NIPS 2018 is approaching. It is very dense. It is desirable to make a good preparation to make the most out of it. I selected NIPS 2018 (RL) Papers to Read, and collected those relevant to Reinforcement Learning in NIPS 2018. In the following, I will talk about my rough plan, and some reflections at the end. This is highly biased towards reinforcement learning (RL).
Day 1: NIPS 2018 Expo (Dec 2)
NIPS 2018 has an industry day Expo program, squeezing lots of great content into one single day. There are two parallel sessions for Talks and Panels, five parallel workshops in the morning and in the afternoon respectively, and nine parallel demonstrations in the morning and in the afternoon respectively.
It must be helpful to break the one day Expo program into two or even three days. Academic research is important; so is industrial research/applications.
I will attend as many Expo programs as possible, and will focus on RL: DL and RL in Intelligent Transportation Systems, Make reinforcement learning in touch with industry, and Deep RL for Online Order Dispatching and Driver Repositioning. It is possible that RL is part of other programs; and it is interesting to see the research/applications of machine learning/AI in industry.
Day 2: Tutorials (Dec 3)
All NIPS tutorials are great to attend. If I have to pick three, I will pick: Adversarial Robustness: Theory and Practice, Unsupervised Deep Learning, and, Automatic Machine Learning.
Day 3-5: Invited Talks, Papers and Demonstrations (Dec 4-6)
The NIPS program is so designed that an invited talk is not in parallel with another program. So we won’t miss them. Joelle Pineau will talk about Reproducible, Reusable, and Robust Reinforcement Learning. Note there is a special Test of Time Award invited talk on Dec 4 at 9:20-9:40.
There are poster sessions twice every day, at 10:45-12:45, and 5:00-7:30. So are demonstration sessions, except for the evening session on Dec 6. Those sessions provide great chances for discussions.
BTW, I attended ACL 2017, and learned that, for ACL, papers selected for oral presentations do not have poster presentations. This is actually a big loss for authors with oral presentations.
It would be very helpful to read some papers before attending the conference. I selected 50+ NIPS 2018 (RL) Papers to Read, and collected those relevant to Reinforcement Learning in NIPS 2018.
Day 6-7: Workshops (Dec 7-8)
Most NIPS workshops are great. I list some below.
- Deep Reinforcement Learning
- Infer to Control: Probabilistic RL and Structured Control
- Reinforcement Learning under Partial Observability
- Wordplay: Reinforcement and Language Learning in Text-based Games
- Imitation Learning and its Challenges in Robotics
- Meta-Learning
- Modeling the Physical World: Learning, Perception, and Control
- Continual Learning
- Causal Learning
- Relational Representation Learning
- Integration of Deep Learning Theories
- Critiquing and Correcting Trends in Machine Learning
- Conversational AI workshop — today’s practice and tomorrow’s potential
- Machine Learning for Systems
- Machine Learning for Intelligent Transportation Systems
- Machine Learning for Health (ML4H): Moving beyond supervised learning in healthcare
- Challenges and Opportunities for AI in Financial Services: the Impact of Fairness, Explainability, Accuracy, and Privacy
Reflecting NIPS 2017 and More
“If a man keeps cherishing his old knowledge, so as continually to be acquiring new, he may be a teacher of others.” (温故而知新,可以为师矣。)
- Turing Awards in AI: Judea Pearl (2011), Leslie G. Valiant (2010), Edward Feigenbaum and Raj Reddy (1994), Allen Newell and Herbert A. Simon (1975), John McCarthy (1971), Marvin Minsky (1969)
- Judea Pearl, The Seven Tools of Causal Inference with Reflections on Machine Learning, The book of why, Causality
- Tutorial: Deep Learning: Practice and Trends, by Nando de Freitas, Scott Reed, Oriol Vinyals
- Tutorial: Reinforcement Learning with People, by Emma Brunskill
- Tutorial: Statistical Relational Artificial Intelligence: Logic, Probability and Computation, by Luc De Raedt, David Poole, Kristian Kersting, Sriraam Natarajan
- Tutorial: Engineering and Reverse-Engineering Intelligence Using Probabilistic Programs, Program Induction, and Deep Learning, by Josh Tenenbaum, Vikash K Mansinghka (ICML 2018)
- Test of Time Award, speech, Reflections, Addendum
- Invited Talk: Pieter Abbeel: Deep Learning for Robotics; Meta-learning
- Invited Talk: Lise Getoor: The Unreasonable Effectiveness of Structure
- David Silver, AlphaZero; Deep RL with Subgoals; Principle of deep RL
- Michael Bowling, Artificial Intelligence Goes All-In
- Doina Precup, Deep RL with Temporal Abstraction, slides
- Demis Hassabis, Learning from first principles; Creativity and AI
- Rich Sutton, The next big step in AI: Planning with a Learned Model
- Andrew Barto, A history of RL, IJCAI 2018; intrinsic motivation
- Yann LeCun, Learning world model: the next step towards AI, IJCAI 2018; Discussion with Chris Manning
- Yoshua Bengio, Disentangled representation; DL for AI
- Juergen Schmidhuber, Learning to learn
- Michael Jordan, Perspectives and Challenges, SysML 2018
- Jitendra Malik, Award for Research Excellence, IJCAI 2018
- Dawn Song, AI and Security, ICML 2018 invited talk
- Yuxi Li, Deep Reinforcement Learning, arXiv, 2018 (intro)