Samsung patents method to fix errors in simplified AI calculations

1. A computing system comprising: a neural network system configured to drive an artificial neural network (ANN); and a quantization system configured to quantize the ANN, wherein the quantization system is configured to quantize parameters of the ANN to generate quantized parameters of the ANN, generate a quantization error of the parameters of the ANN based on the parameters of the ANN and the quantized parameters, generate a compensation bias based on the quantized parameters and the quantization error of the parameters of the ANN, and transmit the quantized parameters and the compensation bias to the neural network system, wherein the parameters of the ANN comprise a weight and a bias of the ANN and the quantized parameters include a quantized weight and a quantized bias, wherein the compensation bias is based on a result of at least one of a second multiply-accumulate (MAC) operation or a result of a third MAC operation, wherein the second MAC operation includes a reference sample and a quantization error of the weight as inputs and the third MAC operation includes a quantization error of the reference sample and the quantized weight as inputs, and wherein the neural network system is configured to receive a quantized input sample, receive the quantized parameters transmitted from the quantization system, receive the generated compensation bias transmitted from the quantization system, perform a first MAC operation based on the quantized input sample and the quantized parameters as inputs to the ANN, and generate a final operation result by applying the compensation bias to a result of the first MAC operation.