18 January 2024 in IEEE Transactions on Pattern Analysis and Machine Intelligence
Label-noise learning (LNL) aims to increase the model's generalization given training data with noisy labels. To facilitate practical LNL algorithms, researchers have proposed different label noise types, ranging from class-conditional to instance-dependent noises. In this paper, we introduce a novel label noise type called BadLabel , which can significantly degrade the performance of existing LNL algorithms by a large margin. BadLabel is crafted based on the label-flipping attack against standard classification, where specific samples are selected and their labels are flipped to other labels so that the loss values of clean and noisy labels become indistinguishable. To address the challenge posed by BadLabel, we further propose a robust LNL method that perturbs the labels in an adversarial manner at each epoch to make the loss values of clean and noisy labels again distinguishable. Once we select a small set of (mostly) clean labeled data, we can apply the techniques of semi-supervised learning to train the model accurately. Empirically, our experimental results demonstrate that existing LNL algorithms are vulnerable to the newly introduced BadLabel noise type, while our proposed robust LNL method can effectively improve the generalization performance of the model under various types of label noise. The new dataset of noisy labels and the source codes of robust LNL algorithms are available at https://github.com/zjfheart/BadLabels .
2 November 2023 in Neural Networks
In the transfer learning paradigm, models that are pre-trained on large datasets are used as the foundation models for various downstream tasks. However, this paradigm exposes downstream practitioners to data poisoning threats, as attackers can inject malicious samples into the re-training datasets to manipulate the behavior of models in downstream tasks. In this work, we propose a defense strategy that significantly reduces the success rate of various data poisoning attacks in downstream tasks. Our defense aims to pre-train a robust foundation model by reducing adversarial feature distance and increasing inter-class feature distance. Experiments demonstrate the excellent defense performance of the proposed strategy towards state-of-the-art clean-label poisoning attacks in the transfer learning scenario.
9 February 2024 in IEEE Transactions on Artificial Intelligence
Adversarial training is an effective method to improve the model’s adversarial robustness. To realize a considerable tradeoff between clean accuracy and adversarial robustness, surrogate loss minimization can be utilized to regularize for robustness. This study advances beyond previous research efforts by taking a further step. It is informed by two theoretical perspectives. First, adversarial examples are inevitable within a unit sphere surrounding clean data. Second, geodesics can characterize the shortest distance between two points in Riemannian geometry. Accordingly, this study employs geodesic distance as a regularized surrogate loss item to capture the minimal divergence between the distribution of natural examples and the distribution of adversarial examples. This approach yields a more compact upper bound of risk errors than previous studies, which is beneficial for improving adversarial robustness. Based on the theoretical insight, this study proposes a metric of Geodesic Loss and a framework of Geodesic Adversarial Training to boost the adversarial robustness of neural networks. The empirical study on the diverse datasets demonstrates the considerable performance of the proposed method against diverse attacks, including white-box attacks, black-box corruptions, adaptive attacks, and auto attacks. Our code is available at: https://github.com/momo1986/GeodesicAdversarialTraining .
16 October 2024 in IEEE Transactions on Geoscience and Remote Sensing
Accurate recognition of marine species in drone-captured images is essential in maintaining the stability of coastal zone ecosystems. Unmanned Aerial Vehicle (UAV) remote sensing images usually lack paired supervised signals and suffer from color distortion and blurring due to the interaction of ambient light with cross-medium transmission between air and water. However, current algorithms mainly focus on supervised training methods and also ignore the interaction involved in the cross-medium transmission of light in water. In this paper, for UAV in coastal zones, we propose an unsupervised image water removal model-DewaterGAN, which is based solely on low-tide and high-tide images without paired supervised signals and also preserves color and texture in the water removal process. Specifically, our approach involves two key steps: an unsupervised training CycleGAN network accomplishes domain transitions from low-tide level to high-tide level, and a physics-based attention module guides image water removal and maintains authenticity. Additionally, we utilize evaluation metrics of image restoration PSNR and SSIM to quantitatively analyze the performance of the model. We also employed several non-reference metrics (UIQM, UCIQE, NIQE, BRISQUE, LIQE, ILNIQE and CLIPIQA) to evaluate the visual quality of the image de-watering process. Extensive experiments conducted on both our water removal dataset and public datasets validate the efficacy of our model. The code is at: https://github.com/yfq-yy/Dewater.git.