Nicola Novello

I am currently a fourth-year PhD student at University of Klagenfurt, Austria, under the supervision of Professor Andrea M. Tonello and Professor Deniz Gündüz. Previously, I received my MSc and BSc degrees with honors from the University of Padova, Italy. I am interested in the design of objective functions for discriminative and generative models, with a particular focus on Trustworthy ML.

Machine Unlearning

Designing novel objective functions and frameworks to effectively erase concepts from diffusion models.

f-DMU SPACE

Information-Theoretic Deep Learning

Designing objective functions to estimate and maximize mutual information for ML and communication engineering.

f-DIME

Robust Classification

Exploring advanced objective functions (using f-Divergence) to enhance classification with and without label noise.

f-PML f-PML (LN)

Statistical Dependency Modeling

Developing sophisticated neural estimators to capture complex dependency structures and multivariate continuous distributions.

CODINE

news

May 13, 2026	I have been recognized as a Gold Reviewer at ICML 2026!
May 01, 2026	“A Unified Framework for Diffusion Model Unlearning with f-Divergence” accepted at ICML 2026!
Jul 22, 2025	“Copula density neural estimation” accepted at IEEE TNNLS!
Jun 01, 2025	I have started my 2 months visiting period at Princeton University!

selected publications

ICML

A Unified Framework for Diffusion Model Unlearning with f-Divergence

Nicola Novello, Federico Fontana, Luigi Cinque, Deniz Gunduz, and Andrea M Tonello

In International Conference on Machine Learning, 2026

Abs PDF Code

This paper presents the first f-divergence-based framework for DM unlearning (f-DMU). We present two classes of loss functions and we theoretically study their convergence properties. We show how H-DMU (derived from the squared Hellinger distance) leads to improved prior preservation.
NeurIPS

Mutual Information Estimation via f -Divergence and Data Derangements

Nunzio A Letizia, Nicola Novello, and Andrea M Tonello

Advances in Neural Information Processing Systems, 2024

Abs PDF Code

This paper presents an f-divergence-based class of low-variance mutual information neural estimators. Furthermore, it shows how random permutation should be replaced by a data derangement operation, to avoid upper-bounded estimates.
ICML

f -Divergence Based Classification: Beyond the Use of Cross-Entropy

Nicola Novello and Andrea M Tonello

In International Conference on Machine Learning, 2024

Abs PDF Code

This paper presents an f-divergence-based class of objective functions for the estimation of p_X|Y. Then, it shows its applicability for both when X is continuous and when X is discrete (classification).
IEEE TNNLS

Copula density neural estimation

Nunzio A Letizia, Nicola Novello, and Andrea M Tonello

IEEE Transactions on Neural Networks and Learning Systems, 2025

Abs PDF Code

This paper presents an f-divergence-based neural copula estimator, that outperforms other sota neural copula estimators.