Processing method and accelerating device

A data processing method, comprising:

• storing, by a storage unit, multiple input neurons, at least an output neuron, multiple weights, and an instruction of a neural network;
• performing, by a coarse-grained pruning unit, coarse-grained pruning on the weights of the neural network to obtain pruned weights;
• storing, by the coarse-grained pruning unit, the pruned weights and position information of one or more target weights into the storage unit, wherein an absolute value of each of the target weights is greater than a second given threshold;
• selecting, by the coarse-grained pruning unit, one or more weights from weights of the neural network through a sliding window;
• setting, by the coarse-grained pruning unit, at least a portion of the selected weights to 0;
• determining, by the coarse-grained pruning unit, position information of the target weights among the pruned weights based on the position of the selected weights that are set to zero;
• receiving, by a coarse-grained selection unit, the multiple input neurons and the position information of the target weights, and the target weights;
• selecting, by the coarse-grained selection unit, one or more input neurons corresponding to the target weights according to the position information of the target weights;
• performing, by an operation unit, training according to the pruned weights, and the weights that have been set to 0 in the training process remain 0;
• performing, by the operation unit, a neural network operation according to aft the target weights and the selected input neurons corresponding to the target weights to generate the at least an one output neuron;
• and
• transmitting, by the operation unit, the output neuron to the storage unit as an input neuron of a next layer.

An electronic device, comprising:

• a processing device that includes:
• a storage unit configured to store multiple input neurons, at least one output neuron, multiple weights, and an instruction of a neural network;
• a coarse-grained pruning unit configured to perform coarse-grained pruning on the weights of the neural network to obtain weights and store the pruned weights and position information of none or more target weights into the storage unit, wherein an absolute value of each of the target weights is greater than a second given threshold, and the coarse-grained pruning unit is specifically configured to:
• select one or more weights from weights of the neural network through a sliding window;
• and
• when the selected one or more weights meet a preset condition, set the selected weights to zero;
• determine position information of the target weights among the pruned weights based on the position of the selected weights that are set to zero;
• a coarse-grained selection unit configured to receive the multiple input neurons and the position information of the target weights, and the target weights, and select one or more input neurons corresponding to the target weights according to the position information of the target weights;
• and
• an operation unit configured to perform training according to the pruned weights, and the selected weights that have been set to 0 in the training process remain 0, perform a neural network operation according to the target weights and the selected input neurons corresponding to the target weights to generate the at least one output neuron, and transmit the output neuron to the storage unit as an input neuron of a next layer.