Robocrunch        AI
  Tweet  

Animesh Garg    @animesh_garg   ·   9/15/2021
Data Augmentation helps with performance improvement in Offline RL. - across the board with most envs and most algorithms! Surprising finding and hence the name!
 Reply      Retweet   6      Like     32    






  Similar Tweets  

DeepMind    @DeepMind   ·   9/16/2021
In part two of the model-free lecture, Hado explains how to use prediction algorithms for policy improvement, leading to algorithms - like Q-learning - that can learn good behaviour policies from sampled experience: https://t.co/o5qgfTekqN #DeepMindxUCL @ai_ucl
 Reply      Retweet   2      Like     7    

Posted by:
DeepMind


Chelsea Finn    @chelseabfinn   ·   6/25/2021
When training video prediction models on broad datasets, they fail to even fit the training data. FitVid is a simple model that addresses this problem & yields good performance with data augmentation. Led by @babaeizadeh, with @msaffar3 @SurajNair_1 @svlevine @doomie
 Reply      Retweet   28      Like     120    



Samarth Sinha    @_sam_sinha_   ·   9/14/2021
Our paper on “Surprisingly Simple Self-Supervision for Offline RL” got accepted at CoRL 2021! Find out how simple data augmentations from states can help Q-learning algorithms on a variety of robotics tasks! https://t.co/ZwMG5mHcXE J/w: @AjayMandlekar @animesh_garg
 Reply      Retweet   1      Like     3    



Brenden Lake    @LakeBrenden   ·   7/6/2021
Self-supervised learning works well on vision+language data, but is it relevant to how children learn words? @wkvong considers 7 behavioral phenomena, finding recent algorithms capture many (and learn quickly), but fail when mutual exclusivity is required https://t.co/8NWkewc3Rv
 Reply      Retweet   24      Like     128    



Xiaolong Wang    @xiaolonw   ·   7/2/2021
Introducing 𝗦𝗩𝗘𝗔, Stabilizing Value Estimation under Augmentation. While RAD / DrQ provided great studies on how data aug affects RL, most augmentations make training unstable. SVEA stabilizes the training across various augmentations and generalize RL w/ Vision Transformers.
 Reply      Retweet   2      Like     24    



Nils Reimers    @Nils_Reimers   ·   6/3/2021
Small & Fast Models 🏎️💨 We added several small & fast models, for optimal encoding speed on GPU & CPU. Multi-Lingual Models 🇺🇳 Multi-lingual models for 50+ languages are available. They achieve by far the best performance across all available multilingual models for many tasks. h
 Reply      Retweet   2      Like     21    



Arsenii Ashukha    @senya_ashuha   ·   6/5/2021
Why test-time data augmentation drives uncertainty on imagenet in a single image
 Reply      Retweet   4      Like     32    



François Chollet    @fchollet   ·   9/3/2021
In our example, we kept the data augmentation stage in float32 (created before setting the global policy) since it's meant to be run on CPU as part of the TF data pipeline.
 Reply      Retweet        Like     14    



Sergey Levine    @svlevine   ·   7/27/2021
Offline RL+meta-learning is a great combo: take data from prior tasks, use it to meta-train a good RL method, then quickly adapt to new tasks. But it's really hard. With SMAC, we use online self-supervised finetuning to make offline RL work: https://t.co/mtgZTivlVs A thread:
 Reply      Retweet   43      Like     162    



Ettore Randazzo    @RandazzoEttore   ·   9/14/2021
Very cool work! It's great to see that having compact and powerful models such as NCA can allow for quality diversity algorithms to be used efficiently.
 Reply      Retweet   2      Like     1    



TensorFlow    @TensorFlow   ·   6/30/2021
🏆⚡️ The latest MLPerf Benchmark results are out and Google's #TPU v4 has set new performance records! Now you can train some of the most common ML models in seconds. Learn more https://t.co/IZGQh2p2pY
 Reply      Retweet   42      Like     175    



Leo Boytsov    @srchvrs   ·   9/6/2021
1. Without finetuning Google LM has very low zero shot performance in MT tasks, in fact, much lower than OpenAI claimed foe their GPT3 model. 2. It's low but not zero esp De-En has unbelievably high BLEU 20. 3. Finetuned model codenamed FLAN unsurprisingly performs much better
 Reply      Retweet   1      Like     3    



Pin-Yu Chen    @pinyuchenTW   ·   6/21/2021
Join our live session #CVPR2021 @CVPRConf to know more about how data poisoning affects the certification performance of randomized smoothing for non-robust and robust models at test time!
 Reply      Retweet   1      Like     17    



Max Jaderberg    @maxjaderberg   ·   7/15/2021
My team at @DeepMind is now hiring, and we are looking for some amazing people who want to join us to build the agents and learning algorithms of the future! https://t.co/onDPjjJekT 1/n
 Reply      Retweet   60      Like     299    














 
  Relevant People  

Animesh Garg
Machine Learning for Robotics. Assistant Professor in AI @UofTCompSci & @VectorInst. Also at @UofTRobotics & @NvidiaAI. Here via @StanfordAILab, @berkeley_ai
Animesh Garg 36.2

Nils Reimers
NLP researcher at @huggingface • Creator of SBERT (https://t.co/MKKOMfuQ4C)
Nils Reimers 22.3

DeepMind
Our team research and build safe AI systems. We're committed to solving intelligence, to advance science and humanity.
DeepMind 60.0

Sergey Levine
Associate Professor at UC Berkeley
Sergey Levine 43.3

Chelsea Finn
CS Faculty @Stanford. Research scientist @GoogleAI. PhD from @Berkeley_EECS, EECS BS from @MIT #BlackLivesMatter
Chelsea Finn 45.1

TensorFlow
TensorFlow is a fast, flexible, and scalable open-source machine learning library for research and production.
TensorFlow 56.5

Stanford HAI
Advancing AI research, education, policy, and practice to improve the human condition.
Stanford HAI 47.5




  Hashtags  


   #TPU

   #CVPR2021