Dynamic Block-Wise Local Learning Algorithm for Efficient Neural Network Training

In the backpropagation algorithm, the error calculated from the output of the neural network should backpropagate the layers to update the weights of each layer, making it difficult to parallelize the training process and requiring frequent off-chip memory access. Local learning algorithms locally generate error signals which are used for weight updates, removing the need for backpropagation of er...