一些主题文章集合

一些论文集合

context-based meta RL

https://mp.weixin.qq.com/s/hMTm38gCccxt-Jnz28Xx1A

PEARL, 2019

MQL, 2020

Enhancing Context-Based Meta-Reinforcement Learning Algorithms via An Efficient Task Encode, 2021 华为诺亚 南大 *

Towards Effective Context for Meta-Reinforcement Learning: An Approach Based on Contrastive Learning,2021 华为诺亚 天大 *

META-REINFORCEMENT LEARNING ROBUST TO DISTRIBUTIONAL SHIFT VIA MODEL IDENTIFICATION AND EXPERIENCE RELABELING, C Finn 2021 * (Model-based )

Representation

Decouple exploration and exploitation

Learning a good task embedding

 

 

Sample efficiency

Off-policy tricks

Model-Based RL

Exploration

Unsupervised training: reward/task proposal

Intrinsic rewards

 

 

 

Batch RL

BCQ(icml’19, nipsDRLW’19),

BEAR(nips’19),

REM(iclr’20 rej),

ABM(iclr’20),

BRPO(2002.icml?),

BRAC-QL (iclr’20 rej),

BAIL(icml’20 under review),

D4RL (2004.sergey.dataset)

作者：爱笑的Groza
链接：https://zhuanlan.zhihu.com/p/143291642
来源：知乎
著作权归作者所有。商业转载请联系作者获得授权，非商业转载请注明出处。

 

Skill Discovery based on Mutual Information

option找"元动作"

1. 2016 Variational Information Maximisation for Intrinsically Motivated Reinforcement Learning

2. Variational Intrinsic Control

3. Diversity is All You Need: Learning Skills without a Reward Function

4. Dynamics-Aware Unsupervised Discovery of Skills

5. Explore, Discover and Learn:Unsupervised Discovery of State-Covering Skills

6. Dynamics Aware Embedding

7. Mutual Information State Intrinsic Control