| Abstract |
This talk begins with the quantum feature map of QSVM, explaining that the quantum kernel function actually reflects the data geometry induced by the feature map in Hilbert space. When a noisy CPTP channel in a real quantum device acts on a quantum eigenstate, the original inner-product structure and kernel matrix are distorted, thereby changing the QSVM classification boundary. Therefore, the noise-robustness problem of QSVM can be viewed as a deformation and recovery problem of the feature map geometry induced by the quantum channel. This presentation will introduce how to replace global channel inversion with local channel recovery and discuss the roles of MUB, Choi matrix, quasi-inverse, and diamond norm bounds in this problem. |
