Harnessing Rydberg Dynamics for Quantum Kernel Learning - Victor Drouin-Touchette

Neutral atoms in optical tweezers provide a scalable platform for quantum information processing, with strong interactions enabled by the Rydberg blockade. This talk explores how the resulting dynamics can be harnessed for practical quantum machine learning (QML), focusing on quantum kernel methods (QKMs). While QKMs often suffer from exponential concentration—requiring exponentially many measurements to resolve nontrivial kernels—we propose a novel kernel that avoids this pitfall by exploiting weak ergodicity-breaking in many-body Rydberg dynamics. We present analytical insights from a toy model, extensive simulations on synthetic data, and empirical results on real datasets. Crucially, the proposed kernel is both classically hard to simulate and implementable on current neutral atom quantum processors.