NISQRC in a Flexible Qubit-Bus Architecture 1: Static Inference and Time-series Processing of Quantum Data

Quantum data processing is the most promising application of quantum machine learning (QML), where the learning algorithm can both naturally access the exponential complexity of the input state, and does not suffer from the overhead associated with encoding classical data in the quantum domain and subsequently measuring it. Here we present a robust and flexible superconducting circuit platform for QML that can be readily integrated in the quantum computing stack, and describe how it can be operated as a quantum reservoir computer (QRC) for either static or temporal quantum data. The architecture consists of an array of transmon qubits coupled to a common bus resonator which mediates all-to-all coupling, yielding a complex and expressive analog quantum feature map without the need for well-calibrated gate sequences. The bus further provides a robust readout channel; we show that through tuning the measurement strength one can optimize for output noise or memory, and thus for static or temporal inference tasks. This weak measurement regime, when combined with intrinsic energy-damping, enables non-unital dynamics essential for quantum sequence data processing and is the continuous analog of the NISQRC algorithm [Hu et al, Nat Commun 15, 7491 (2024)].