.png){kind=icon}
.png){kind=icon}
Return
Meta-learning representations for learning from multiple annotators
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2026.132744.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
We propose a meta-learning method for learning from multiple noisy annotators. In many applications such as crowdsourcing services, labels for supervised learning are given by multiple annotators. Since the annotators have different skills or biases, provided labels can be noisy. To learn accurate classifiers, existing methods require many noisily annotated data. However, sufficient data might be unavailable in practice. To overcome the lack of data, our method uses labeled data obtained in different but related tasks. Our method embeds each example in tasks into a latent space by using a neural network and constructs a probabilistic model for learning a task-specific classifier while estimating annotators’ abilities in the latent space. This neural network is meta-learned to improve the expected test classification performance when the classifier is adapted to a given small amount of annotated data. This adaptation is performed by maximizing the posterior probability via the expectation-maximization (EM) algorithm. Since each step in the EM algorithm is computed in closed-form and is differentiable, our method can efficiently backpropagate the loss through the EM algorithm to meta-learn the neural network. We demonstrate the effectiveness of our method with real-world datasets containing synthetic noise and real-world crowdsourcing datasets.