DelocalizedDanny

Our recent work on #MachineLearning on small dataset with @BernaertsGroup is published on the cover of #PolymerInternational of @WileyPolymers. Had as much fun creating this, as the paper. #AI #artificialintelligence #ML #materiomics #ScienceAsArt @uhasselt @imoimomec

📜alert: We are excited to share our latest research on SnV color centers in #diamond 💎. We explore the impact of comp.methods on cost an accuracy, hoping to contribute to a route to #frugalcomputing. We identify the SnV0 ZPL at 663nm. #QuATOMs #DFT sciencedirect.com/science/articl…

@burkov Seems they missed modelling small data( <50 samples) though I've doing this with real experimental data since 2020. 🤔 And inference models are even cheaper.😇 pubs.aip.org/aip/jap/articl…

This paper really is groundbreaking. It solves a long-standing embarrassment in machine learning: despite all the hype around deep learning, traditional tree-based methods (XGBoost, CatBoost, random forests, etc) have dominated tabular data—the most common data format in real-world applications—for two decades. Deep learning conquered images, text, and games, but spreadsheets remained stubbornly resistant. This paper's (published in Nature by the way) main contribution is a foundation model that finally beats tree-based methods convincingly on small-to-medium datasets, and does so very fast. TabPFN in 2.8 seconds outperforms CatBoost tuned for 4 hours—a 5,000× speedup. That's not incremental; it's a different regime entirely. The training approach is also fundamentally different. GPT trains on internet text; CLIP trains on image-caption pairs. TabPFN trains on entirely synthetic data—over 100 million artificial datasets generated from causal graphs. TabPFN generates training data by randomly constructing directed acyclic graphs where each edge applies a random transformation (using neural networks, decision trees, discretization, or noise), then pushes random noise through the root nodes and lets it propagate through the graph—the intermediate values at various nodes become features, one becomes the target, and post-processing adds realistic messiness like missing values and outliers. By training on millions of these synthetic datasets with very different structures, the model learns general prediction strategies without ever seeing real data. The inference mechanism is also unusual. Rather than finetuning or prompting, TabPFN performs both "training" and prediction in a single forward pass. You feed it your labeled training data and unlabeled test points together, and it outputs predictions immediately. There's no gradient descent at inference time—the model has learned how to learn from examples during pretraining. The architecture respects tabular structure with two-way attention (across features within a row, then across samples within a column), unlike standard transformers that treat everything as a flat sequence. So, the transformer has basically learned to do supervised learning. Talk to the paper on ChapterPal: chapterpal.com/s/a1899430/acc… Download the PDF: nature.com/articles/s4158…

Guest editing a special issue on AI in carbon materials. More info on the flyer. Show us what quality AI can be for materials research. dannyvanpoucke.be/special-issue-…

Een historisch moment, de ondertekening van een samenwerkingsovereenkomst tussen @uhasselt, @Unimaas, @OU_Nederland, @UniversiteLiege, @RWTH Trinationaal onderwijs onder de noemer ‘The art of working together’. @Rianneletschert @TheoBastiaens

🚨📜The result of a very enjoyable collaboration with Goedele Roos and Jane Murray. It's marvelous working with someone who has such a long experience with ESP. The paper provides an interesting glimpse into the history of ESP. The pdf can be dl'ed: …mistry-europe.onlinelibrary.wiley.com/doi/epdf/10.10…

Fortunately, no-one will ever see this post ;-)

1984 update to 2025? Looks like a lot of people never read this classic, though it seems to be unfortunately relevant and all to realistic. Maybe time to so now. There is more relevant dystopian fiction (minority report and Gattaca come to mind.)... goodreads.com/book/show/6143…

@Peretti_Romain is this in the realm of possibilities for you? science.org/doi/10.1126/sc…

🎉Excited to share our paper on the impact of strain on the germanium vacancy color center in #diamond! A great collaboration with the #QuATOMs group combining theory and experiments: sciencedirect.com/science/articl… @KenHaenen @PauliusPbd @DannyVanpoucke @ThijsvanWijk @RaniMaryJoy

@rmaryjoy @WBGM_UHasselt free paper access via authors.elsevier.com/a/1kPZ91zUAZr7H

Of course, calculations alone are just busywork, so we got some marvelous collaboration from @rmaryjoy from the @WBGM_UHasselt group.

📜🚨 From MSc to Paper: "The impact of strain on the GeV-color center in #diamond" We are the first to extrapolate to exp. conc. of Ge in 💎 using #DFT, highlighting the risks of single cell calculations. #QuATOMs #materiomics sciencedirect.com/science/articl…

Molecular dynamics simulations in mixed reality! With @labriataphd and @lucien_krapp we’re scaling our multi-user WebXR platform to enable immersive molecular simulations, fully based on web technologies. Stay tuned, lots of cool stuff coming soon! @threejs #WebXR #Quest3

En zo starten we het derde jaar van de master in #materiomics opleiding. #materials #physics #chemistry

A big Congratulations from the #QuATOMs group as well. You did marvelous, and put a high standard for those who will follow in your footsteps.

@j_bertolotti Infinite loop gif...only one needed. 😇

Group picture is here! #DFT2024

Big thanks to the organisers of #dft2024 for the opportunity to present our recent work on diamond. As usual it was a lot of fun presenting without a single equation 😈👨‍💻

Ilaria Ciofini @group_ctm is on the stage for the opening of #DFT2024! The detailed program is now online coms.events/DFT2024/index.…. Enjoy!