Dual-Primal Graph VAEs for Noisy Label Aggregation

2026-08-11

AI Summary

This work addresses the challenge of label noise in crowdsourced data by proposing a novel graph variational autoencoder (Graph VAE) framework that models true labels as latent variables. The approach performs end-to-end unsupervised representation learning by employing graph attention networks (GATs) for both encoding and decoding on the original crowdsourcing data graph and its dual graph. Notably, it is the first to leverage a dual graph structure for label aggregation, eliminating the need for pseudo-labels or synthetic data. The framework also naturally incorporates side information from neural classifiers to enhance performance. Extensive experiments on multiple crowdsourcing benchmark datasets demonstrate state-of-the-art results, confirming the effectiveness of the proposed architecture in both denoising and classification accuracy.

Abstract

Inferring the ground-truth from noisy crowdsourced labels is an important theoretical and practical problem. Neural network-based methods offer an alternative to classical Bayesian models which require specifying a family of generative models used for inference. However, current models either still rely on fairly simple generative models for inference or require pseudo-labels or synthetic data to train the aggregate classifier. We propose a graph VAE architecture in which the decoder and encoder use GAT-based message passing on the adjacency graph of a crowdsourced dataset and its dual, respectively. The ground-truth labels are treated as latent variables, enabling unsupervised representation learning without needing to train a separate classifier. We show our model achieves state of the art performance on crowdsourcing benchmarks. We then demonstrate the generality of our approach by showing how the original crowdsourcing graph can be augmented to incorporate side information such as representations from neural network classifiers trained on the noisy labels to substantially boost their classification performance at test time.

Problem

• noisy label aggregation
• crowdsourcing
• ground-truth inference
• label denoising

Innovation

• Graph VAE
• Noisy Label Aggregation
• Dual-Primal Graph
• GAT-based Message Passing
• Unsupervised Representation Learning