FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence

TLDR

Semi‑supervised learning (SSL) leverages unlabeled data to improve model performance. The paper demonstrates the power of combining consistency regularization and pseudo‑labeling in SSL. FixMatch generates pseudo‑labels from weakly‑augmented unlabeled images, keeps only high‑confidence predictions, and trains the model on strongly‑augmented versions of the same images, with extensive ablation studies to identify key factors. FixMatch achieves state‑of‑the‑art performance on standard SSL benchmarks, such as 94.93 % accuracy on CIFAR‑10 with 250 labels and 88.61 % with 40 labels per class, and its code is publicly available.

Abstract

Semi-supervised learning (SSL) provides an effective means of leveraging unlabeled data to improve a model's performance. In this paper, we demonstrate the power of a simple combination of two common SSL methods: consistency regularization and pseudo-labeling. Our algorithm, FixMatch, first generates pseudo-labels using the model's predictions on weakly-augmented unlabeled images. For a given image, the pseudo-label is only retained if the model produces a high-confidence prediction. The model is then trained to predict the pseudo-label when fed a strongly-augmented version of the same image. Despite its simplicity, we show that FixMatch achieves state-of-the-art performance across a variety of standard semi-supervised learning benchmarks, including 94.93% accuracy on CIFAR-10 with 250 labels and 88.61% accuracy with 40 -- just 4 labels per class. Since FixMatch bears many similarities to existing SSL methods that achieve worse performance, we carry out an extensive ablation study to tease apart the experimental factors that are most important to FixMatch's success. We make our code available at https://github.com/google-research/fixmatch.