IFISC researchers propose a new protocol for data processing with Quantum Reservoir Computing

• Researchers at IFISC (UIB-CSIC) propose a new protocol for processing sequential data using quantum machine learning.
• The study proposes a way to efficiently include quantum measurement while preserving the quantum advantage that characterises these systems.

Researchers at the Institute for Cross-Disciplinary Physics and Complex Systems, IFISC (UIB-CSIC), in Mallorca, propose the first protocol that includes the effect of measurement in the processing of temporal data sequences using quantum systems. Examples of these computational tasks are handwriting recognition or the prediction of chaotic series. The advantage of using quantum systems for these purposes lies in the large processing power provided by the Hilbert space of quantum states, an exponential advantage over classical systems. Moreover, it has now been shown that this advantage can be achieved even in non-ideal situations, where the effect of quantum measurement is taken into account.

The implementation of quantum reservoir computing as a computational method for processing time series data has a lot of potential, but faces several challenges. One of them, common to all quantum computing, is that, due to its stochastic nature, it is necessary to repeat the processing of the information several times and to calculate averages with the results obtained in order to improve accuracy. The other problem is that quantum systems are strongly affected by measurements, i.e. the process of obtaining the processed information. In an implementation of quantum reservoir computing this is especially relevant, as it can impair the quality of the processing at different times. To prevent the next steps from being affected by past measurements, the experiment would have to be restarted by reintroducing the data into the system from the beginning, which is clearly inefficient. In addition, it would be necessary to store the data in an external memory. The researchers have analysed different protocols for time series processing, including the rewinding and restarting protocols, and have proposed an alternative based on weak measurements that allows continuous online monitoring of the data without external storing, operating in real time.

This online protocol proposed by the researchers, presented in the journal npj Quantum Information, introduces the effect of the measurement on data processing. Typically, weak measurements provide less information and are noisier, but in this way of obtaining processing results the quantum system does not "collapse" as a whole, making it possible to identify situations in which effective data processing is achieved in both accuracy and resources.

The study establishes the advantage of quantum reservoirs in realistic scenarios and is expected to pave the way for efficient experimental implementations involving continuous time series processing with quantum systems. In addition, this research may also contribute to the development of concrete applications such as quantum time tomography, quantum recurrent neural networks or quantum neuromorphic computing, among other advances.

Image: Schematic of the protocol proposed by the researchers. The measurements of the quantum reservoir that processes the time series are weak, so the quantum system does not "collapse" in its entirety.

Mujal, P., Martínez-Peña, R., Giorgi, G.L., Soriano, M.C., Zambrini, R., Time-series quantum reservoir computing with weak and projective measurements. npj Quantum Inf 9, 16 (2023). https://doi.org/10.1038/s41534-023-00682-z

• El diari de la UIB