Learning in Light: Modular and Photonic Reservoirs at Work

Quantum reservoir computing (QRC) offers a promising approach for harnessing complex quantum dynamics to perform machine learning on near-term devices. We present advances in quantum extreme reservoir computing (QERC) that unify photonic and modular perspectives. Using boson sampling interferometers, our quantum optical reservoir computing (QORC) model encodes data into few-photon interference patterns, producing high-dimensional feature maps that enable accurate classification of benchmark datasets such as MNIST, K-MNIST, and Fashion-MNIST. We further demonstrate that modular reservoir principles—where finite-range interactions and sparse inter-module links yield performance comparable to fully connected networks—also apply to photonic architectures, guiding scalable interferometer designs. This integrated framework shows that modest connectivity suffices to capture complex dynamics, making photonic reservoirs experimentally practical. Beyond benchmarks, the approach holds promise for real-world applications such as medical image classification, where efficient quantum-enhanced feature extraction could play a transformative role.