Edoardo Ponti

Fast and Expressive Multi-Token Prediction with Probabilistic Circuits

Sample-efficient Integration of New Modalities into Large Language Models

Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling

Bootstrapping World Models from Dynamics Models in Multimodal Foundation Models

Neurosymbolic Diffusion Models

Training Plug-and-Play Knowledge Modules with Deep Context Distillation

Post-hoc Reward Calibration: A Case Study on Length Bias

Universal Cross-Tokenizer Distillation via Approximate Likelihood Matching

Inference-Time Hyper-Scaling with KV Cache Compression