dodoseung/deep-learning-pytorch

The pytorch implementation of deep learning

Deep Learning Pytorch

This project is a set of repositories written by Seungwon Do including various deep learning algorithms using Pytorch.

Reinforcement Learning

Model-free Reinforcement Learning

1. Deep Q Network (DQN)
   • Implementation
     • Numeric state and discrete action
     • Image state and discrete action
   • Reference paper
     • Playing Atari with Deep Reinforcement Learning
2. Vanilla Policy Gradient (VPG)
   • Implementation
     • Numeric state and discrete action
3. Actor Critic (AC)
   • Implementation
     • Numeric state and discrete action
4. Advantage Actor-Critic (A2C)
   • Implementation
     • Numeric state, discrete action, and training per step
     • Numeric state, discrete action, and training per episode
   • Reference paper
     • Asynchronous Methods for Deep Reinforcement Learning
5. Asynchronous Advantage Actor Critic (A3C)
   • Implementation
     • Numeric state and discrete action
   • Reference paper
     • Asynchronous Methods for Deep Reinforcement Learning
6. Deep Deterministic Policy Gradient (DDPG)
   • Implementation
     • Numeric state, continuous action, and deterministic action
   • Reference paper
     • Continuous control with deep reinforcement learning
7. Trust Region Policy Optimization (TRPO)
   • Implementation
     • Numeric state and discrete action
     • Numeric state, continuous action, and stochastic action
   • Reference paper
     • Trust Region Policy Optimization
8. Proximal Policy Optimization (PPO)
   • Implementation
     • Numeric state, discrete action, and gae
     • Numeric state, continuous action, stochastic action, and gae
   • Reference paper
     • Proximal Policy Optimization Algorithms
9. Twin Delayed Deep Deterministic Policy Gradient (TD3)
   • Implementation
     • Numeric state, continuous action, and deterministic action
   • Reference paper
     • Addressing Function Approximation Error in Actor-Critic Methods
10. Soft Actor-Critic (SAC)
    • Implementation
      • Numeric state, discrete action, and entropy
      • Numeric state, continuous action, and stochastic action
      • Numeric state, continuous action, stochastic action, and entropy
    • Reference paper
      • Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor
      • Soft Actor-Critic Algorithms and Applications
      • Soft Actor-Critic for Discrete Action Settings

Computer Vision

Image Classification and Semantic Segmentation

1. VGG
   • Implementation
     • VGG model
     • Image classification with CIFAR10 dataset
   • Reference paper
     • Very Deep Convolutional Networks for Large-Scale Image Recognition
2. ResNet
   • Implementation
     • ResNet model
     • Image classification with CIFAR10 dataset
   • Reference paper
     • Deep Residual Learning for Image Recognition
3. U-Net
   • Implementation
     • U-Net model
     • Semantic segmentation with Cityscapes dataset
   • Reference paper
     • U-Net: Convolutional Networks for Biomedical Image Segmentation
4. WaveMix-Lite
   • Implementation
     • WaveMix-Lite model
     • Image classification with CIFAR10 dataset
     • Image classification with Places-365 dataset
     • Semantic segmentation with Cityscapes dataset
   • Reference paper
     • WaveMix-Lite: A Resource-efficient Neural Network for Image Analysis
5. ViT
   • Implementation
     • ViT model
     • Image classification with CIFAR10 dataset
     • Image classification with Places365 dataset
   • Reference paper
     • An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Generative Model

1. Auto Encoder (AE)
   • Implementation
     • Denoise with fashion MNIST dataset
2. Variational Auto Encoder (VAE)
   • Implementation
     • VAE model
     • Image generation with EMNIST dataset
   • Reference paper
     • Auto-Encoding Variational Bayes
3. Residual Encoder-Decoder Network (REDNet)
   • Implementation
     • Super resolution with CIFAR10 dataset
   • Reference paper
     • Image Restoration Using Very Deep Convolutional Encoder-Decoder Networks with Symmetric Skip Connections
4. Generative Adversarial Network (GAN)
   • Implementation
     • GAN model
     • Image generation with EMNIST dataset
   • Reference paper
     • Generative Adversarial Nets
5. Wasserstein Generative Adversarial Network (WGAN)
   • Implementation
     • WGAN model
     • Image generation with CIFAR10 dataset
   • Reference paper
     • Wasserstein GAN
6. Wasserstein Generative Adversarial Network Gradient Penalty (WGAN-GP)
   • Implementation
     • WGAN-GP model
     • Image generation with CIFAR10 dataset
   • Reference paper
     • Improved Training of Wasserstein GANs
7. Cycle Generative Adversarial Network (CycleGAN)
   • Implementation
     • CycleGAN model with U-net
     • Style transfer with apple2orange dataset
   • Reference paper
     • Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks

Natural Language Processing

Pre-trained Language Model

1. Transformer
   • Implementation
     • Transformer Model
     • Translator DE to EN
   • Reference paper
     • Attention Is All You Need
2. Generative Pre-Training 2 (GPT-2)
   • Implementation
     • GPT2 Model
   • Reference paper
     • Language Models are Unsupervised Multitask Learners