van der Schaar Lab @ AISTATS 2026

Excited to kick off the year with some great news: three papers from our group have been accepted to the 29th Annual Conference on Artificial Intelligence and Statistics (AISTATS) 2026, our first conference of the year.

AISTATS is a prestigious international conference bridging machine learning, statistics and data science. This year held in Tangier, Morocco from 2nd – 5th of May will see researchers from around the world come together to present their research and exchange ideas.

The three papers accepted this year ranges from AI and creativity to DNA sequencing, which are listed below:

1. The Reasoning-Creativity Trade-off: Toward Creativity-Driven Problem Solving
   Max Ruiz Luyten, Mihaela van der Schaar
2. Interpretable DNA Sequence Classification via Dynamic Feature Generation in Decision Trees
   Nicolas Huynh, Krzysztof Kacprzyk, Ryan M Sheridan, David L. Bentley, Mihaela van der Schaar
3. Fact-Augmented Lookahead for LLM Agents: Simple Online Memory, No Finetuning
   Samuel Holt, Max Ruiz Luyten, Thomas Pouplin, Mihaela van der Schaar

We have “The Reasoning-Creativity Trade-off: Toward Creativity-Driven Problem Solving” by Max Ruiz Luyten. This paper studies how common reasoning pipelines for large language models can unintentionally suppress diversity and creativity. It introduces a principled framework for building LLMs that remain both correct and creatively flexible.

A wonderful collaborative piece led by Nicolas “Interpretable DNA Sequence Classification via Dynamic Feature Generation in Decision Trees”. This work presents a new approach to DNA sequence modelling that dynamically constructs human-interpretable features within decision trees, enabling strong predictive performance across a diverse range of genomic tasks.

Finally, “Fact-Augmented Lookahead for LLM Agents: Simple Online Memory, No Finetuning” by our wonderful alumnus Samuel Holt now working at DeepMind and students Max and Thomas Pouplin. This paper proposes an LLM agent framework that learns from atomic facts extracted during interaction and uses them for lookahead planning, improving decision-making and yielding higher-quality solutions.

Congratulations to all co-authors and collaborators – we look forward to sharing more details on each paper in the coming weeks.