Prevenire l'Obsolescenza dei Modelli ML: Il Caso Bioretics
Friday, May 30
16:15 - 16:45
Il machine learning sta rivoluzionando l’industria, ma mantenere modelli performanti in ambienti dinamici è una sfida. Esploreremo tecniche di monitoraggio e riaddestramento dei modelli per garantire prestazioni costanti, mostrando un esempio pratico su un sistema di riconoscimento difetti frutta.
Il machine learning sta trasformando l’industria e il business, consentendo decisioni più rapide e accurate. Tuttavia, mantenere modelli di machine learning performanti in ambienti produttivi, spesso caratterizzati da dati in continua evoluzione, è una sfida significativa.
Questo talk ha l’obiettivo di esplorare le tecniche fondamentali per il monitoraggio e il riaddestramento dei modelli di ML, al fine di garantire prestazioni costanti nel tempo. Attraverso un esempio pratico, mostreremo come applicare queste soluzioni a un modello di machine learning basato su immagini utilizzato per il riconoscimento dei difetti della frutta, dimostrando l’importanza di un approccio strutturato alla manutenzione dei modelli.
Verrà descritta la piattaforma alla base della soluzione sviluppata e lo stack tecnologico impiegato, interamente basata su librerie python, in parte open source e in parte sviluppate internamente.
Il talk è rivolto a developers, data scientist e ML engineer interessati a migliorare la robustezza e l’affidabilità dei sistemi di machine learning in produzione.
Lorenzo Bisi
Lorenzo Bisi obtained his PhD in Information Engineering in 2022 at Politecnico, with a thesis on Reinforcement Learning algorithms specialized in risk-aversion contexts. A significant part of his research focused on applying Machine Learning techniques to develop algorithmic trading strategies. Since 2022, he has been working at ML Cube as an AI specialist, overseeing the development of Artificial Intelligence models for consulting projects. Here, he applies Machine Learning, Computer Vision, and Generative AI techniques to create customized solutions for clients in various industrial sectors.
Lavelli
I received my Master’s degree in Computer Science and Engineering from Politecnico di Milano in 2019.
After graduation, I worked as a research fellow at AIRLab of Politecnico di Milano until 2021. The research activity focused on the application of Reinforcement Learning and Machine Learning in the automotive, manufacturing and energy fields collaborating with Scuderia Ferrari, Pirelli and RSE.
Since 2021 I have been working as Senior Machine Learning Engineer in ML cube, an AI spin-off company of Politecnico di Milano. I am Technical Product Owner of ML cube Platform, a product dedicated to AI supervision and maintenance. The new modules of this product are dedicated to GenAI and LLM given the high need for monitoring and supervision of these complex systems.
I have been working with Python for more than 7 years for the development of AI algorithms and more traditional applications. I am now getting interested in Rust and its use as a backend for highly efficient Python modules.
Giovanni Giacometti
I currently work as a Machine Learning Engineer at ML cube, a Politecnico di Milano spin-off, where I’m involved in the development of the ML cube Platform, an MLOps solution that offers data and model monitoring, drift explainability and performance-safe retraining.
I hold a Master’s degree in Artificial Intelligence from Politecnico di Milano and a Bachelor’s degree in Computer Science from Università Politecnica delle Marche. During the Master, I had the opportunity to spend a semester at the IT University of Copenhagen as an Erasmus student, broadening my academic and cultural perspective.
My expertise lies at the intersection of AI, Machine Learning and Software Engineering, with a focus on:
- 🚀 Backend: mostly Python (FastAPI, Pydantic), a bit of Rust.
- 🤖 AI, Machine Learning & Data: LangGraph, Pydantic AI, Scikit-learn, PyTorch, Numpy, OpenCV, Polars, Pandas.
- 📅 Databases: SQL & NoSQL.
- 🔼 Deployment: Docker, Kubernetes, Terraform, Github Actions.
- ☁️ Cloud: AWS, GCP