CV

PhD researcher, advised by Pietro Michiardi and Giulio Franzese

EURECOM

• Information estimation with discrete diffusion: proposed INFO-SEDD, a discrete diffusion-based estimator of mutual information and entropy for high-dimensional discrete data; outperforms SOTA estimators and unlocks applications in genomics and summary evaluation. First author, ICLR 2026.
• Thermodynamic sampling schedules: derived optimization-free schedules for uniform discrete diffusion from entropy production, dynamical activity and a variational form of the discrete 1-Wasserstein distance; on a 7B diffusion LM they match the uniform schedule on MBPP with up to half the function evaluations. First author, under review at the PriGM workshop at NeurIPS 2026.
• Benchmarking mutual information estimation: unified existing benchmarks under a copula-theoretic framework and built a suite of synthetic, flow-based and real-image tasks; showed that no estimator class (non-parametric, discriminative, generative) dominates and identified the failure modes of each. First author, under review at JMLR.
• Optimal noise allocation for diffusion training: co-developed a theory showing that, under convexity, the optimal training schedule concentrates on finitely many noise levels. Co-author, with Radboud University and Sony AI.
• Mentored students in the Advanced Statistical Inference course and supervised semester projects.
• Contributed to the MUSECOM2 project in collaboration with the University of Oulu, Bosch and the Czech Technical University in Prague.

Research intern

EURECOM

• Conducted research on functional diffusion models, synthesizing the experimental results and theoretical findings into a master’s thesis.

Honours program

Polytechnic of Turin and Polytechnic of Milan

Alta Scuola Politecnica

• Honours program for the best 160 students of the Polytechnic of Turin and of the Polytechnic of Milan.
• Developed weakly supervised segmentation methods for waste sorting, contributing to a co-authored publication at the ICCV 2025 Workshops.

MSc

Polytechnic of Turin

Data Science and Engineering

• Relevant coursework: machine learning, deep learning, statistics, information theory, natural language processing, big data, learning theory, optimization on networks, graph theory, artificial intelligence, data ethics, data protection.
• Final grade: 110/110 cum laude.

BSc

Polytechnic of Turin

Computer Engineering

• Completed the Percorso Talenti honours program for the best 200 students in the cohort.
• Final grade: 110/110.

Top Reviewer

Conference on Neural Information Processing Systems (NeurIPS 2026)

Information Estimation with Discrete Diffusion

International Conference on Learning Representations (ICLR)

WS2: Weakly Supervised Segmentation using Before-After Supervision in Waste Sorting

IEEE/CVF International Conference on Computer Vision (ICCV) Workshops

INFO-SEDD: Continuous Time Markov Chains as Scalable Information Metrics Estimators

ICLR 2025 Workshop on Deep Generative Models: Theory, Principle and Efficacy

Improved Sampling Schedules for Discrete Diffusion Models

arXiv preprint arXiv:2602.06849

Towards Diverse and Comprehensive Benchmarks for Mutual Information Estimation

arXiv preprint arXiv:2607.03487

From Atoms to Entropy: Optimal Noise Allocation for Diffusion Training in the Convex Regime

arXiv preprint arXiv:2607.20540

Improved Plastic Sorting and Ecodesign (IPSE), with Alta Scuola Politecnica, Wisort and Corepla

• Designed and implemented a data collection system for images of a conveyor belt in a sorting plant.
• Built a semantic segmentation dataset with the custom data collection system.
• Implemented supervised and unsupervised semantic segmentation models.
• Integrated foundation models (SAM) to refine unsupervised segmentation predictions.

Predictive models for humanitarian aid, with the United Nations

• Designed a data augmentation strategy for time series based on causal graphs and kernel density estimation, reducing XGBoost mean squared error by 26%.
• Built a dataset of social tensions in the Sahel region.
• Completed a social impact assessment of the project’s output.