11^th FNIP seminar |T3-T2
AI-driven research and research-driven AI: from Photonics to Neuroscience

Watch live: REGISTER HERE

Schedule

15:15 – 15:30 → Introduction with M. Bruzzone, F. Lorenzi, M. Brondi
15:30 – 16:00 → Daniel Brunner (CNRS, FEMTO-ST, FR) – End-to-end training and scaling laws for computing with physical systems
16:00 – 16:30 → William Buchser (WashU, St. Louis, US) – From Single-Cell Phenotyping to Spatial Atlases: Image-First Spatial Omics in the Nervous System
16:30 – 17:00 → Tommaso Dorigo (LTU, SE & INFN, IT) – A Second AI Revolution for Fundamental Science
17:00 – 17:15 → Follow-up session – Open questions

Speaker biographies and talk abstracts here

AI-driven research and research-driven AI will gather three internationally recognized researchers to explore the convergence of artificial intelligence, advanced computation, and data-intensive scientific discovery across photonics, neuroscience, and particle physics. The event highlights how modern machine learning, physical computing substrates, and simulation-driven inference are transforming both experimental design and the extraction of knowledge from complex, high-dimensional datasets.

The first talk, by Daniel Brunner, presents recent advances in physical neural computing based on semiconductor laser photonic architectures. His work demonstrates a fully hardware-embedded neural network in which computation is performed directly in the physical substrate, avoiding digital pre- and post-processing. The system is trained end-to-end using model-free approaches and enables exploration of scaling laws linking physical system properties, computational capacity, and performance in high-dimensional photonic neural networks.

The second contribution, by William Buchser, introduces an image-first framework that uses biophysical simulation to generate synthetic training data for tissue segmentation from subcellular level to large anatomical regions, and discusses extensions to 3D modeling and applications in large-scale neuroscience platforms. The talk will also touch on advances in single-cell phenotyping of human iPSC-derived neurons and the broader potential of spatial omics as quantitatively structured data.

The third lecture, by Tommaso Dorigo, outlines how the rise of deep learning marked a transformation in data analysis methodologies and how a new AI-driven paradigm is emerging, aimed at optimizing the design and interpretation of complex experiments. Particular attention will be given to co-design strategies that integrate hardware and software, addressing the challenges posed by high-dimensional detector parameter spaces, stochastic physical processes, and multi-objective optimization. He will also highlight the potential of neuromorphic and AI-based approaches for in-situ dimensionality reduction and enhanced inference in large-scale physics experiments.