可訓練的核函數在量子變分線路中的效益

Abstract

量子運算在機器學習領域的應用越來越受到關注。在各種量子學習模型中，變分量子電路因其易於實作的特性而顯得尤其突出。透過了解這個模型的數學特性，可以發現具有可訓練的量子核函數的模型似乎有提升表現的潛力。我針對不同的問題測試了幾種具有此特性的方法，結果證實這些方法確實提高了模型的表達能力，也突破了使用固定核函數的模型的表現限制。

The application of quantum computing to machine learning problems has grown increasingly popular. Among the various quantum models, variational quantum circuits are particularly notable for their ease of implementation. Investigating the mathematical properties of such models reveals that those equipped with trainable quantum kernels may achieve enhanced performance. We tested several methods possessing this characteristic across different problems, and the results confirm that such methods not only improve model expressibility but also surpass the performance limits of models with fixed kernels.