| 摘要 |
The invention proposes a convolutional neural network (100) comprising a plurality of artificial neurons arranged in one or several convolution layers, each convolution layer comprising one or more output matrices (14), each output matrix comprising a set of output neurons, each output matrix being connected to an input matrix, comprising a set of input neurons, by artificial synapses associated with a convolution matrix. The convolution matrix comprises the weight coefficients associated with the output neurons of the output matrix, the output value of each output neuron being determined from the input neurons of the input matrix to which the output neuron is connected and the weight coefficients of the convolution matrix associated with the output matrix. Each synapse is constituted by a set of memristive devices comprising at least one memristive device, each set of memristive devices storing a weight coefficient of the convolution matrix. In response to a change in the output value of an input neuron of an input matrix, the neural network is capable of dynamically associating each set of memristive devices storing a coefficient of the convolution matrix with an output neuron connected to the input neuron. The neural network further comprises an accumulator (140, 121) for each output neuron, the accumulator being configured to accumulate the values of the weight coefficients stored in the sets of memristive devices dynamically associated with the output neuron, the output value of the output neuron being determined from the value accumulated in the accumulator. |
