publications

2025

1. arXiv

   

   How compositional generalization and creativity improve as diffusion models are trained

   Alessandro Favero^*, Antonio Sclocchi^*, Francesco Cagnetta, Pascal Frossard, and 1 more author

   arXiv preprint, 2025

   PDF
2. ICLR

   

   Unraveling the latent hierarchical structure of language and images via diffusion models

   Antonio Sclocchi^*, Alessandro Favero^*, Noam Levi^*, and Matthieu Wyart

   International Conference on Learning Representations, 2025

   PDF
3. ICLR

   

   LiNeS: Post-training layer scaling prevents forgetting and enhances model merging

   Ke Wang, Nikolaos Dimitriadis, Alessandro Favero, Guillermo Ortiz-Jimenez, and 2 more authors

   International Coference on Learning Representations, 2025

   PDF
4. PNAS

   

   A phase transition in diffusion models reveals the hierarchical nature of data

   Antonio Sclocchi, Alessandro Favero, and Matthieu Wyart

   Proceedings of the National Academy of Sciences, 2025

   PDF

2024

1. CVPR

   

   Multi-modal hallucination control by visual information grounding

   Alessandro Favero, Luca Zancato, Matthew Trager, Siddharth Choudhary, and 4 more authors

   IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024

   PDF
2. ICML

   

   Task addition and weight disentanglement in closed-vocabulary models

   Adam Hazimeh^*, Alessandro Favero^*, and Pascal Frossard

   ICML 2024 Efficient Systems for Foundation Models Workshop, 2024

   PDF
3. PhysRevX

   

   How deep neural networks learn compositional data: The Random Hierarchy Model

   Francesco Cagnetta, Leonardo Petrini, Umberto Tomasini, Alessandro Favero, and 1 more author

   Physical Review X, 2024

   PDF
4. JSTAT

   

   Computational complexity of deep learning: fundamental limitations and empirical phenomena

   Boaz Barak, Annabelle Carrell, Alessandro Favero, Weiyu Li, and 2 more authors

   Journal of Statistical Mechanics: Theory and Experiment, 2024

   PDF

2023

1. NeurIPS

   

   Task arithmetic in the tangent space: Improved editing of pre-trained models

   Guillermo Ortiz-Jimenez^*, Alessandro Favero^*, and Pascal Frossard

   Advances in Neural Information Processing Systems, Oral (top 0.54%) , 2023

   PDF Video
2. ICML

   

   What can be learnt with wide convolutional neural networks?

   Francesco Cagnetta^*, Alessandro Favero^*, and Matthieu Wyart

   International Conference on Machine Learning, PMLR, 2023

   PDF

2021

1. NeurIPS

   

   Locality defeats the curse of dimensionality in convolutional teacher-student scenarios

   Alessandro Favero^*, Francesco Cagnetta^*, and Matthieu Wyart

   Advances in Neural Information Processing Systems, 2021

   PDF
2. NeurIPS

   

   Relative stability toward diffeomorphisms indicates performance in deep nets

   Leonardo Petrini, Alessandro Favero, Mario Geiger, and Matthieu Wyart

   Advances in Neural Information Processing Systems, 2021

   PDF

2020

1. Thesis

   Spectral analysis of infinitely wide convolutional neural networks

   Alessandro Favero

   Master’s Thesis, Sorbonne Université and Politecnico di Torino, 2020

   PDF