Abstract: Quantized neural networks significantly reduce storage requirements and computational complexity by lowering the numerical precision of weights and activations. Among these, binary neural ...