Quantum-inspired machine learning on FPGA

by Dr Lorenzo Borella (DFA)

Thursday 5 Jun 2025, 17:15 → 17:30 Europe/Rome

1/1-3 - Aula B (Dipartimento di Fisica e Astronomia - Edificio Marzolo)

Tensor Networks (TNs) offer a powerful framework for representing quantum many-body systems and have recently gained attention for their application in Machine Learning, achieving performance comparable to traditional supervised methods. Unlike classical approaches, TNs can leverage quantum-inspired features—such as quantum entanglement and data correlations—to enable a form of learning informed by the internal structure of the data. These features can be exploited to construct highly compressed models without sacrificing accuracy, a particularly valuable property in resource-constrained environments such as FPGAs, where ML models can be embedded to perform tasks with sub-microsecond latency. This project investigates the use of Tree Tensor Networks (TTNs) as ML classifiers in real-time, high-frequency scenarios, leveraging the low-latency capabilities of Field-Programmable Gate Arrays (FPGAs). As a demonstrator, a TTN classifier designed for deployment within the trigger pipelines of High Energy Physics (HEP) experiments was implemented, achieving fully pipelined execution with sub-microsecond latency. This result highlights the feasibility and efficiency of combining quantum-inspired ML models with specialized hardware for fast, reliable scientific data processing in time-critical environments.