PyTorch is an open-source Python library for building and training deep learning models. It is used here to build generative models from scratch, with classification tasks providing a practical starting point. Compared with TensorFlow, its appeal is ease of use and flexibility. section 1.1.3 section 2.0
A computational graph tracks operations and the data flowing between them. PyTorch creates and changes this graph as a model runs, which helps with varying architectures and debugging. The graph supports the gradient calculations needed to update model parameters. GPU training can also reduce training time compared with CPU training. section 1.1.3
Classify clothing images
Start with labeled grayscale clothing images and convert their raw pixels into floating-point PyTorch tensors. Build a network with dense layers to predict one of ten clothing types. Use cross-entropy loss to measure predictions against labels, then use the Adam optimizer with a learning rate of 0.001 to update the weights. Keep a validation set to assess the model and help choose training settings. section 2.2 section 2.2.1
The same training foundations carry into generative work: a GAN discriminator makes a binary real-or-fake classification, while text generation can require predicting the next character or word from many possibilities. section 2.0
