Author name cluster

Kaibo Wang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

6 papers

1 author row

TIST Journal 2026 Journal Article

Multi-Stage Robust Federated Learning: Addressing Label Noise under Data Heterogeneity and Imbalance

Kaibo Wang
Anqi Zhang
Tangyou Liu
Wenqian Zhang
Guanglin Zhang

Federated Learning (FL) enables collaborative model training while preserving data privacy, but the presence of noisy labels in local datasets remains a significant challenge, particularly under heterogeneous noise conditions and class imbalance. In this work, we introduce a novel Multi-Stage Robust Federated Learning (MRFL) framework to address these issues. In the warm-up noise detection stage, MRFL computes per-class average losses on each client and employs a Gaussian mixture model to accurately identify clients with substantial label noise. In the subsequent noise-robust training stage, a robust loss function and noise solver are designed to distinguish clean from noisy samples, while semi-supervised learning is used to recover valuable information from tail classes. Moreover, a robust weighted aggregation strategy is adopted to mitigate the adverse effects of noisy clients. Extensive experiments on CIFAR-10/100-LT and ICH datasets demonstrate that MRFL outperforms state-of-the-art methods in federated noisy label learning scenarios characterized by data heterogeneity and imbalance.

Details DOI

AAAI Conference 2026 Conference Paper

TweezeEdit: Consistent and Efficient Image Editing with Path Regularization

Jianda Mao
Kaibo Wang
Yang Xiang
Kani Chen

Recent progress in training-free image editing has enabled existing text-to-image diffusion models to be directly adapted into text-guided image editors without additional training. However, existing methods often over-align with target prompts while inadequately preserving source image semantics. These approaches generate target images explicitly or implicitly from the inversion noise of the source images, termed the inversion anchors. We identify this strategy as suboptimal for semantic preservation and inefficient due to elongated editing paths. We propose TweezeEdit, a tuning- and inversion-free framework for consistent and efficient image editing. Our method addresses these limitations by regularizing the entire denoising path rather than relying solely on the inversion anchors, ensuring source semantic retention and shortening editing paths. Guided by gradient-driven regularization, we efficiently inject target prompt semantics along a direct path using a consistency model. Extensive experiments demonstrate TweezeEdit's superior performance in semantic preservation and target alignment, outperforming existing methods. Remarkably, it requires only 12 steps (1.6 seconds per edit), underscoring its potential for real-time applications. The appendix is available in the extended version.

PDF Details DOI

EAAI Journal 2025 Journal Article

Ship re-identification in foggy weather: A two-branch network with dynamic feature enhancement and dual attention

Wei Sun
Fei Guan
Xiaorui Zhang
Xinyi Shen
Kaibo Wang

Most existing ship re-identification methods are only applicable under sunny days. However, the presence of fog makes ship re-identification difficult. Therefore, this study proposes a two-branch network with dynamic feature enhancement and dual attention, which includes the defogging and re-identification branch, employing an end-to-end training approach to simultaneously learn defogging and re-identification tasks. To address issues related to blurring images and changes in ship appearance caused by fog, this study proposes a dynamic feature enhancement module, utilizing the deformable group convolution to adaptively adjust the shapes and positions of kernels. The adjustment enables better matching with ship actual outlines and size changes, thereby accurately locating salient regions in ship images. To further improve the network’s ability in capturing ship details in foggy weather and recover clear and fog-free images, a dual attention module fusing multiple pooling operations and using convolution kernels of different sizes is proposed, which strengthens the network’s attention to capture ship characteristics in different channels and spaces. To enhance the discrimination of learned ship vital attributes, this study introduces the adaptive sparse pair loss, dynamically adjusting the network’s attention to positive and negative sample pairs with different attributes. In addition, a new dataset is constructed specifically for foggy ship re-identification. The atmospheric scattering model is employed for the synthesis of foggy ship images, effectively addressing the challenge associated with training and evaluating foggy ship re-identification networks. Experimental results show that the proposed method outperforms existing ship re-identification methods in foggy weather.

Details DOI

NeurIPS Conference 2025 Conference Paper

Towards a Golden Classifier-Free Guidance Path via Foresight Fixed Point Iterations

Kaibo Wang
Jianda Mao
Tong Wu
Yang Xiang

Classifier-Free Guidance (CFG) is an essential component of text-to-image diffusion models, and understanding and advancing its operational mechanisms remains a central focus of research. Existing approaches stem from divergent theoretical interpretations, thereby limiting the design space and obscuring key design choices. To address this, we propose a unified perspective that reframes conditional guidance as fixed point iterations, seeking to identify a golden path where latents produce consistent outputs under both conditional and unconditional generation. We demonstrate that CFG and its variants constitute a special case of single-step short-interval iteration, which is theoretically proven to exhibit inefficiency. To this end, we introduce Foresight Guidance (FSG), which prioritizes solving longer-interval subproblems in early diffusion stages with increased iterations. Extensive experiments across diverse datasets and model architectures validate the superiority of FSG over state-of-the-art methods in both image quality and computational efficiency. Our work offers novel perspectives for conditional guidance and unlocks the potential of adaptive design.

PDF Details

NeurIPS Conference 2024 Conference Paper

DiffHammer: Rethinking the Robustness of Diffusion-Based Adversarial Purification

Kaibo Wang
Xiaowen Fu
Yuxuan Han
Yang Xiang

Diffusion-based purification has demonstrated impressive robustness as an adversarial defense. However, concerns exist about whether this robustness arises from insufficient evaluation. Our research shows that EOT-based attacks face gradient dilemmas due to global gradient averaging, resulting in ineffective evaluations. Additionally, 1-evaluation underestimates resubmit risks in stochastic defenses. To address these issues, we propose an effective and efficient attack named DiffHammer. This method bypasses the gradient dilemma through selective attacks on vulnerable purifications, incorporating $N$-evaluation into loops and using gradient grafting for comprehensive and efficient evaluations. Our experiments validate that DiffHammer achieves effective results within 10-30 iterations, outperforming other methods. This calls into question the reliability of diffusion-based purification after mitigating the gradient dilemma and scrutinizing its resubmit risk.

PDF Details DOI

EAAI Journal 2023 Journal Article

Parallel crosschecking neural network based fault-tolerant flight parameter estimation and faulty sensor identification

Wanyong Zou
Ni Li
Ban Wang
Kaibo Wang
Shuhui Bu
Ming Zhou
He Shen

Reliable flight parameter estimation is crucial for the safe operation of unmanned air vehicles. Recent studies on flight parameter estimation based on distributed airflow sensors have shown promising results; however, how to extract reliable flight parameters and locate faults in case of sensor failures are not studied yet. This paper proposes a novel fault-tolerant flight parameter estimation method with capabilities to identify and isolate faulty sensors for a flight parameter estimation system based on distributed airflow sensors. First, the distributed sensors are grouped into subsets using combinations. Second, feedforward backpropagation neural networks of the same structure are designed and used by all subsets for flight parameter estimation. The crosschecking between these parallel networks improves the robustness of the estimation. In case of sensor failures, the sensor groups that contain any faulty sensors will result in wrong flight parameter estimations. Especially, a directional random sample consensus algorithm is developed and used to identify the subsets that contain faulty sensors. Then, the ratio of the faulty subsets is used to identify the number of faulty sensors. Finally, the faulty sensors will be identified according to the frequencies of their appearance in the subsets that contain faulty sensors. The effectiveness of the proposed method is validated by using wind tunnel tests. The results show that the proposed method can effectively provide reliable flight parameters and identify faulty sensors.

Details DOI