Quantum Reservoir Computing (QRC) combines classical reservoir computing with quantum computing. Reservoir computing (RC) is a computational framework derived from recurrent neural networks (RNNs) that offers an efficient and flexible approach for processing temporal data. Unlike traditional RNNs, where all weights are trained, reservoir computing simplifies the learning process by keeping most of the network (i.e. the "reservoir") fixed and only training the output layer. This significantly reduces computational complexity while retaining the network's ability to model dynamic and nonlinear systems.
This repository walks you through setting up a hybrid reservoir computing for time-series prediction and signal processing — blending classical ESNs with quantum reservoir computing (QRC) on qBraid. You can also further optimize your training process by utilizing GPUs on qBraid. To learn more, check out the qBraid publication: arXiv:2505.22837.
✨ What you’ll get
- Classical ESN baseline: Quickly stand up a robust echo state network and tune spectral radius, sparsity, and leak rate.
- Quantum Reservoir Computing (QRC): Build a quantum reservoir (fixed random circuit + linear readout), run on simulator or hardware, and compare against classical baselines.
- Hybrid & composable models: Combine classical and quantum reservoirs (e.g., concatenation or onion-style multi-scale reservoirs) and measure accuracy/latency tradeoffs.
- End-to-end workflow: Data prep → reservoir design → training (linear readout) → evaluation → visual diagnostics (loss curves, scatter/forecast plots).
📈 Outputs
- Forecast plots (prediction vs. ground truth)
- Error metrics (MSE/MAE/R²)
- Reservoir diagnostics (state norms, spectral radius checks)
- Ablations (shots, circuit depth, reservoir size)
🧭 Tips & troubleshooting
- Interested in trying this for your own dataset? Use AI tools on qBraid Lab to start the QRC-tool. Just provide a CSV file with your time-series data and watch how custom qBraid-AI tools help you get going on your first QRC results
- GPU not used? Ensure your kernel is bound to a GPU-enabled image on qBraid; classical training benefits most.
- Quantum simulator slow? Reduce shots or circuit depth; start with simulators before hardware.
- Different SDKs? qBraid supports multiple quantum stacks; swap Qiskit for your preferred framework with minimal changes.
- Numerical stability: normalize inputs; tune ridge penalty; verify spectral radius < 1 for ESN echo state property.