An Uncertainty-Aware Pseudo-Label Selection Framework using Regularized Conformal Prediction

TL;DR

This paper introduces an uncertainty-aware pseudo-label selection framework using conformal prediction to improve semi-supervised learning by reducing noisy labels and fixing model calibration issues.

Contribution

It proposes a novel framework that leverages conformal regularization to select reliable pseudo-labels, enhancing semi-supervised learning without domain-specific data augmentation.

Findings

Improved pseudo-label quality through uncertainty sets.

Enhanced model calibration and reduced noise.

Better semi-supervised learning performance.

Abstract

Consistency regularization-based methods are prevalent in semi-supervised learning (SSL) algorithms due to their exceptional performance. However, they mainly depend on domain-specific data augmentations, which are not usable in domains where data augmentations are less practicable. On the other hand, Pseudo-labeling (PL) is a general and domain-agnostic SSL approach that, unlike consistency regularization-based methods, does not rely on the domain. PL underperforms due to the erroneous high-confidence predictions from poorly calibrated models. This paper proposes an uncertainty-aware pseudo-label selection framework that employs uncertainty sets yielded by the conformal regularization algorithm to fix the poor calibration neural networks, reducing noisy training data. The codes of this work are available at: https://github.com/matinmoezzi/ups conformal classification