A method of generating a neural network may be provided. A method may include applying non-linear quantization to a plurality of synaptic weights of a neural network model. The method may further include training the neural network model. Further, the method ma include generating a neural network output from the trained neural network model based on or more inputs received by the trained neural network model.