Python 3.7

Deep Learning Models

A collection of various deep learning architectures, models, and tips for TensorFlow and PyTorch in Jupyter Notebooks.

Traditional Machine Learning

Multilayer Perceptrons

Convolutional Neural Networks

Basic

Concepts

  • Replacing Fully-Connnected by Equivalent Convolutional Layers
       [PyTorch: GitHub | Nbviewer]

AlexNet

DenseNet

  • DenseNet-121 Digit Classifier Trained on MNIST
       [PyTorch: GitHub | Nbviewer]
  • DenseNet-121 Image Classifier Trained on CIFAR-10
       [PyTorch: GitHub | Nbviewer]

Fully Convolutional

  • Fully Convolutional Neural Network
       [PyTorch: GitHub | Nbviewer]

LeNet

MobileNet

Network in Network

  • Network in Network CIFAR-10 Classifier
       [PyTorch: GitHub | Nbviewer]

VGG

ResNet

  • ResNet and Residual Blocks
       [PyTorch: GitHub | Nbviewer]
  • ResNet-18 Digit Classifier Trained on MNIST
       [PyTorch: GitHub | Nbviewer]
  • ResNet-18 Gender Classifier Trained on CelebA
       [PyTorch: GitHub | Nbviewer]
  • ResNet-34 Digit Classifier Trained on MNIST
       [PyTorch: GitHub | Nbviewer]
  • ResNet-34 Object Classifier Trained on QuickDraw
       [PyTorch: GitHub | Nbviewer]
  • ResNet-34 Gender Classifier Trained on CelebA
       [PyTorch: GitHub | Nbviewer]
  • ResNet-50 Digit Classifier Trained on MNIST
       [PyTorch: GitHub | Nbviewer]
  • ResNet-50 Gender Classifier Trained on CelebA
       [PyTorch: GitHub | Nbviewer]
  • ResNet-101 Gender Classifier Trained on CelebA
       [PyTorch: GitHub | Nbviewer]
  • ResNet-101 Trained on CIFAR-10
       [PyTorch: GitHub | Nbviewer]
  • ResNet-152 Gender Classifier Trained on CelebA
       [PyTorch: GitHub | Nbviewer]

Normalization Layers

  • BatchNorm before and after Activation for Network-in-Network CIFAR-10 Classifier
       [PyTorch: GitHub | Nbviewer]
  • Filter Response Normalization for Network-in-Network CIFAR-10 Classifier
       [PyTorch: GitHub | Nbviewer]

Metric Learning

  • Siamese Network with Multilayer Perceptrons
       [TensorFlow 1: GitHub | Nbviewer]

Autoencoders

Fully-connected Autoencoders

Convolutional Autoencoders

  • Convolutional Autoencoder with Deconvolutions / Transposed Convolutions
       [TensorFlow 1: GitHub | Nbviewer]
       [PyTorch: GitHub | Nbviewer]
  • Convolutional Autoencoder with Deconvolutions and Continuous Jaccard Distance
       [PyTorch: GitHub | Nbviewer]
  • Convolutional Autoencoder with Deconvolutions (without pooling operations)
       [PyTorch: GitHub | Nbviewer]
  • Convolutional Autoencoder with Nearest-neighbor Interpolation
       [TensorFlow 1: GitHub | Nbviewer]
       [PyTorch: GitHub | Nbviewer]
  • Convolutional Autoencoder with Nearest-neighbor Interpolation — Trained on CelebA
       [PyTorch: GitHub | Nbviewer]
  • Convolutional Autoencoder with Nearest-neighbor Interpolation — Trained on Quickdraw
       [PyTorch: GitHub | Nbviewer]

Variational Autoencoders

Conditional Variational Autoencoders

  • Conditional Variational Autoencoder (with labels in reconstruction loss)
       [PyTorch: GitHub | Nbviewer]
  • Conditional Variational Autoencoder (without labels in reconstruction loss)
       [PyTorch: GitHub | Nbviewer]
  • Convolutional Conditional Variational Autoencoder (with labels in reconstruction loss)
       [PyTorch: GitHub | Nbviewer]
  • Convolutional Conditional Variational Autoencoder (without labels in reconstruction loss)
       [PyTorch: GitHub | Nbviewer]

Generative Adversarial Networks (GANs)

Graph Neural Networks (GNNs)

  • Most Basic Graph Neural Network with Gaussian Filter on MNIST
       [PyTorch: GitHub | Nbviewer]
  • Basic Graph Neural Network with Edge Prediction on MNIST
       [PyTorch: GitHub | Nbviewer]
  • Basic Graph Neural Network with Spectral Graph Convolution on MNIST
       [PyTorch: GitHub | Nbviewer]

Recurrent Neural Networks (RNNs)

Many-to-one: Sentiment Analysis / Classification

  • A simple single-layer RNN (IMDB)
       [PyTorch: GitHub | Nbviewer]
  • A simple single-layer RNN with packed sequences to ignore padding characters (IMDB)
       [PyTorch: GitHub | Nbviewer]
  • RNN with LSTM cells (IMDB)
       [PyTorch: GitHub | Nbviewer]
  • RNN with LSTM cells (IMDB) and pre-trained GloVe word vectors
       [PyTorch: GitHub | Nbviewer]
  • RNN with LSTM cells and Own Dataset in CSV Format (IMDB)
       [PyTorch: GitHub | Nbviewer]
  • RNN with GRU cells (IMDB)
       [PyTorch: GitHub | Nbviewer]
  • Multilayer bi-directional RNN (IMDB)
       [PyTorch: GitHub | Nbviewer]
  • Bidirectional Multi-layer RNN with LSTM with Own Dataset in CSV Format (AG News)
       [PyTorch: GitHub | Nbviewer]

Many-to-Many / Sequence-to-Sequence

  • A simple character RNN to generate new text (Charles Dickens)
       [PyTorch: GitHub | Nbviewer]

Ordinal Regression

  • Ordinal Regression CNN — CORAL w. ResNet34 on AFAD-Lite
       [PyTorch: GitHub | Nbviewer]
  • Ordinal Regression CNN — Niu et al. 2016 w. ResNet34 on AFAD-Lite
       [PyTorch: GitHub | Nbviewer]
  • Ordinal Regression CNN — Beckham and Pal 2016 w. ResNet34 on AFAD-Lite
       [PyTorch: GitHub | Nbviewer]

Tips and Tricks

  • Cyclical Learning Rate
       [PyTorch: GitHub | Nbviewer]
  • Annealing with Increasing the Batch Size (w. CIFAR-10 & AlexNet)
       [PyTorch: GitHub | Nbviewer]
  • Gradient Clipping (w. MLP on MNIST)
       [PyTorch: GitHub | Nbviewer]

Transfer Learning

  • Transfer Learning Example (VGG16 pre-trained on ImageNet for Cifar-10)
       [PyTorch: GitHub | Nbviewer]

Visualization and Interpretation

  • Vanilla Loss Gradient (wrt Inputs) Visualization (Based on a VGG16 Convolutional Neural Network for Kaggle’s Cats and Dogs Images)
       [PyTorch: GitHub | Nbviewer]
  • Guided Backpropagation (Based on a VGG16 Convolutional Neural Network for Kaggle’s Cats and Dogs Images)
       [PyTorch: GitHub | Nbviewer]

PyTorch Workflows and Mechanics

Custom Datasets

  • Custom Data Loader Example for PNG Files
       [PyTorch: GitHub | Nbviewer]
  • Using PyTorch Dataset Loading Utilities for Custom Datasets — CSV files converted to HDF5
       [PyTorch: GitHub | Nbviewer]
  • Using PyTorch Dataset Loading Utilities for Custom Datasets — Face Images from CelebA
       [PyTorch: GitHub | Nbviewer]
  • Using PyTorch Dataset Loading Utilities for Custom Datasets — Drawings from Quickdraw
       [PyTorch: GitHub | Nbviewer]
  • Using PyTorch Dataset Loading Utilities for Custom Datasets — Drawings from the Street View House Number (SVHN) Dataset
       [PyTorch: GitHub | Nbviewer]
  • Using PyTorch Dataset Loading Utilities for Custom Datasets — Asian Face Dataset (AFAD)
       [PyTorch: GitHub | Nbviewer]
  • Using PyTorch Dataset Loading Utilities for Custom Datasets — Dating Historical Color Images
       [PyTorch: GitHub | Nbviewer]
  • Using PyTorch Dataset Loading Utilities for Custom Datasets — Fashion MNIST
       [PyTorch: GitHub | Nbviewer]

Training and Preprocessing

Improving Memory Efficiency

  • Gradient Checkpointing Demo (Network-in-Network trained on CIFAR-10)
       [PyTorch: GitHub | Nbviewer]

Parallel Computing

  • Using Multiple GPUs with DataParallel — VGG-16 Gender Classifier on CelebA
       [PyTorch: GitHub | Nbviewer]
  • Distribute a Model Across Multiple GPUs with Pipeline Parallelism (VGG-16 Example)    [PyTorch: GitHub | Nbviewer]

Other

  • PyTorch with and without Deterministic Behavior — Runtime Benchmark
       [PyTorch: GitHub | Nbviewer]
  • Sequential API and hooks
       [PyTorch: GitHub | Nbviewer]
  • Weight Sharing Within a Layer
       [PyTorch: GitHub | Nbviewer]
  • Plotting Live Training Performance in Jupyter Notebooks with just Matplotlib
       [PyTorch: GitHub | Nbviewer]

Autograd

  • Getting Gradients of an Intermediate Variable in PyTorch
       [PyTorch: GitHub | Nbviewer]

TensorFlow Workflows and Mechanics

Custom Datasets

  • Chunking an Image Dataset for Minibatch Training using NumPy NPZ Archives
       [TensorFlow 1: GitHub | Nbviewer]
  • Storing an Image Dataset for Minibatch Training using HDF5
       [TensorFlow 1: GitHub | Nbviewer]
  • Using Input Pipelines to Read Data from TFRecords Files
       [TensorFlow 1: GitHub | Nbviewer]
  • Using Queue Runners to Feed Images Directly from Disk
       [TensorFlow 1: GitHub | Nbviewer]
  • Using TensorFlow’s Dataset API
       [TensorFlow 1: GitHub | Nbviewer]

Training and Preprocessing

  • Saving and Loading Trained Models — from TensorFlow Checkpoint Files and NumPy NPZ Archives
       [TensorFlow 1: | Nbviewer]

GitHub

GitHub - rasbt/deeplearning-models: A collection of various deep learning architectures, models, and tips
A collection of various deep learning architectures, models, and tips - GitHub - rasbt/deeplearning-models: A collection of various deep learning architectures, models, and tips