Quantum Reservoir Computing for time-series processing

Time-series processing is the task of inferring, from a finite record of a signal, its future values or functions of its past. Statistical models built on autoregression capture the linear temporal correlations of a signal, while deep neural networks learn nonlinear representations of its history at a high training cost. Quantum reservoir computing offers a third route - the time series is embedded into the state of a quantum system evolving under a fixed Hamiltonian, expectation values of measured Pauli observables provide a classical, nonlinear representation of the signal's history, and only a linear readout is trained.

In this talk, we will present the quantum reservoir computing paradigm and quantum extreme learning machines, outlining the requirements on the reservoir dynamics. We will then review our recent results on the role of memory in quantum reservoirs, and discuss open research directions, in particular, quantum reservoirs based on non-Markovian open quantum systems, where the memory of the bath can serve as a computational resource.

[1] "Harnessing quantum back-action for time-series processing" G. Franceschetto, M. Płodzień, M. Lewenstein, A. Acín, P. Mujal Phys. Rev. X 16, 012417 (2026)

[2] "From quantum reservoirs to quantum extreme learning machines through a nearest-neighbor spin chain with tunable quantum memory" C. Ramon-Escandell, A. Riera, M. Płodzień arXiv:2608.28440