ML4Sci #35: Announcing the new ML4Sci Slack Workspace

Also, recommendation systems for nanoporous materials and AI for semiconductor chips

Hi, I’m Charles Yang and I’m sharing (roughly) monthly issues about applications of artificial intelligence and machine learning to problems of interest for scientists and engineers.

If you enjoy reading ML4Sci, send us a ❤️. Or forward it to someone who you think might enjoy it!

Welcome back to ML4Sci and apologies for landing in the inbox on a Tuesday.

First, some housekeeping. I’m excited to announce that by popular request, we’re starting a slack workspace to help build out the ML4Sci community! I realize that newsletters are fairly static, one-to-many networks and I’m hoping that this slack workspace can help facilitate more discussion and serendipitous introductions. You can click here to join and introduce yourself in the #introductions channel.

I also realized that my recent newsletters have been focusing quite heavily on AI-specific topics while neglecting papers that focus on the intersection of AI and sciences. I’ll be trying to realign focus again which also means that I’ll be switching to a monthly schedule to for these newsletters (reading papers is hard!). But don’t worry - you can always join the ML4Sci workspace and keep the conversations going!

And now, for the news.

Department of Machine Learning

Are benchmarks all you need? Apparently not. CNN architectures that perform better on ImageNet do not necessarily perform better on other datasets. This just reinforces the point that I’ve been making for awhile now: we need domain-specific architectures in ML4Sci, which means we will need domain experts who understand the data and can design novel AI models.

Near-Future Science

🧪[ACS] “The lab of the future is now” aka “self-driving labs”. Nice to see that many of the articles we’ve covered at ML4Sci appear in this article!

❤️[BBC] The NHS approves an AI algorithm for CAT scans after heart attacks

Manufacturing ever-more advanced semiconductor chips is becoming the new “Space Age” competiton - the technological benchmark by which nation-states (US, China, EU) jostle for dominance. DeepForestSci had some coverage on how AI is helping advance this field

One problem AI still can’t solve: math proofs. GPT-2 and GPT-3 never break 10% accuracy on this math theorem proving dataset

🌠 “A Living Review of ML for Particle Physics”. Last week, I highlighted the Living Journal of Computational Molecular Science. Software is allowing us to have living, interactive scientific repositories of knowledge!

[Chemrxiv] Recommendation engines for nanoporous materials. The spurt of growth in recommendation systems, powering Netflix, Airbnb, and more, is also helping scientists impute missing properties of materials

💼 always wanted to work at Facebook but don’t know how to code? Facebook is now hiring MOF-specialists… (incidentally, MOF’s are another material class that is seeing huge amounts of AI research due to their large design space)

The Science of Science

“Science as craft industry” by Freeman Dyson from 1998. A remarkably prescient essay that seems particularly relevant as AI-whisperers are really more craftsmen/women and less scientists per se

📖The University of California and Elsevier reach open-access publishing deal. Another slow step towards changing the way we publish science

Protecting Privacy gone wrong: the authors behind ImageNet released a new version of the dataset with the faces in the dataset obscured. But one researcher details how they systematically excluded their work from the cited literature in the paper. Power imbalance, hierarchy, and exculsion in citation

A twitter thread outlining why virtual research seminars should stay:

🌎Out in the World of Tech

Microsoft plans to role out (AI-driven) text predictions for Word

[TheVerge] Check out this crazy AI-generated deepfake video of Tom Cruise thats circulating on TikTok (and why making these kinds of deepfakes is still really hard)

[Wired] Replacing the Simpson’s voice actors with deepfakes. Is this the automation everyone keeps telling us to be worried about? On a more serious note, there are some interesting questions about copyright - who owns your voice? Is it copyright infringement when TikTok users do it and not when your studio does it?

Thanks for Reading!

I hope you’re as excited as I am about the future of machine learning for solving exciting problems in science. You can find the archive of all past issues here and click here to subscribe to the newsletter.

Have any questions, feedback, or suggestions for articles? Contact me at or on Twitter @charlesxjyang