Prototype-based Neural Network Layers: Incorporating Vector Quantization