- What does a British ML researcher say when they see you initialize all of your network weights to 0? - This is not correct innit

If you want to hear me ramble for an hour about how ML can be used to deal with chemical reactions, this is it. Thanks MLSS for the invite ❤️

Oh wow. I know this isn't too far from your standard corporate-email-to-bullet-points kind of thing, yet when done directly from a 1h+ video it still feels a bit like magic ✨️

We've built generative models of drug-like molecules in the past, now it's time for materials. Very exciting work from our #AI4Science team Microsoft Research 🚀

If you're at NeurIPS and interested in retrosynthesis, then come talk to our ex-intern Austin!

We've been secretly developing syntheseus (our synthesis planning library) for a while now, currently ironing out the remaining quirks and preparing docs/tutorials. If you want a sneak peek into what we've been up to, chat to Austin at NeurIPS or visit github.com/microsoft/synt… ⚗️

After several rounds of super-successful AI4Science internships, here comes another one. Apply and come work with us on cool things 🚀

I'm at #ICLR2024; let me know if you want to chat! Also, drop into the MSR booth on Wednesday & Friday for talks and AMAs around #AI4Science ⚗️

reviewers: you need more math in your paper me: say no more me: proceeds to take 7 lines to prove that if something happens with probability p, then it doesn't happen x times with probability (1-p)^x

Why do people working on drug discovery in the US find it hard to collaborate with people in Europe? Because Americans represent the molecules using the SMILES notation while Europeans prefer SKILOMETERS 🙃

Happy to see more people building on top of our syntheseus framework! pubs.acs.org/doi/10.1021/ac…