Code Collection

This post serves as a code collection of my personal implementations for different algorithms, which cover general deep learning algorithms and some paper algorithms.

Content

• General Algorithms
  • Deep Deterministic Policy Gradient
  • Mixture Density Networks
  • Neural Machine Translation
  • Residual Networks
  • Variational Autoencoder
• Paper Algorithms
  • Deep Q-Network
  • Large-Scale Study of Curiosity-Driven Learning
  • Prioritized Experience Replay
  • World Models

General Algorithms

This category contains stand alone implementation of general machine learning algorithms.

Deep Deterministic Policy Gradient

• Reinforcement learning algorithm.
• Policy optimization algorithm that works well in continous action space.
• Repository: DDPG

Mixture Density Networks

• Neural network architecture.
• Network model that learns a mixture of probability distributions of the input variables.
• Repository: MDN

Neural Machine Translation

• Neural network architecture.
• Sequence to sequence model that transforms an input sequence into an output sequence.
• Repository: NMT

Residual Networks

• Neural network architecture.
• Network model that uses residual connection to resolve vanishing gradient problem of deep neural networks.
• Repository: ResNet

Variational Autoencoder

• Neural network architecture.
• Generative model that learns useful features for image reconstruction while restricting the features close to normal distribution.
• Repository: VAE

Paper Algorithms

This category contains implementation of specific algorithms proposed in some papers.

Deep Q-Network

• Reinforcement learning algorithm.
• The first algorithm that achieves human-level performance on Atari games with raw-pixel input.
• Repository: Deep Q-Network

Large-Scale Study of Curiosity-Driven Learning

• Reinforcement learning algorithm.
• Explore what will happen if an agent is trained purely by its own curiosity.
• Repository: Large-Scale Study of Curiosity-Driven Learning

Prioritized Experience Replay

• Reinforcement learning algorithm.
• An extension to Deep Q-Network (DQN). Samples with higher training error is more likely to be sampled for training.
• Repository: PER

World Models

• Reinforcement learning algorithm.
• Build an agent inspired by our own cognitive system: a vision model, a memory model, and a controller model.
• Repository: World Models

Twitter  facebook