Table of Contents

Part 1: Getting Started

Introduction (AVAILABLE)

Part 2: Inference (AVAILABLE)

Representing layers and connections (AVAILABLE)

Bias and activation function (AVAILABLE)

Fully connected inference layers (AVAILABLE)

Increasing performance with batch processing (AVAILABLE)

Sharing memory (AVAILABLE)

GPU computing with CUDA and OpenCL (AVAILABLE)

Part 3: Learning (AVAILABLE)

Gradient descent and backpropagation (AVAILABLE)

The forward pass (AVAILABLE)

The activation and its derivative (AVAILABLE)

The backward pass (AVAILABLE)

Part 4: A simple neural networks API (AVAILABLE)

Inference API (AVAILABLE)

Training API (AVAILABLE)

Initializing weights (AVAILABLE)

Regression: learning a known function (AVAILABLE)

Part 5: Training optimizations (AVAILABLE)

Weight decay (AVAILABLE)

Momentum and Nesterov momentum (AVAILABLE)

Adaptive learning rates (AVAILABLE)

Regression: Boston housing prices (AVAILABLE)

Dropout (AVAILABLE)

Stochastic gradient descent (AVAILABLE)

Classification: IMDB sentiments (AVAILABLE)

Part 6: Tensors (AVAILABLE)

Classification and metrics: MNIST handwritten digits recognition (AVAILABLE)

Tensors and ND arrays (AVAILABLE)

Tensor transformations (AVAILABLE)

DNNL: Tensors on the CPU (AVAILABLE)

Tensor-based neural networks (AVAILABLE)

cuDNN: Tensors on the GPU (AVAILABLE)

Part 7: Convolutional networks (AVAILABLE)

The convolution operation (AVAILABLE)

Convolutional layers on the CPU with DNNL (AVAILABLE)

Convolutional neural networks (CNN): Fashion-MNIST (AVAILABLE)

CNN on the GPU with cuDNNL (AVAILABLE)

Part 8: Recurrent neural networks (IN PRORESS)

Recurrent layers (AVAILABLE)

Recurrent neural networks (RNN) (AVAILABLE)

Recurrent layers on the CPU with DNNL (AVAILABLE)

Recurrent layers on the GPU with cuDNN (AVAILABLE)

(Future editions, TBD)

Appendix

Setting up the environment and the JVM (AVAILABLE)