@felix-s-k
Assistant Professor of Digital Chemistry (@uni-wuppertal.bsky.social) – building digital tools for synthesis and catalysis. Former: U Toronto (Postdoc) | U Münster (PhD) | MIT (Visiting Researcher) https://fsk-lab.github.io
More news from our group – we have an open PhD position in computational catalyst design! If you're interested in applying data science to real-world challenges from the lab, and want to be part of an interdisciplinary team, we'd be happy to hear from you.
Check fsk-lab.github.io for details!
BoTier can be used as a plugin for BoTorch – and is available from `pip`.
pypi.org/project/boti...
It's a great feeling to see our first paper out @digital-discovery.rsc.org! This wasn't planned – but necessity was the mother of invention: We needed BayesOpt for a hierarchy of objectives, some of which are experiment inputs.
So here's BoTier – open-source & ready to use! doi.org/10.1039/D5DD...
Open PhD position in computational materials chemistry
As part of my Villum Young Investigator project "Entropy in materials design: Accelerated discovery of disordered solid electrolytes", I am looking for PhD student to join my group.
Please share with anyone interested!
#compchem #chemsky
More BayesOpt work out on arXiv!
What we chemists call "finding general conditions" is rather tricky for BO. We try to benchmark how it's done best. TLDR: Exploration of both parameter and task space is key. Check the thread by the one and only @spschmid.bsky.social – thanks to him and the team! 🍛
🚀 Looking for reaction conditions that work well for multiple substrates? CurryBO can help🍛
Now out on arXiv: arxiv.org/abs/2502.18966
A short explanation thread 👇
Come to #accelerate25 the premier conference on #aiformaterials #selfdrivinglabs #chemsky Please RT
link 📈🤖
BoTier: Multi-Objective Bayesian Optimization with Tiered Composite Objectives (Haddadnia, Grashoff, Strieth-Kalthoff) Scientific optimization problems are usually concerned with balancing multiple competing objectives, which come as preferences over both the outcomes of an experiment (e.
🔄 Updated Arxiv Paper
Title: BoTier: Multi-Objective Bayesian Optimization with Tiered Composite Objectives
Authors: Mohammad Haddadnia, Leonie Grashoff, Felix Strieth-Kalthoff
Read more: https://arxiv.org/abs/2501.15554
Lastly, a huge thanks to Mohammad and Leonie for making it happen!
The code is available at github.com/fsk-lab/botier, and it can be directly installed from #pip. pypi.org/project/boti... . Feel free to give it a try, and leave any feedback!
...and their hierarchy in a modular fashion – and has helped us in practical optimization problems already. It's fully auto-differentiable, and can be used as a composite objective with #botorch, which gives a big boost in optimization efficiency (check the paper for systematic benchmarks).
In the lab, we often face multi-objective problems, where objectives come as preferences over both the outcomes of an experiment (e.g. maximize the yield of a reaction) and experiment inputs (e.g. minimize the load of an expensive catalyst). BoTier helps you to express these preferences ...
My first post here on #bsky – what better occasion than our group's first preprint? BoTier (Multiobjective Optimization with Multiple Tiers of Objectives) is now on arXiv. This project was actually born out of necessity – more details in the 🧵 below. Please check it out!
arxiv.org/abs/2501.15554