Speaker
Description
Quantum computing promises new ways to tackle high-dimensional, combinatorial optimization problems that appear throughout scientific modelling. Flood prediction is one such area: modern neural networks can support large-scale, data-driven flood mapping, but their performance is highly sensitive to hyperparameter choices, and repeated tuning is computationally expensive. In this work, we investigate a Hybrid Quantum Genetic Algorithm (HQGA) as a NISQ-era quantum optimization primitive for hyperparameter search in a flood-prediction neural network.
Our case study targets binary classification of flooded versus non-flooded areas for a simulated event along the Saskatoon River, Canada. HQGA encodes candidate hyperparameter configurations (e.g., multilayer perceptron depth and width, learning rate, dropout, batch size and epochs) in a parameterized quantum circuit, while classical genetic operators perform selection, crossover and mutation. We compare HQGA against two strong classical baselines for hyperparameter optimization—a genetic algorithm (GA) and Bayesian optimization (BO)—under a shared search space and budget. Each method is run for 20 batches of 20 runs (400 HPO runs per method); every selected configuration is then evaluated by training the neural network with multiple initialization seeds and testing on both a held-out test set and an “entire region” evaluation set.
Across all methods, mean accuracy, F1-score, precision and recall are comparable, showing that the quantum-enhanced optimizer reaches a similar performance regime to the classical baselines. The key quantitative advantage of HQGA is stability: over 400 runs, it exhibits a statistically significant reduction in the variance of evaluation metrics and in the number of generations required to reach near-optimal performance, compared with GA and BO. This indicates that HQGA yields consistently strong hyperparameter settings with fewer poor runs.
From a broader quantum-computing perspective, our results provide an application-driven benchmark of a hybrid quantum optimizer embedded in a realistic environmental modelling workflow. They show that, even in the NISQ regime and without claiming a formal quantum speedup, quantum-enhanced search can already act as a robust, practically useful component in flood-prediction pipelines, and outline a concrete path for scaling quantum optimization to more complex hydrological and climate-risk models as hardware matures.
| Keyword-1 | Quantum Machine learning |
|---|---|
| Keyword-2 | Quantum optimization |
| Keyword-3 | Quantum computation |