Frame Quantization of Neural Networks