A Novel Transfer Matrix Method Based on Kernel Maximum Mean Discrepancy for Noisy Samples

The previous section highlighted the limitations of the traditional transfer matrix method when encountering noisy samples and inaccurate posterior estimates of noise probabilities. These limitations stem from the discrepancy between probability distributions A and B. To address this, we introduce a new metric to evaluate the difference between probability distributions. To overcome this challenge, we propose a novel transfer matrix method based on kernel maximum mean discrepancy (MMD). This method leverages kernel functions to quantify the divergence between different probability distributions and applies this measurement to the estimation of the transfer matrix. By minimizing the MMD, we obtain more accurate estimations of the transfer matrix. Experimental results validate the effectiveness of our proposed method in scenarios involving noisy samples and inaccurate posterior estimates of noise probabilities.