xLSTM for stock market prediction: arxiv.org/abs/2408.12408 "xLSTM-TS model consistently outperformed other state-of-the-art models, demonstrating superior predictive accuracy across multiple datasets...highlighting its robustness and ability to handle...financial time series."

Diverse Hits in De Novo Molecule Design: Diversity-Based Comparison of Goal-Directed Generators #DrugDesign pubs.acs.org/doi/10.1021/ac… Sohvi Luukkonen Günter Klambauer Computational Drug Discovery Leiden LIT AI Lab & ELLIS Unit Linz LeidenScience ENG #JCIM Vol64 Issue15 #Letter

Machine Learning Potentials: A Roadmap Toward Next-Generation Biomolecular Simulations Neural network potentials are very powerful, but they appear too slow to be ever used for macromolecules.. this work provides a possible roadmap... P: arxiv.org/abs/2408.12625

Jetzt ist schon wieder was passiert: unser Art&Science Projekt **BruQner -- the sound of Entanglement** wird ein opening act beim Ars Electronica Festival nächsten Mittwoch (4.9), 22:00 im Linzer Dom. Eintritt ist frei! This should be fun! youtu.be/m5-0rCh3GbY?si…

Cross-Modal Learning for Chemistry Property Prediction LLMs and GNNs work together to allow for zero- and few-shot property prediction. Very similar and already shown in CLAMP (arxiv.org/abs/2303.03363).. still impressive on QM9. P: arxiv.org/abs/2408.14964

This is crazy...

ACEGEN: Reinforcement Learning of Generative Chemical Agents for Drug Discovery #DrugDiscovery pubs.acs.org/doi/10.1021/ac… Carles Navarro Maciej Majewski Gary Tresadern vmoens Gianni De Fabritiis Acellera Therapeutics #JCIM Vol64 Issue15 #MachineLearning #DeepLearning

Mamba or Transformer for Time Series Forecasting? Mixture of Universals (MoU) Is All You Need Uses patches and "feature extraction" at different levels. Combines Mamba and transformer.. .. I think that is not all we need ;) P: arxiv.org/abs/2408.15997

Anthropic CEO Dario Amodei says AI could increase the rate of discovery in biology by 100x, compressing the total amount of progress that occurred in the 20th and 21st centuries to just a few years

A bioactivity foundation model using pairwise meta-learning Siamese networks used again for few-shot learning (we have better models by now, but ok). They predict the activity difference between compounds. Remarkable again: Fingerprints+MLP encoder P: nature.com/articles/s4225…

xLSTM to detect ASD: arxiv.org/abs/2408.16924 "2sGCN-AxLSTM demonstrates an unprecedented accuracy of 89.6% in early detection of ASD." 2sGCN-AxLSTM "outperforms conventional GCN-based and CNN-based methods, as well as the traditional LSTM-based method"

Meta Flow Matching: Integrating Vector Fields on the Wasserstein Manifold Flow matching extended to population-level predictions. Uses KNN graphs and GNNs to embed instances of the population. P: arxiv.org/abs/2408.14608

Towards Symbolic XAI -- Explanation Through Human Understandable Logical Relationships Between Features Abstractions on top of traditional XAI methods are used. These are combined with logical operators to provide explanations... P: arxiv.org/abs/2408.17198

I've heard rumors that ELLIS ML4Molecules workshop will be back in December! Stay tuned for more updates! Here is a list of previous workshops: moleculediscovery.github.io/workshop2023/ moleculediscovery.github.io/workshop2022/ moleculediscovery.github.io/workshop2021/

Ein österreichisches Forschungsprojekt will am Vorbild des menschlichen Gehirns eine #KI entwickeln, die Aussagen auch auf Wahrheit, Schlüssigkeit und Logik prüfen kann. #Exzellenzcluster #FWF-funded JKU - Johannes Kepler Universität Linz TU Graz TU Wien via orf_at #WissenAktuell ➡️ science.orf.at/stories/322658…

If our model is correct, we are in for a lot of high flows for the next days... Also surreal to see flood warnings from our own model on Google Maps for the first time in my own region. #floodalert #Vienna #AustriaGo

Reminder that Google Floodforecasting and Floodhub are based on the famous LSTM networks! sites.research.google/floodforecasti… To all the people living in central Europe now affected by floodings.. stay safe!!

HOW MOLECULES IMPACT CELLS: Unlocking Contrastive PhenoMolecular Retrieval Valence Labs builds on our CLOOME approach ( nature.com/articles/s4146… ) Instead of co-training encoders, they pre-train each in uni-modal way. P: arxiv.org/abs/2409.08302