Inference method and device using spiking neural network

An inference method using a spiking neural network, performed by a computing device including a processor, comprising:

• inputting an input signal to the spiking neural network, the spiking neural network including a plurality of layers;
• outputting plural output spikes by an artificial neuron, of the spiking neural network, with respect to plural inputs to the artificial neuron, where the plural inputs are dependent on the input signal;
• and
• determining a result of the artificial neuron corresponding to the plural inputs based on respective first output spikes, of the plural output spikes, that are output after a predetermined time from an operating time point of the artificial neuron with respect to the plural inputs,
• wherein the spiking neural network is generated based on a pre-learned analog-valued neural network (ANN), and
• wherein the determining of the result further comprises ignoring respective second output spikes, of the output spikes, that are output before the predetermined time, where each of the respective first output spikes and the respective second output spikes may include one or more output spikes.

An inference device using a spiking neural network, comprising:

• a processor which outputs a determined result corresponding to input data using the spiking neural network including a plurality of artificial neurons, for inference;
• and
• a memory which stores instructions executable by the processor, wherein execution of the instructions by the processor configures the processor to:
• perform the determination of the result corresponding to the input data, based on respective first output spikes of plural output spikes from the artificial neuron, that are output after a predetermined time from an operating time point of the artificial neuron with respect to the plural inputs, the determination further comprising:
• determining, with a sum of weightings of plural inputs to an artificial neuron of the plurality of artificial neurons being a negative value representing that a membrane potential of the artificial neuron is negative, that a neuron result of the artificial neuron is a value, which is negative;
• and
• ignoring, in the determination of the result, respective second output spikes, of the output spikes, that are output before the predetermined time, where each of the respective first output spikes and the respective second output spikes may include one or more output spikes;
• and
• perform the output of the determined result.