We are pleased to announce the Second AI_INFN Advanced Hackathon, following the successful inaugural edition held in Padova in 2024, building on the tradition established by ML_INFN since 2020.
This year the AI_INFN Advanced Hackathon will be hosted by INFN Pavia, from Monday, November 24 to Thursday, November 27, 2025 in the rooms of the Almo Collegio Borromeo.
This four-day event is primarily designed for early-career scientists (PhD students, postdocs, and Master’s students) to provide a comprehensive overview of Machine Learning and Artificial Intelligence activities within INFN, combined with intensive hands-on practice on cutting-edge ML topics.
Thanks to the support of INFN Pavia and the Italian Research Center on High Performance Computing, Big Data and Quantum Computing (ICSC), participation in the event will be free of charge. As in previous editions, participants must be employed by or associated with INFN, or provide a reference letter from an INFN supervisor.
Hands-on sessions
Participants will engage in hands-on exercises with machine learning techniques using resources provided by INFN CNAF, ReCaS Bari, and INFN Milano-Bicocca.
The hackathon will feature a balanced mix of lectures and practical sessions, fostering both theoretical understanding and technical skills development. Participants will have the opportunity to work with state-of-the-art tools and methodologies while engaging with the broader INFN AI research community.
This year, the exercises that will be discussed in the long hands-on sessions in the afternoons are the following:
Quantum Optimization with QUBO: Graph Coloring and Ising Models
Explore the formulation of optimization problems using QUBO and solve them with quantum hardware. The exercise focuses on graph coloring, with applications in finding ground states of physical systems modeled by Ising Hamiltonians. Ideal for understanding quantum-inspired approaches to physics problems.
Explainable Graph Transformers for ATLAS Data
Hands-on with a graph-transformer model featuring a mixture-of-experts orchestrator for selective activation. Based on ATLAS HEP data, this exercise introduces key concepts in GNNs, attention mechanisms, and explainable AI. Participants will learn how activation patterns reveal insights into model predictions.
Physics-Informed Neural Networks for Gel Dosimetry
Apply PINNs to model dose distributions in gel-based medical physics applications. This exercise blends domain knowledge with deep learning to solve PDEs relevant to radiation therapy. Suitable for exploring interpretable ML in biomedical contexts.
Signal Compression and FPGA Deployment with the Virgo experiment
Combine two advanced workflows: autoencoder-based compression of Virgo gravitational wave signals and FPGA deployment through DNN transpilation. Participants will use Keras/TensorFlow and HLS4ML to explore ML on edge devices. A cross-domain exercise bridging astrophysics and high-energy instrumentation.
Bottom-up workshop contributions
For the first time, the Advanced Hackathon will feature research talks selected through an open abstract submission process. Contributions from both participants and other researchers are welcome, with the goal of fostering a strong network of Machine Learning experts in nuclear, subnuclear, and applied physics research.
