioning

confidence: 99%

Section: A Iris Datasetmentioning

confidence: 99%

See 3 more Smart Citations

SWAT: A Spiking Neural Network Training Algorithm for Classification Problems

Wade

McDaid

Santos

et al. 2010

IEEE Trans. Neural Netw.

165

128

View full text Add to dashboard Cite

show abstract

A Proof of a Key Formula in the Error-Backpropagation Learning Algorithm for Multiple Spiking Neural Networks

Yang

Yang²,

Fan

2014

Advances in Neural Networks – ISNN 2014

View full text Add to dashboard Cite

Learning Mechanisms in Networks of Spiking Neurons

McGinnity

Maguire

et al.

Studies in Computational Intelligence

View full text Add to dashboard Cite

In spiking neural networks, signals are transferred by action potentials. The information is encoded in the patterns of neuron activities or spikes. These features create significant differences between spiking neural networks and classical neural networks. Since spiking neural networks are based on spiking neuron models that are very close to the biological neuron model, many of the principles found in biological neuroscience can be used in the networks. In this chapter, a number of learning mechanisms for spiking neural networks are introduced. The learning mechanisms can be applied to explain the behaviours of networks in the brain, and also can be applied to artificial intelligent systems to process complex information represented by biological stimuli.

show abstract

Learning under weight constraints in networks of temporal encoding spiking neurons

Cited by 46 publications

References 14 publications

SWAT: A Spiking Neural Network Training Algorithm for Classification Problems

SWAT: A Spiking Neural Network Training Algorithm for Classification Problems

A Proof of a Key Formula in the Error-Backpropagation Learning Algorithm for Multiple Spiking Neural Networks

Learning Mechanisms in Networks of Spiking Neurons

Contact Info

Product

Resources

About