Speakers
Description
Forecasting chaotic dynamical systems is challenging due to sensitivity to initial conditions and long-term dependencies. We benchmark analog Quantum Reservoir Computing (QRC) on the Pasqal neutral-atom platform against classical forecasters (MLP, LSTM, CRC) using Mackey-Glass, Hénon, and Lorenz-63 datasets from ReservoirPy. Temporal inputs are encoded as continuous detuning waveforms driving interacting Rydberg atoms, whose many-body dynamics act as a fixed, untrained reservoir with classical linear readout, avoiding barren-plateau issues typical of variational algorithms.
We evaluate forecasting accuracy (MSE), memory capacity, and generalization under a sliding-window protocol, systematically varying qubit number, array topology, Rabi frequency, and probing time. Given realistic near-term constraints, results progress from noiseless to shot-noise-limited simulations, with selective hardware validation. This characterizes how far Rydberg-induced fading-memory dynamics can practically reconstruct chaotic attractors under current hardware limitations.
| Topic | Hybrid quantum-classical approaches for learning and optimization |
|---|