← Search

Fiorenzo Parascandolo

2 accepted papers

2026

Restoring Exploration after Post-Training: Latent Exploration Decoding for Large Reasoning Models

ICML 2026poster

Large Reasoning Models (LRMs) have recently achieved strong mathematical and code reasoning performance through Reinforcement Learning (RL) post-training. However, we show that modern reasoning post-training induces an unintended exploration collapse: temperature-based sampling no longer increases p…

Cited by 0SourceScholar
2025

Causal Graphical Models for Vision-Language Compositional Understanding

ICLR 2025poster

Recent work has empirically shown that Vision-Language Models (VLMs) struggle to fully understand the compositional properties of the human language, usually modeling an image caption as a “bag of words”. As a result, they perform poorly on compositional tasks, which require a deeper understanding o…