Stefan

Data scientist working on RL for protein optimization. Building my next startup. Interest in economics, politics, culture. Loves biking, coding, travelling.🇪🇺

[Stefan @Stefan_Vrest]

@clmt works great. thanks!
to be honest the Gemini/AI API documentation is a bit inconsistent and confusing

[Stefan @Stefan_Vrest]

@thegrahamcooke nah

[Stefan @Stefan_Vrest]

Sad to see how @Google avoids tax. Taxation is a system to achieve equality and fairness in our societies. The fact that @Google does not want to participate in this, gives a bad taste to their great products and @DeepMind‘s excellent research.
@googleeurope @GoogleAI @GoogleUK

[Stefan @Stefan_Vrest]

DALLE creates images from text captions. https://t.co/m74cxfkGwJ

[Sergey Levine @svlevine]

New lectures released for the deep RL class: special lecture on open topics in algorithms by Aviral Kumer (#17), variational inference (#18), control as inference (#19), and inverse RL (#20):

https://t.co/Y674PBH6TS

[Sergey Levine @svlevine]

New deep RL course lectures!
https://t.co/Y674PBH6TS

Lecture 13+14: exploration and unsupervised RL
Lecture 15+16: offline RL and RL theory, by Aviral Kumar!
Will try to update each week in the future as new lectures come out.

[PyTorch @PyTorch]

PreferredNet is moving ChainerRL to the PyTorch ecosystem. With a heavy emphasis on reproducibility, PFRL users have the ability to choose amongst the library's several algorithms and combine them with a multitude of features to produce novel RL systems. https://t.co/FcZCg72Hki

[Stefan @Stefan_Vrest]

How Lyft built their ML framework. Both researchers and engineers can co-develop using the same framework, which allows to create an iteration cycle from idea-to-production at a rapid pace.
#AI #DataScience #ML https://t.co/7qqp4KrTcP

[Google AI @GoogleAI]

Announcing Menger, a massively large-scale distributed #ReinforcementLearning infrastructure that scales to several thousands of actors, which we show can achieve up to 8.6x reduction in training time on chip placement tasks. Learn more below: https://t.co/vBa6ziCHsH

[Ming-Yu Liu @liu_mingyu]

Introducing #Imaginaire
a #PyTorch library with optimized implementations of several #GAN image and video synthesis methods developed at #NVIDIA
code https://t.co/fWMHEwhS3q
video https://t.co/BUkEE7RB3D
By @liu_mingyu @tcwang0509 @arunmallya @xunhuang1995

[Stefan @Stefan_Vrest]

About protein design, the folding-problem and the role of machine learning
#DataScience #proteomics #AI #proteindesign https://t.co/qi0LcKwRZk

[Stefan @Stefan_Vrest]

How Offline Reinforcement Learning Will Change How We Use Machine Learning https://t.co/MJc82VvRJr

[Yujia Li @liyuajia]

Nice blog post explaining the work done at DeepMind in collaboration with Google using graph neural networks to improve travel time estimates in Google Maps. GNNs are already serving queries in many cities around the world :)

[Hiroshi Sasaki @HiroshiSasak1]

DNA nanoscope: Microscope-free "imaging" method for complex nanostructures by leveraging DNA molecular ruler and long-read @nanopore sequencing. Congrats to Nikhil and co-authors!
https://t.co/BvhzNGMuY1

[Microsoft Research @MSFTResearch]

Microsoft researchers propose a new natural language processing paradigm—one in which models are pretrained from scratch entirely within a specialized domain. Learn about the new BLURB benchmark, leaderboard, and state-of-the-art model for biomedical NLP: https://t.co/xHX7Z26pSa

[Kristian Kersting @kerstingAIML]

Have some exciting news about #AI research in 🇩🇪. Today the Hessian Center For AI has been announced. Great team: @stefanroth, @Jan_R_Peters, @ConstantinRoth1, @cbinnig & many more. 22 founding members & 20 new #AI profs. Thrilled to co-direct it together with the great @m_mezini

[François Chollet @fchollet]

I think it's clear that for many smaller companies that invested in deep learning, it turned out not to be essential and got cut post-Covid as part of downsizings. There are somewhat fewer people doing deep learning now than half a year ago, for the first time since at least 2010

[lewis martin @lewischewis]

love this idea - quickly visualize PDB files directly in the terminal. As an example this is 3JAE, a glycine receptor. https://t.co/6FOTJuv551

[Soumith Chintala @soumithchintala]

really cool robot from a new startup HelloRobot.

A payload of 1.5kg, the simple and effective arm, the funky manipulator and the lightweight nature, all very interesting!
$18k for now, so still out of reach of most folks, but cheaper than the $200k ones

https://t.co/oBoocIpifq

[Stefan @Stefan_Vrest]

Transformers for protein folding.
Great to see that transformers (finally) made it to this exciting application! @salesforce
#AI #ML #NLP #Folding

[Richard Socher @RichardSocher]

Fascinating insight: Trained solely on unsupervised language modeling, the Transformer's attention mechanism recovers high-level structural (folding) and functional properties of proteins!
Blog: https://t.co/ClWXxxNbpW
Paper: https://t.co/0oK7coEMV6
Code: https://t.co/3z3CVQ51kS