A Deep Generative Model for Semi-Supervised Classification with Noisy Labels

09/16/2018

∙

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.

READ FULL TEXT

A Deep Generative Model for Semi-Supervised Classification with Noisy Labels

Semi-supervised source localization in reverberant environments with deep generative modeling

Deep Dive into Semi-Supervised ELBO for Improving Classification Performance

SHOT-VAE: Semi-supervised Deep Generative Models With Label-aware ELBO Approximations

DGPose: Disentangled Semi-supervised Deep Generative Models for Human Body Analysis

Boosting Generative Models by Leveraging Cascaded Meta-Models

Semi-supervised Neural Chord Estimation Based on a Variational Autoencoder with Discrete Labels and Continuous Textures of Chords

Semi-crowdsourced Clustering with Deep Generative Models

A Deep Generative Model for Semi-Supervised Classification with Noisy Labels

Related Research

Semi-supervised source localization in reverberant environments with deep generative modeling

Deep Dive into Semi-Supervised ELBO for Improving Classification Performance

SHOT-VAE: Semi-supervised Deep Generative Models With Label-aware ELBO Approximations

DGPose: Disentangled Semi-supervised Deep Generative Models for Human Body Analysis

Boosting Generative Models by Leveraging Cascaded Meta-Models

Semi-supervised Neural Chord Estimation Based on a Variational Autoencoder with Discrete Labels and Continuous Textures of Chords

Semi-crowdsourced Clustering with Deep Generative Models