Awesome-AGI-Paper

Our dream is creating real AI who is suitable for all domains, here are some papers we high recommanded

Representation learning

AutoEncoder

• Auto-Encoding Variational Bayes[ORIGINAL]
• vq-vae2 [SOTA]
• Adversarial Collaboration: Joint Unsupervised Learning of Depth, Camera Motion, Optical Flow and Motion Segmentation

Disentangled-representations

• Learning Disentangled Joint Continuous and Discrete Representations
• Understanding disentangling in β-VAE
• Isolating Sources of Disentanglement in VAEs
• Ladder Variational Autoencoders
• Disentangling by Factorising[FACTOR-VAE]
• VARIATIONAL INFERENCE OF DISENTANGLED LATENT CONCEPTS FROM UNLABELED OBSERVATIONS[DIPVAE]

General theory of VAE

• Recent Advances in Autoencoder-Based Representation Learning
• Are Disentangled Representations Helpful for Abstract Visual Reasoning?

GAN

• f-GAN: Training Generative Neural Samplers using Variational Divergence Minimization
• INFO-GAN
• WGAN
• Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks

DRL

Task-Agnostic Reinforcement Learning

• Workshop at ICLR, 06 May 2019, New Orleans Building agents that explore and learn in the absence of rewards[24papers]

Successor Representations

• Predicting the Future with Multi-scale Successor Representations

OPTIONS SKILL

• Learning Abstract Options
• Successor Options: An Option Discovery Framework for Reinforcement Learning

Model base

• LEARNING AWARENESS MODELS ICLR 2018
• Curiosity-driven Exploration by Self-supervised Prediction ICML 2017 [curiosity][ICM]
• Learning Latent Dynamics for Planning from Pixels ICML 2018 [MPC][VAE] [planet]
• Dynamics-Aware Unsupervised Discovery of Skills
• INFOBOT
• EMI
• Unsupervised Discovery of Decision States for Transfer in Reinforcement Learning

Model free

• TRPO
• PPO
• DDPG
• TD3
• SAC
• RECURRENT REINFORCEMENT LEARNING:A HYBRID APPROACH

Meta learning

• MBMPO

Memory

HUBE

• A BIOLOGICALLY INSPIRED VISUAL WORKING MEMORY FOR DEEP NETWORKS

NTM

• Learning to Remember More with Less Memorization

Episode memory

• Learning to Learn without Forgetting by Maximizing Transfer and Minimizing Interference

Association memory

• Dense Associative Memory is Robust to Adversarial Inputs

Few shot learning

• Meta-Transfer Learning for Few-Shot Learning

Mutual infomation

• Unsupervised Discovery of Decision States for Transfer in Reinforcement Learning

• Dynamics-Aware Unsupervised Discovery of Skills

• [Contrastive Bidirectional Transformer for Temporal Representation Learning]

• [Contrastive Multiview Coding]

• Infobot

• DIM

• cpc

• EMI

• MINE

• Learning Belief Representations for Imitation Learning in POMDPs

• INFORMATION ASYMMETRY IN KL-REGULARIZED RL

• Exploiting Hierarchy for Learning and Transfer in KL-regularized RL

• Learning to Share and Hide Intentions using Information Regularization

General

• THE THERMODYNAMICS OF MACHINE LEARNING

Key point

Curiosity-driven Exploration by Self-supervised Prediction

Learning a predictor and use predict error as internal reward, and they jointly train an inverse dynamics model for encoder, who project observation to a space that is invariant to parts of the environment that do not affect the agent or task.

EMI

• Learn the representation of states and the action such that the representation of the corresponding next state following linear dynamics
• Intrinsic reward augmentation
• https://github.com/snu-mllab/EMI

THE THERMODYNAMICS OF MACHINE LEARNING

In this work we offer an information-theoretic framework for representation learn- ing that connects with a wide class of existing objectives in machine learning. We develop a formal correspondence between this work and thermodynamics and dis- cuss its implications.

RECURRENT REINFORCEMENT LEARNING:A HYBRID APPROACH

propose a new family of hybrid models that combines the strength of both supervised learning (SL) and reinforcement learning (RL), trained in a joint fashion: The SL component can be a recurrent neural networks (RNN) or its long short-term memory (LSTM) version, which is equipped with the desired property of being able to capture long-term dependency on history, thus providing an effective way of learning the representation of hidden states. The RL component is a deep Q-network (DQN) that learns to optimize the control for maximizing long-term rewards. Extensive experiments in a direct mailing campaign problem demonstrate the effectiveness and advantages of the proposed approach

Dynamics-Aware Unsupervised Discovery of Skills

Graphical models, Information Bottleneck and Unsupervised Skill Learning