Publications

All Publications

My research interests are centered on advancing diffusion-based generative models and multimodal information theory. I aim to refine diffusion models for generating high-quality outputs across different modalities, tackling the challenge of any-to-any generation to enable seamless cross-modal translations. Concurrently, my work explores multimodal information theory to better understand and quantify the complex interplay between various data types.

For the full publication list, please refer to:

Selected Publications

[1]
Bounoua, M., Franzese, G. and Michiardi, P., S\(\Omega\)i: Score-based o-INFORMATION estimation.
[2]
Franzese, G., Bounoua, M. and Michiardi, P., MINDE: Mutual information neural diffusion estimation. arXiv preprint arXiv:2310.09031. (2023).
[3]
Bounoua, M., Franzese, G. and Michiardi, P., Multi-modal latent diffusion. arXiv preprint arXiv:2306.04445. (2023).
[4]
Tran, B.-H., Franzese, G., Michiardi, P. and Filippone, M., One-line-of-code data mollification improves optimization of likelihood-based generative models. to appear in NeurIPS. (2023).
[5]
Franzese, G., Corallo, G., Rossi, S., Heinonen, M., Filippone, M. and Michiardi, P., Continuous-time functional diffusion processes. to appear in NeurIPS. (2023).
[6]
Franzese, G., Rossi, S., Yang, L., Finamore, A., Rossi, D., Filippone, M. and Michiardi, P., How much is enough? A study on diffusion times in score-based generative models. Entropy. 25, 4 (2023), 633.
[7]
Franzese, G., Milios, D., Filippone, M. and Michiardi, P., Revisiting the effects of stochasticity for hamiltonian samplers. ICML (2022), 6744–6778.