Semi-Supervised Variational Adversarial Active Learning via Learning to Rank and Agreement-Based Pseudo Labeling

TL;DR

This work addresses the cost of labeling in deep learning by enhancing pool-based active learning with semi-supervised training signals. It introduces SS-VAAL, which combines clustering-assisted pseudo labeling to robustly utilize unlabeled data with a learning-to-rank loss predictor that embeds task information into a latent space via a VAE–discriminator framework. The method demonstrates consistent gains over VAAL, TA-VAAL, and MAOAL on image classification benchmarks including CIFAR-10/100, Caltech-101, and ImageNet, as well as semantic segmentation on Cityscapes, and provides extensive ablations and an ImageNet result analysis. Overall, SS-VAAL enables more effective target-task learning by leveraging unlabeled data throughout training and by explicitly ranking predicted losses, leading to improved sample efficiency and performance.

Abstract

Active learning aims to alleviate the amount of labor involved in data labeling by automating the selection of unlabeled samples via an acquisition function. For example, variational adversarial active learning (VAAL) leverages an adversarial network to discriminate unlabeled samples from labeled ones using latent space information. However, VAAL has the following shortcomings: (i) it does not exploit target task information, and (ii) unlabeled data is only used for sample selection rather than model training. To address these limitations, we introduce novel techniques that significantly improve the use of abundant unlabeled data during training and take into account the task information. Concretely, we propose an improved pseudo-labeling algorithm that leverages information from all unlabeled data in a semi-supervised manner, thus allowing a model to explore a richer data space. In addition, we develop a ranking-based loss prediction module that converts predicted relative ranking information into a differentiable ranking loss. This loss can be embedded as a rank variable into the latent space of a variational autoencoder and then trained with a discriminator in an adversarial fashion for sample selection. We demonstrate the superior performance of our approach over the state of the art on various image classification and segmentation benchmark datasets.

Figures (15)

• Figure 1: An overview of SS-VAAL. First, a loss prediction module attached to the target model predicts losses on the input data. Next, the predicted losses along with the actual target losses are transformed into ranking losses via a pretrained ranking function. Unlabeled samples are then passed to the target model and subsequently through a k-means algorithm to acquire pseudo labels for additional training. Finally, a discriminator following a variational autoencoder is trained in an adversarial manner to select unlabeled samples for annotation.
• Figure 2: The detailed architecture of SS-VAAL. (Stage 1) A loss prediction module is attached to the target model to predict losses on the input data. These predicted losses, along with the actual losses obtained from the target model, are transformed into ranking losses via a pretrained ranking function. Features of the labeled samples are extracted from the target model to fit a k-means algorithm. (Stage 2) Unlabeled samples are processed through the target model to obtain initial pseudo labels. The k-means algorithm, already fit with labeled features, is also applied to the unlabeled samples to obtain clustering labels for them. Initial pseudo and clustering labels are combined to determine the final pseudo labels. These unlabeled samples and their pseudo labels are then used for additional training of the target model. (Stage 3) Both labeled and unlabeled samples are fed into an encoder network to learn the latent variables. The learned and rank variables are trained adversarially with a discriminator. Sample selection is based on the predicted probability from the discriminator.
• Figure 3: An overview of the SoDeep sorter architecture. A pretrained differentiable DNN sorter converts the raw scores into ranks given by the target model. A loss is then applied to the predicted rank, backpropagated through the differentiable sorter, and used to update the weights.
• Figure 4: Image classification comparison on the (a) CIFAR-10, (b) CIFAR-100, and (c) Caltech-101 datasets.
• Figure 5: Semantic segmentation results on the Cityscapes dataset.