Michael Janner @michaeljanner

@openai jannerm.github.io Joined February 2017

Tweets

21
Followers

1K
Following

93
Likes

57

Ilya Sutskever @ilyasut

2 years ago

🤍

6K 4K 32K 28.8M 4K

4 0 15 5K 0

Michael Janner @michaeljanner

2 years ago

OpenAI is nothing without its people

7 17 290 20K 3

Sam Altman @sama

2 years ago

🤍

5K 4K 70K 32.0M 1K

0 0 29 8K 2

We figured out how to train diffusion models with RL to generate images aligned with user goals! Our RL method gets ants to play chess and dolphins to ride bikes. Reward from powerful vision-language models (i.e., RL from AI feedback): rl-diffusion.github.io A 🧵👇

14 177 826 129K 322

Download Image

Sergey Levine @svlevine

3 years ago

Can we do model-based RL just by treating a trajectory like a huge image, and training a diffusion model to generate trajectories? Diffuser does exactly this, guiding generative diffusion models over trajectories with Q-values! diffusion-planning.github.io 🧵->

6 102 620 0 151

Download Gif

Sergey Levine @svlevine

4 years ago

Can we replace RL with one big sequence model? Trajectory transformer models state/action/reward sequences one token (dimension) at a time -- this allows long-horizon prediction and offline RL w/o separate actors, critics, constraints, etc.: trajectory-transformer.github.io A thread:

8 136 619 0 130

Download Gif

Igor Mordatch @IMordatch

4 years ago

Can RL algorithms be replaced with transformer-based language models? We’ve looked at this question with our work on Decision Transformer: Website: sites.google.com/corp/berkeley.… Code: github.com/kzl/decision-t… 1/8

17 250 1K 0 281

Download Gif

Sergey Levine @svlevine

6 years ago

Model-Based Reinforcement Learning: Theory and Practice bair.berkeley.edu/blog/2019/12/1… New blog post by @michaeljanner about the taxonomy of model-based RL methods, when we should use models, and the state-of-the-art MBPO algorithm for sample-efficient RL.

0 79 321 0 60

Sergey Levine @svlevine

6 years ago

When should we use a model to improve RL? We've analyzed this theoretically and empirically, with a monotonic improvement result, error accumulation study (vid below), and proposing the most efficient RL method yet, MBPO people.eecs.berkeley.edu/~janner/mbpo/ w/ @michaeljanner, J. Fu, M. Zhang

2 52 224 0 33

Download Video

Sergey Levine @svlevine

6 years ago

How can we use object-based physics prediction models to construct structures out of blocks? Michael Janner will be presenting our work on predictive models tomorrow (Wed) at 11 am at #ICLR youtube.com/watch?v=CXS7dR…

0 17 65 0 5

Tejas Kulkarni @tejasdkulkarni

8 years ago

Our NIPS paper on self-supervised intrinsic image (shape,reflectance,lighting) decomposition. This is work from MIT days. Someday this might become useful for robot manipulation or image editing -- papers.nips.cc/paper/7175-lea….