Abstract: Method and apparatus for training an artificial neural network having an input layer, an output layer and at least one intervening hidden layer. A training circuit stores, in a memory, node input and output values and a loss function value at the output layer responsive to input data applied to the input layer. For a selected node, a set of downstream nodes interconnected along a single chain path from the selected node to the output layer are identified. The parameter values of the selected node are adjusted, a new node output value for the selected node is determined, and an updated loss function value at the output layer is generated by feeding forward the new node output value to each of the downstream nodes using chain isolation optimization without backpropagation. Look up tables can be used to eliminate mathematical calculations for the updates to each node.

Abstract: Apparatus and method for implementing an Artificial Neural Network (ANN) section which eliminates the need for backpropagation during training. The ANN section has a plurality of integer gate logic (IGL) nodes respectively arranged into an input layer, an output layer, and at least one hidden layer. Each node has multiple inputs and a single output, and uses a non-differentiable activation function to emulate Boolean logic functions (including XOR), near-Boolean functions, and unknown functions, based on one or more selectable weight values. A chain isolation optimization process is used to select and isolate each node during training to assess the impact of the different weight parameters on the output. Enhanced error functions, batch processing scheduling, and random node selection techniques can be used during training. The nodes can be singly connected, or arranged into convolutional filters or localized fully interconnected layers.