Deep Learning Models

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

Traditional Machine Learning

• Perceptron
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Logistic Regression
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Softmax Regression (Multinomial Logistic Regression)
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Softmax Regression with MLxtend's plot_decision_regions on Iris
     [PyTorch: GitHub | Nbviewer]

Multilayer Perceptrons

• Multilayer Perceptron
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Multilayer Perceptron with Dropout
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Multilayer Perceptron with Batch Normalization
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Multilayer Perceptron with Backpropagation from Scratch
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]

Convolutional Neural Networks

Basic

• Convolutional Neural Network
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Convolutional Neural Network with He Initialization
     [PyTorch: GitHub | Nbviewer]

Concepts

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

Fully Convolutional

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

LeNet

• LeNet-5 on MNIST
     [PyTorch: GitHub | Nbviewer]
• LeNet-5 on CIFAR-10
     [PyTorch: GitHub | Nbviewer]
• LeNet-5 on QuickDraw
     [PyTorch: GitHub | Nbviewer]

AlexNet

• AlexNet on CIFAR-10
     [PyTorch: GitHub | Nbviewer]

VGG

• Convolutional Neural Network VGG-16
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• VGG-16 Gender Classifier Trained on CelebA
     [PyTorch: GitHub | Nbviewer]
• Convolutional Neural Network VGG-19
     [PyTorch: GitHub | Nbviewer]

DenseNet

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

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]

Network in Network

• Network in Network CIFAR-10 Classifier
     [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

• Autoencoder (MNIST)
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Autoencoder (MNIST) + Scikit-Learn Random Forest Classifier
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]

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

• Variational Autoencoder
     [PyTorch: GitHub | Nbviewer]
• Convolutional Variational Autoencoder
     [PyTorch: GitHub | Nbviewer]

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)

• Fully Connected GAN on MNIST
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Fully Connected Wasserstein GAN on MNIST
     [PyTorch: GitHub | Nbviewer]
• Convolutional GAN on MNIST
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Convolutional GAN on MNIST with Label Smoothing
     [TensorFlow 1: GitHub | Nbviewer]
     [PyTorch: GitHub | Nbviewer]
• Convolutional Wasserstein GAN on MNIST
     [PyTorch: GitHub | Nbviewer]

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]
• Bidirectional Multi-layer RNN with LSTM with Own Dataset in CSV Format (Yelp Review Polarity)
     [PyTorch: GitHub | Nbviewer]
• Bidirectional Multi-layer RNN with LSTM with Own Dataset in CSV Format (Amazon Review Polarity)
     [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]

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]

Training and Preprocessing

• Generating Validation Set Splits
  [PyTorch]: GitHub | Nbviewer]
• Dataloading with Pinned Memory
     [PyTorch: GitHub | Nbviewer]
• Standardizing Images
     [PyTorch: GitHub | Nbviewer]
• Image Transformation Examples
     [PyTorch: GitHub | Nbviewer]
• Char-RNN with Own Text File
     [PyTorch: GitHub | Nbviewer]
• Sentiment Classification RNN with Own CSV File
     [PyTorch: GitHub | Nbviewer]

Parallel Computing

• Using Multiple GPUs with DataParallel -- VGG-16 Gender Classifier on CelebA
     [PyTorch: GitHub | Nbviewer]

Other

• 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: GitHub | Nbviewer]