A Deep Generative Model for Semi-Supervised Classification with Noisy Labels
This work addresses label noise in semi-supervised classification, which is a common issue in real-world datasets, but it appears incremental as it builds upon existing deep generative models.
The authors tackled the problem of classification with noisy labels by proposing a new semi-supervised deep generative model called Mislabeled VAE (M-VAE), which explicitly models label noise and outperforms existing models that do not account for such noise.
Class labels are often imperfectly observed, due to mistakes and to genuine ambiguity among classes. We propose a new semi-supervised deep generative model that explicitly models noisy labels, called the Mislabeled VAE (M-VAE). The M-VAE can perform better than existing deep generative models which do not account for label noise. Additionally, the derivation of M-VAE gives new theoretical insights into the popular M1+M2 semi-supervised model.