HQKAN: quantum-inspired network cuts federated ECG costs by 37%

A new arXiv paper (2608.13914) introduces HQKAN, a hybrid quantum-inspired Kolmogorov-Arnold network designed to tackle a practical problem: training ECG classifiers across hospitals and wearable devices without sharing raw biosignal data. Federated learning allows collaboration while keeping data local, but arrhythmia classification suffers from limited client-side samples, imbalanced labels, and non-IID data distributions. The authors—Chun-Hua Lin, Samuel Yen-Chi Chen, and colleagues from Taiwan's top institutions—compared HQKAN against a standard multilayer perceptron (MLP) using federated averaging (FedAvg).

The results are striking. On the MIT-BIH dataset (five-class arrhythmia classification), HQKAN improved aggregate and minority-class metrics while using 37.35% fewer trainable parameters and reducing communication cost by 24.89%. On the INCART dataset (three-class), the improvements were even larger: 44.81% fewer parameters and 36.41% lower communication overhead. The quantum-inspired component—likely a tensor-network or parameterized quantum circuit analogy—gives the model a compact representation that captures complex patterns without the parameter bloat of dense layers.

For real-world deployment, this matters because hospitals and wearables often run on limited bandwidth and edge hardware. A 25–36% reduction in communication cost means faster federated training rounds and less battery drain on wearable ECG monitors. The accuracy gains on minority classes (e.g., rare arrhythmia types) are clinically crucial—those are exactly the patterns most likely to be missed. While this is a research preprint, it points toward a practical path where quantum-inspired architectures can solve mundane but pressing healthcare infrastructure problems, not just theoretical physics.