Jonas Verhellen @jverhell@qoto.org

📊 Excited to announce that "Bayesian Illumination: Inference and Quality-Diversity Accelerate Generative Molecular Models" is now available! Key takeaway: Bayesian Illumination is 100x more effective than either genetic algorithms or deep generative models.

📄 Preprint: https://doi.org/10.26434/chemrxiv-2024-tqf0x
💻 GitHub: https://github.com/Jonas-Verhellen/Bayesian-Illumination
📰 Docs: https://jonas-verhellen.github.io/Bayesian-Illumination/

Bayesian Illumination has been accepted at the ICML ML4LMS workshop! More updates to come!

Simple but magnificent 🤤. A little bit of Rudolph on my plate for Christmas dinner. 🦌❄️

🇧🇪 #kerstavond 🇳🇴 #julaften

To conclude my visit to #Trondheim, I got to see a performance by a local artist: #Aurora #Borealis. Not to be confused with that other Aurora who played the day before.

Jumping into a fjord at -4°C ... ✅

Visiting #Trondheim and thie as high as the sun gets this time of year.

#norge #norway #winter

Had a great time at the Nordic #AI Meet, presenting our work on distilling #neuroscience into neural networks. #computerscience #norway #conference

#RemembranceDay

Now that I have the opportunity, let me #introduce our quality-diversity #compchem algorithm to Mastadon. It's based on @janhjensen 's generative algorithm. We avoid stagnation issues, and illuminates opportunities across chemical space, and outperforms #deeplearning approaches!

Code: https://github.com/Jonas-Verhellen/Argenomic

Paper: https://pubs.rsc.org/en/content/articlelanding/2020/sc/d0sc03544k

#drugdesign #optimisation #algorithm #chemistry #generative