Branislav Kveton

Learning user preferences for thousands of items can be achieved with just a handful of evaluations, thanks to a novel approach that leverages effective dimension in graph-based bandit problems.

Correcting errors early in the diffusion process matters more than fixing them later: Stepwise-Flow-GRPO leverages this insight to dramatically improve RL-based flow model training.

Eye-tracking data can boost click prediction in carousel interfaces, but surprisingly, better click prediction doesn't always mean a better model of user behavior.

Stop wasting compute on LLM evals: a variance-adaptive querying strategy slashes estimation error by focusing on the most uncertain prompt-response pairs.