GMM Discriminant Analysis with Noisy Label for Each Class
This addresses the problem of noisy labels in real-world datasets for machine learning practitioners, offering an incremental improvement over existing methods.
The paper tackles classification with noisy labels by proposing a Gaussian Mixture Discriminant Analysis method that incorporates flipping and class probabilities, using EM algorithms, and demonstrates it outperforms four state-of-the-art methods on synthetic and real-world datasets.
Real world datasets often contain noisy labels, and learning from such datasets using standard classification approaches may not produce the desired performance. In this paper, we propose a Gaussian Mixture Discriminant Analysis (GMDA) with noisy label for each class. We introduce flipping probability and class probability and use EM algorithms to solve the discriminant problem with label noise. We also provide the detail proofs of convergence. Experimental results on synthetic and real-world datasets show that the proposed approach notably outperforms other four state-of-art methods.