Тяжела ты, шапка Мономаха...

!pip install

After talking to several of my senior colleagues in academia, I somehow remember the most famous tanka by Ono no Komachi....

AI comedy night - exploring career pivot to stand up comedy www.youtube.com/watch?v=jNP7...

🙏 Off to the APS March meeting to give Adler Prize lecture on closing the characterization loop via reward driven workflows. DM if you would like to meet and grab a coffee! What makes this conference special is that it is also the first talk of Dmitri Kalinin!

My belief in humanity is restored

Automated STEM in action. So many particles, so little time... But now AE STEM can do it www.linkedin.com/posts/sergei...

Pro tip - if you have to wake up at night a lot to take care of baby, luminescent tape on all edges minimizes probability of toe damage

Newcomer to automated experiment world - ML-enabled nanoindentation www.linkedin.com/posts/activi...

Ferrofluid and Frank Sinatra. It doesn't get better.

A question to my colleagues in computer vision - what is the best way to identify objects and data variability in Esher drawings?

Dune Prophecy made Dune very small

Python is successfully installed, but now no mouse is detected....

Russo - Iranian XMas. Vodka is from Texas.

شب یالدا

Welcome to Knoxville, Vasily Kalinin!

Asking for advice from molecular discovery and ML community re practical databases - will appreciate leads! www.linkedin.com/posts/sergei...

Showing the automated labs in University of Tennessee Tickle College of Engineering to Steven Spurgeon . The Spectra 300 with HPC control, Jupiter AFM, and the fleet of pipetting robots and SpinBot at Mahshid Ahmadi lab. It works, and it works together!

Machine learning and AI can help only those who know how to help themselves

Hackaton launch! www.linkedin.com/posts/sergei...

SVK -> Fellow, National Academy of Inventors research.utk.edu/oried/2024/1...

Reinforcement learning is not the right algorithm for the cartpole problem. LLMs are not the right algorithm for math or causal inference. It's amazing to see the ML community trying to use fantastic tools for tasks they are not built for.

With a defined reward function, everything is an optimization problem. And we know how to solve optimization problems.

If you plan to attend MRS in Boston - will be happy to connect! www.linkedin.com/posts/sergei...

Intro to Machine Learning for Materials Science: from ontologies and NLP to causality and decision making www.linkedin.com/posts/sergei...

Sproggy guards the turkey

How important is quantum mechanics for materials science? www.linkedin.com/posts/sergei...

So Microsoft now can datamine all Word and PowerPoint docs as we type them unless opt out? www.linkedin.com/posts/sergei...

It's a very special feeling when you are 5 minutes over time in class, and the students look at you expecting the lecture to continue.

AI/ML for scanning probe microscope - use digital twin to transition from in silico to in materio! www.linkedin.com/posts/sergei...

For the last 20 years, we were scaling theory. For the last 5, synthesis. It's time for characterization and feedback to theory. Which also means that we need light models that can be updated based on individual experiments, and ways to integrate them into more universal ones.

And reward based workflows for image analysis: unsupervised, explainable, robust, and allowing for human in the loop tuning. arxiv.org/abs/2411.12612

Reward based microscopy optimization - how to make things work better when there is no training data, but we can build human heuristics or physics based reward function arxiv.org/abs/2411.12065

Join the microscopy hackaton (and see the actual video of the hacker infiltrating the microscope room) www.linkedin.com/posts/sergei...

I know everyone else noticed too, but I'm still amazed, happy and unexpectedly distracted by the old chemtwitter suddenly re-forming on chemsky. Made my week so much better and I hope it'll make everyone else's life better, too. Great to see everyone here.