Spiking neural network device that updates synaptic weight based on output frequency and learning method of spiking neural network device
A spiking neural network device according to an embodiment includes a synaptic element, a neuron circuit, a determinator, a synaptic depressor, and a synaptic potentiator. The synaptic element has a variable weight and outputs, in response to input of a first spike signal, a synaptic signal having intensity adjusted in accordance with the weight. The neuron circuit outputs a second spike signal in a case where the synaptic signal is inputted and a predetermined firing condition for the synaptic signal is satisfied. The determinator determines whether or not the weight is to be updated on a basis of an output frequency of the second spike signal by the neuron circuit. The synaptic depressor performs depression operation for depressing the weight in a case where it is determined that the weight is to be updated. The synaptic potentiator performs potentiating operation for potentiating the weight.
1 . A spiking neural network device comprising:
a synaptic circuit configured to have a variable weight, and configured to output, in response to input of a first spike signal, a synaptic signal whose intensity is adjusted in accordance with the weight;
a neuron circuit configured to input the synaptic signal and output a second spike signal in a case where an integrated value of the inputted synaptic signal exceeds a first threshold;
a determination circuit configured to determine whether or not the weight is to be updated on a basis of an output frequency that is a frequency of outputting the second spike signal by the neuron circuit, the output frequency being defined with a variable that increases each time the neuron circuit outputs the second spike signal;
a synaptic depression circuit configured to perform depression operation for depressing the weight in a case where it is determined that the weight is to be updated; and
a synaptic potentiation circuit configured to perform potentiating operation for potentiating the weight,
wherein the determination circuit determines that the weight is to be updated in a case where the output frequency is equal to or greater than a second threshold.
2 . The spiking neural network device according to claim 1 ,
wherein the output frequency decreases with time.
3 . The spiking neural network device according to claim 1 ,
wherein the synaptic depression circuit performs the depression operation in a case where it is determined that the weight is to be updated and the neuron circuit outputs the second spike signal.
4 . The spiking neural network device according to claim 1 ,
wherein the synaptic potentiation circuit performs the potentiating operation in a case where a predetermined potentiating condition is satisfied.
5 . The spiking neural network device according to claim 4 ,
wherein the synaptic potentiation circuit performs the potentiating operation in a case where the determination circuit determines that the weight is to be updated and the potentiating condition is satisfied.
6 . The spiking neural network device according to claim 1 ,
wherein the synaptic circuit includes a storage element for storing a value of the weight by a resistance value and a metal oxide semiconductor (MOS) transistor.
7 . The spiking neural network device according to claim 6 ,
wherein the storage circuit has a plurality of resistance states and transition between the plurality of resistance states is stochastic.
8 . The spiking neural network device according to claim 6 ,
wherein the synaptic circuit further includes a diode for connecting the MOS transistor and the storage element to each other.
9 . A learning method of a spiking neural network device,
the spiking neural network device including
a synaptic circuit configured to have a variable weight, and configured to output, in response to input of a first spike signal, a synaptic signal whose intensity is adjusted in accordance with the weight, and
a neuron circuit configured to input the synaptic signal and output a second spike signal in a case where an integrated value of the inputted synaptic signal exceeds a first threshold,
the learning method comprising:
a determination step of determining whether or not the weight is to be updated on a basis of an output frequency that is a frequency of outputting the second spike signal by the neuron circuit, the output frequency being defined with a variable that increases each time the neuron circuit outputs the second spike signal;
a synapse depressing step of performing depression operation for depressing the weight in a case where it is determined that the weight is to be updated; and
a synapse potentiating step of performing potentiating operation for potentiating the weight,
wherein the determination step includes a step of determining that the weight is to be updated in a case where the output frequency is equal to or greater than a second threshold.