RL intro and more

[NirViaje]

Prequest

• https://deepmind.com/learning-resources/-introduction-reinforcement-learning-david-silver
• https://www.davidsilver.uk/teaching/

Several reading materials

• https://www.zhihu.com/column/intelligentunit
• https://www.zhihu.com/column/c_29122335
• https://bair.berkeley.edu/blog/
• http://incompleteideas.net/

Topics

MBRL

• https://worldmodels.github.io/
  • https://arxiv.org/abs/1803.10122

• https://danijar.com/project/dreamerv2/
  • https://ai.googleblog.com/2021/02/mastering-atari-with-discrete-world.html
  • https://zhuanlan.zhihu.com/p/137261461
• http://bzhou.ie.cuhk.edu.hk/robotics/MAIRL_draft.pdf

Contrastive Learning

• https://lilianweng.github.io/lil-log/2021/05/31/contrastive-representation-learning.html

Contrastive Learning on RL

• https://arxiv.org/abs/2012.07975
• https://github.com/PhilipZRH/ferm
• https://zhuanlan.zhihu.com/p/355523266

Transfer RL and MT

CS285, Sergey Levine

• http://rail.eecs.berkeley.edu/deeprlcourse/static/slides/lec-21.pdf

CS330, Chelsea Finn

• http://cs330.stanford.edu/slides/cs330_rl_gc_mt_karol.pdf
• https://ai.stanford.edu/~cbfinn/_files/samsung_ai_forum.pdf

Never Ending

• https://sites.google.com/view/neverendingrl

A never-ending learner has the following desiderata:

• It learns behaviors and skills while solving its tasks.
• It invents new subtasks that may later serve as stepping stones.
• It learns hierarchically, i.e. skills learned now can be built upon later.
• It learns without ergodic or resetting assumptions on the underlying (PO)MDP.
• It learns without episode boundaries.
• It learns in a single life without leveraging multiple episodes of experience.

![Uploading image4.gif…]()

MT/QT-Opt

• https://ai.googleblog.com/2021/04/multi-task-robotic-reinforcement.html
  • MT-Opt and Actionable Models
  • https://arxiv.org/abs/2104.08212
  • https://arxiv.org/abs/2104.07749
• https://ai.googleblog.com/2018/06/scalable-deep-reinforcement-learning.html
  • https://arxiv.org/abs/1806.10293
  • https://github.com/quantumiracle/QT_Opt
  • https://zhuanlan.zhihu.com/p/356080761

Random reading

• https://distill.pub/

to name a few

• https://bair.berkeley.edu/blog/
• https://deepmind.com/research/
• https://ai.googleblog.com/

Your good friend -

• https://www.zhihu.com/search?q=reinforcement%20learning

Tools

• https://code.visualstudio.com/download
• https://docs.conda.io/en/latest/
  • https://docs.conda.io/projects/conda/en/4.6.0/_downloads/52a95608c49671267e40c689e0bc00ca/conda-cheatsheet.pdf
• https://meta-world.github.io/

[t6-thu]

RL course of Shusen Wang: https://www.bilibili.com/video/BV1rv41167yx?p=3

[NirViaje]

optimization and dynamic programming vs. supervised learning

• https://bair.berkeley.edu/blog/2020/10/13/supervised-rl/

[NirViaje]

• https://cs330.stanford.edu/slides/cs330_metarl1_2021.pdf

MBRL

• https://worldmodels.github.io/
• http://ai.googleblog.com/2020/03/introducing-dreamer-scalable.html
• https://danijar.com/project/dreamer/
• https://danijar.com/project/dreamerv2/

Meta-RL

RL2: Fast Reinforcement Learning via Slow Reinforcement Learning

• https://arxiv.org/abs/1611.02779
• https://arxiv.org/abs/2005.02934
• http://arxiv.org/abs/1903.08254

Learning to Learn with Probabilistic Task Embeddings

• https://bair.berkeley.edu/blog/2019/06/10/pearl/

EI Seminar - Deepak Pathak - Rapid Adaptation in Robot Learning

• https://www.youtube.com/watch?v=DdweAnoNdtY

Tranformer on RL

• https://arxiv.org/abs/2107.05431

Soft Module

• https://rchalyang.github.io/SoftModule/

[NirViaje]

Grounded Language

Grounding Language to Entities and Dynamics for Generalization in Reinforcement Learning

• https://arxiv.org/abs/2101.07393

Text-based RL Agents with Commonsense Knowledge: New Challenges, Environments and Baselines

• https://arxiv.org/abs/2010.03790

Pierre-Yves Oudeyer

• https://developmentalsystems.org/

Talks

Felix Hill

• https://slideslive.com/38955292/some-reasons-to-do-embodied-machine-learning
• https://slideslive.com/38930841/embodied-language-learning-and-the-power-of-prediction

• https://slideslive.com/38953564/grounded-language-learning-fast-and-slow

Related

• https://en.wikipedia.org/wiki/Fast_mapping

[NirViaje]

Energy-Based Model, Boltzmann Distribution, Maximum Entropy

• https://jaan.io/how-does-physics-connect-machine-learning/
  • 机器学习是如何借鉴物理学思想的？从伊辛模型谈起（万字长文）
  • https://cloud.tencent.com/developer/article/1134088

[NirViaje]

NewtonianVAE

• https://arxiv.org/abs/2006.01959
• https://jaques.xyz/newtonianvae/

Learning Physical Constraints with Neural Projections

• https://proceedings.neurips.cc/paper/2020/file/37bc5e7fb6931a50b3464ec66179085f-Paper.pdf

[NirViaje]

Language grounding and RL

Vision-Language Navigation Policy Learning and Adaptation Chapter, thesis, UCSB, 2020

• https://escholarship.org/uc/item/56087605
• https://eric-xw.github.io/