Newest papers

Predicting Unseen Scientific Interactions: Conditional Flow Models Hallucinate Non-Additive Effects

Sam Christian Jan Gijsen, Moritz Vinzent Seiler, Kerstin Ritter · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.20288 · cited by 0 works (OpenAlex) · open-access PDF

French

Les modèles génératifs conditionnels sont de plus en plus utilisés pour la prédiction scientifique au-delà du soutien de la formation directe. Cependant, les questions scientifiques concernent souvent si une intervention change l'effet d'une autre. De telles interactions peuvent être d'intérêt primordial tout en expliquant peu de la variation totale, permettant aux modèles d'apparaître précis dans l'ensemble sans récupérer l'effet scientifique pertinent. Il reste incertain de savoir comment ces modèles apprennent les interactions hallucinées et quand ceux-ci extrapolent avec succès à des conditions invisibles. Nous étudions donc la récupération des interactions dans des modèles ciblés en utilisant un simulateur de série de temps physiologique, des données d'interaction de dose de médicament, des perturb

Machine translation (Meta M2M100); the English original follows.

Abstract

Conditional generative models are increasingly used for scientific prediction beyond the direct training support. Yet scientific questions often concern whether one intervention changes the effect of another. Such interactions can be of primary interest while explaining little of the total variation, allowing models to appear accurate overall without recovering the relevant scientific effect. It remains unclear how these models learn interactions and when these successfully extrapolate to unseen conditions. We therefore study interaction recovery in flow-matching models using a physiological time-series simulator, drug dose-response data, combinatorial cell perturbations, and molecular conformer generations, where we separate recovery from hallucinated interaction structure. In the controlled setting, individual effects are learned much earlier than interactions and models hallucinate interactions where none exist. Across settings, we find qualitatively different generalization problems, with successful prediction depending on whether condition structure continues from observed to held-out targets. When it does, interactions are recoverable. If not, models can generate plausible responses while missing the interaction. We furthermore find that hallucinations increase as targets become less identifiable, and that successful interaction recovery in output space can fail under nonlinear readouts. Finally, we show that aggregate fidelity metrics can conceal whether models learn and extrapolate interactions.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Hallucinating, Generalization, Artificial intelligence, Overfitting, Computer science, Machine learning

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange