Eliciting and Learning with Soft Labels from Every Annotator
This work addresses the challenge of reducing annotation costs for practitioners in machine learning by enabling the use of soft labels from fewer annotators, though it is incremental as it builds on existing soft label approaches.
The paper tackles the problem of efficiently eliciting soft labels from individual annotators to improve model generalization, robustness, and calibration, and shows that learning with these labels achieves comparable performance to prior methods while requiring far fewer annotators (N=248 in a CIFAR-10 test set study).
The labels used to train machine learning (ML) models are of paramount importance. Typically for ML classification tasks, datasets contain hard labels, yet learning using soft labels has been shown to yield benefits for model generalization, robustness, and calibration. Earlier work found success in forming soft labels from multiple annotators' hard labels; however, this approach may not converge to the best labels and necessitates many annotators, which can be expensive and inefficient. We focus on efficiently eliciting soft labels from individual annotators. We collect and release a dataset of soft labels (which we call CIFAR-10S) over the CIFAR-10 test set via a crowdsourcing study (N=248). We demonstrate that learning with our labels achieves comparable model performance to prior approaches while requiring far fewer annotators -- albeit with significant temporal costs per elicitation. Our elicitation methodology therefore shows nuanced promise in enabling practitioners to enjoy the benefits of improved model performance and reliability with fewer annotators, and serves as a guide for future dataset curators on the benefits of leveraging richer information, such as categorical uncertainty, from individual annotators.