This repository serves as a playground for experimental deep learning implementations in Keras, providing reference implementations of neural network architectures including Variational Autoencoders, various GAN variants (Wasserstein, Least Squares, Generalized Loss Sensitive), normalization techniques (weight, layer), compression methods (binary/ternary/XNOR networks), and modern architectures like QRNN, gated CNNs, and Squeeze-and-Excitation networks. All scripts default to TensorFlow backend with channels-first data format.