Quantum Embedding Kernels for Evaluating Properties of Carbon Fiber Fabrication

HSU-KAI CHENG^1*, CHUN-WEI PAO¹

¹Research Center for Applied Sciences, Academia Sinica, Taipei, Taiwan

* Presenter:HSU-KAI CHENG, email:kaicheng0108@gmail.com

The development of pitch-based carbon fibers remains to be an important issue in materials science with focus on optimizing targeted properties and reducing high production costs. One difficulty in carbon fiber fabrication can be traced back to a large number of controllable parameters during multi-stage fabrication. Each unique set of parameters represents a high-dimensional feature vector mapped to the final product with desired characteristics. Handling this high-dimensional feature vector intrinsically portrays a problem with kernel calculations and quantum kernels are believed to be computationally more efficient than classical kernels. In this work, a quantum embedding kernel (QEK) is proposed to categorize high-dimensional feature vectors into four classes according to the ranges of the targeted property, say Young’s Modulus. A quantum circuit is constructed to embed data into the quantum states and the corresponding QEK is trained and provided as a similarity measure between vectors for a support vector classifier. We demonstrated that a high accuracy prediction can be performed using a well-trained QEK based on a five-qubit and six-layer quantum circuit architecture.

Keywords: Quantum machine learning, quantum neural network, Quantum Kernel, carbon fiber