Quantum-based Optimization for Large-Scale Machine Learning (QARECA)
Machine learning has entered the large-scale era, where models with billions of parameters require unprecedented computational resources. As transistor scaling approaches its physical limits, sustaining this growth becomes increasingly difficult, challenging the continuation of Moore's law. This proposal addresses this issue through a three-layer design stack comprising a foundation, an algorithmic, and an architectural layer. The foundation layer reformulates large-scale machine learning using Hilbert vector spaces and tensor decomposition to reduce parameter dimensionality while enabling efficient quantum training. The algorithmic layer introduces co-evolutionary algorithms that jointly optimize Hilbert-space parameters and tensor cores, overcoming barren plateaus and improving scalability and training efficiency. The architectural layer develops a quantum-centric supercomputing framework integrating GPUs and quantum processors, supported by advanced distribution and error-mitigation techniques to enable practical execution on NISQ devices. The proposed approach will be validated in national security, healthcare, and climate change applications, strengthening Italy's leadership in quantum technologies for large-scale machine learning.
Funding body:
Ministero dell'Università e della RicercaPrincipal Investigator:
Giovanni AcamporaBudget:
€ 1.900.000,00Date:
July 9, 2026
Category:
