Erasing the Bias: Fine-Tuning Foundation Models for Semi-Supervised Learning
This work addresses the problem of subpar performance in semi-supervised learning for practitioners, offering a novel solution with broad applicability.
The paper tackles the challenge of deploying semi-supervised learning methods by introducing FineSSL, which adapts pre-trained foundation models to address biases and cognitive deviations, resulting in new state-of-the-art performance on multiple benchmarks and a sixfold reduction in training cost.
Semi-supervised learning (SSL) has witnessed remarkable progress, resulting in the emergence of numerous method variations. However, practitioners often encounter challenges when attempting to deploy these methods due to their subpar performance. In this paper, we present a novel SSL approach named FineSSL that significantly addresses this limitation by adapting pre-trained foundation models. We identify the aggregated biases and cognitive deviation problems inherent in foundation models, and propose a simple yet effective solution by imposing balanced margin softmax and decoupled label smoothing. Through extensive experiments, we demonstrate that FineSSL sets a new state of the art for SSL on multiple benchmark datasets, reduces the training cost by over six times, and can seamlessly integrate various fine-tuning and modern SSL algorithms. The source code is available at https://github.com/Gank0078/FineSSL.