PyTorch-Collections
A collection of AIs made with PyTorch
AIs
Low Level
- Linear regression : Linear regression using only AutoGrad with momentum optimizer, MSE loss
Image
- Classification : Convolutional neural network which classifies cats / dogs
- Object Detection : Detects where dogs / cats are within an image
- Auto Encoder : An auto encoder with pytorch.nn module for CIFAR10 images
- Deep Auto Encoder : An auto encoder with convolutional layers which generates MNIST digits
- Denoiser : Simple denoiser using only fully connected layers for the MNIST dataset
- GAN : Simple Generative Adversarial Network using only fully connected layers, generates MNIST like handwritten digits
Reinforcement Learning
- REINFORCE : REINFORCE algorithm (policy gradient) for gym's CartPole environment
- A2C : Advantage Actor Critic algorithm (inspired by REINFORCE algorithm) for gym's CartPole environment
- A2C : A2C on LunarLander with Experience Buffer
- PPO : PPO using clipped objective
- DQN : Deep Q Leaning implementation with basic replay buffer (CartPole env again)
- DDQN : Double Deep Q Leaning implementation on LunarLander-v2
- PER : Prioritized Experience Replay, this method takes O(n) time to get / O(1) time to add (unlike in the original paper), no IS weights
Images
Denoiser
GAN
REINFORCE
A2C
DDQN
