About

Hi, I’m Michele. I’m a PhD candidate at FAU Erlangen-Nuremberg, and I work on AI safety which we can summarize in two modalities. 1) Try to understand when a data sample (from real-world scenario or model output) is never seen and potentially leading to failure and therefore to high costs (in literature is called distribution shift detection) or 2) certify reliable prediction from deep learning models using uncertainty estimation. I am currently collaborating with Mercedes and with Nokia for those topics. Thanks for stopping by, and feel free to say hi on LinkedIn , Github , or by email .

DOWNLOAD MY RESUME

EXPERIENCE

  • Thesis Erasmus

    DTU, Denmark Sept 2020 - Mar 2021

    Six month Erasmus for my Master Thesis in DTU (Denmark Technical University) supervised by Ole Winther

  • Co-Founded startup at Cohenagen, Denmark

    Prediba 2019-2020

    Founded Prediba, a start-up that deliver predictive maintenance to manifacturing companies interested in imporove its efficiency via AI and digitalization

  • Mentoring at Poste Italiane, Alfamotion

    Freelance 2019

    Held Machine Learning courses starting from pre-requirements, to Machine Learning and Deep Learning topics

  • Developed Official App SIS2017

    Freelance Sep 2016 - Jun 2017

    SIS is an annual italian statistic conference held in 2017 at Florence. The app guides the user to the talks, events and the building. Since performances weren’t critical, the application was developed like a website to code-once and deploy everywhere

PUBLICATIONS

CVPRW · 2026

Forecasting the Past: Gradient-Based Distribution Shift Detection in Trajectory Prediction

Michele De Vita, Julian Wiederer, Vasileios Belagiannis

WACV · 2025

Diffusion model guided sampling with pixel-wise aleatoric uncertainty estimation

Michele De Vita, Vasileios Belagiannis

ICML · 2022

Generalization and robustness implications in object-centric learning

Andrea Dittadi, Samuele Papa, Michele De Vita, Bernhard Schölkopf, Ole Winther, Francesco Locatello