Debiased Self-Training for Semi-Supervised Learning
This addresses the challenge of reducing labeled data requirements for deep neural networks, offering a method to stabilize and improve semi-supervised learning across various tasks, though it is incremental as it builds on existing self-training approaches.
The paper tackles the problem of bias and instability in self-training for semi-supervised learning by proposing Debiased Self-Training (DST), which decouples pseudo-label generation and utilization and adversarially optimizes representations to avoid worst-case errors, resulting in an average improvement of 6.3% against state-of-the-art methods on benchmark datasets and 18.9% against FixMatch on 13 diverse tasks.
Deep neural networks achieve remarkable performances on a wide range of tasks with the aid of large-scale labeled datasets. Yet these datasets are time-consuming and labor-exhaustive to obtain on realistic tasks. To mitigate the requirement for labeled data, self-training is widely used in semi-supervised learning by iteratively assigning pseudo labels to unlabeled samples. Despite its popularity, self-training is well-believed to be unreliable and often leads to training instability. Our experimental studies further reveal that the bias in semi-supervised learning arises from both the problem itself and the inappropriate training with potentially incorrect pseudo labels, which accumulates the error in the iterative self-training process. To reduce the above bias, we propose Debiased Self-Training (DST). First, the generation and utilization of pseudo labels are decoupled by two parameter-independent classifier heads to avoid direct error accumulation. Second, we estimate the worst case of self-training bias, where the pseudo labeling function is accurate on labeled samples, yet makes as many mistakes as possible on unlabeled samples. We then adversarially optimize the representations to improve the quality of pseudo labels by avoiding the worst case. Extensive experiments justify that DST achieves an average improvement of 6.3% against state-of-the-art methods on standard semi-supervised learning benchmark datasets and 18.9%$ against FixMatch on 13 diverse tasks. Furthermore, DST can be seamlessly adapted to other self-training methods and help stabilize their training and balance performance across classes in both cases of training from scratch and finetuning from pre-trained models.