Atomutek

🤘learning to learn.

[Atomutek @atomu_tek]

Power Metal: is it really about dragons?
https://t.co/SPATLDIBwH
#nlp #metal #rhapsody #sabaton #alestorm #manowar #lyrics

[Atomutek @atomu_tek]

@suzatweet My team @mercari_jp is looking for researchers and research engineers in computer vision and NLP ;)

- Research Engineer: https://t.co/3FbRvLQdve
- Researcher: https://t.co/NZubpfw2Ra
- Senior Researcher: https://t.co/SsQuBTve9X

[Jeff Dean @JeffDean]

Nice comprehensive look at the state of federated learning research, including an overview of advances in the last three years (since https://t.co/87U9KkLxl2 in 2016), as well as a nice overview of open problems in the area.

105 pages, 485 references!

[Atomutek @atomu_tek]

@hardmaru "If the paper looks too simple, even if you have great results, you won't pass the reviewers. Add more maths equations".
This was an advice I was given for my field (Robotics). 😓

[Atomutek @atomu_tek]

Interesting paper on how to apply reinforcement learning to growing robots (frogs here): https://t.co/fJrUs6eLJa (from Living Machines 2019 @ Nara)

Would that apply to different humanoid robots?
@hardmaru @julianibarz

[Atomutek @atomu_tek]

@hardmaru That idea is at the back of my mind since the end of thesis...
https://t.co/OOaRkAdNua

[Pierre Sermanet @psermanet]

A major benefit of self-supervision is we can truly scale and adapt on the fly. It could be 10% behind supervised ImageNet, it would still do better in real life. We show in https://t.co/dsikxDuWcI that the longer our model looks at objects, the better it understands them.

[Amazon re:MARS @AmazonreMARS]

#reMARS breakout and leadership sessions are now available - click below to watch! https://t.co/NZkq2GXUm2

[Julian Ibarz @julianibarz]

If we want to scale RL by using methods such as architecture search, or even just grid search of hyper parameters, etc. Doing offline evaluation of policies is critical. Here we propose a new metric that scales well to challenging robotic tasks.

[Atomutek @atomu_tek]

A few days ago, I was wondering if somebody had ever acknowledged a non-academic source in their papers for the origin of their idea. @sidgreddy @ancadianadragan @svlevine did it with a reference to the original Blade Runner dialog between Tyrell and Deckard. Great paper too!

[Sergey Levine @svlevine]

SQIL: an imitation learning method so simple I can summarize in a tweet: drop demonstrations into buffer, set their reward to +1, set reward for all other data to 0, run Q-learning or SAC to train. Why does it work? Find out: https://t.co/GKsYR4zG4l
w/ @sidgreddy @ancadianadragan

[Sergey Levine @svlevine]

Want to try robotic learning on a budget? Check out REPLAB, low-cost setup for robotic learning, with code for learning-based grasping and RL!
https://t.co/QvCo15m2Kt
Presented Wed at #ICRA2019 11-12:45, WeBT1-19.3 220
B. Yang, J. Zhang @dineshjayaraman
https://t.co/tdYAmjTjkz

[Atomutek @atomu_tek]

@anomamarc Maybe. Look into this blog post by a Japanese datascientist: https://t.co/nBhN5sIqzb

[Atomutek @atomu_tek]

Interesting and fun reading about how to get research ideas: https://t.co/dBIgykULkN

[Eric Jang @ericjang11]

One of the best ways to advance the "large-scale Deep RL" agenda for robotics is to provide the world with our system. To that end, we're open-sourcing Tensor2Robot, a library for reducing boilerplate in TensorFlow + robotics research.
https://t.co/3SbQzax2Ee

[Lilian Weng @lilianweng]

One of the hardest problems in robotics is that models trained in simulator normally do not work on real robots. Domain randomization is a simple but powerful idea to close this sim2real gap: https://t.co/MjaiVgVGqf

[Atomutek @atomu_tek]

@suzatweet Comedy club: @punchlinesf near Embarcadero
Robot cafe: @cafexapp
Park: Mission Dolores park
Food: chocolate overload at @LoveGhirardelli

[Atomutek @atomu_tek]

How to create your own walking pattern generator by Sylvain Caron.
https://t.co/cNJrgE7Vy9
It's based on his pymanoid lib (Humanoid robotics prototyping environment) https://t.co/Wt6fBKUhLS

[Atomutek @atomu_tek]

Great article on food distribution chain in Japan by @freaklabs

[Freaklabs @freaklabs]

Introducing a new articles series for hackerfarm: The Food Chain. We'll be looking at food distribution chains in different countries and why it's important to understand them:
https://t.co/KyTFiqZM5a

[Atomutek @atomu_tek]

Fascinating talks for this Robotics symposium @OpenAI 👍🤖
So much food for thought!

[OpenAI @OpenAI]

Our Robotics Symposium is today! Tune in from 9:15amPT to see talks about the future of AI and robotics: https://t.co/0RHDiasJMy

[Atomutek @atomu_tek]

Exciting research by @agilityrobotics in deep reinforcement learning and biped walking!