WSPolypNet: Weakly Supervised Polyp Localization in Colonoscopy Videos
为解决结肠镜视频中息肉定位的密集标注成本问题，提出WSPolypNet框架，仅使用视频级标签通过3D卷积神经网络生成类激活图，并结合多视角策略和MedSAM2细化定位，有效减少空间标注需求。
0 citationsRead paper
Selected work
Representative Papers
Multi-Backbone Self-Supervised Ensembles for Audio Deepfake Detection and a Cross-Track Analysis of Generation-Detection Asymmetry
This work addresses the challenges of poor generalization and performance asymmetry between generation and detection in audio deepfake detection. We propose a multi-backbone self-supervised ensemble framework that integrates representations from WavLM-Large, Wav2Vec2-XLS-R-300M, ECAPA-TDNN, and x-vector, and conducts adversarial exploration using state-of-the-art TTS systems such as F5-TTS and GLM-TTTS. For the first time, we systematically reveal a pronounced asymmetry between generation and detection capabilities, formulate the “architectural insurance” hypothesis, and validate its causal mechanism through a pre-registered falsification experiment. On the ImageCLEF 2026 task, our detector achieves an overall score of 0.9522 (100% accuracy on fake audio, 88.75% on real), while generated speech attains remarkably low WER/CER of 4.99%/2.07%, evading 56.2%–61.4% of detectors. We further report an 11.25% false positive rate on organizer-provided real data, highlighting a critical open issue.
0 citationsRead paper
Adversarial Deepfake Generation and an Investigation of Purification-Based Adversarial Detection
This work addresses the dual challenges of deepfake image generation and detection by proposing a two-stage hot-start, adaptive-weighted multi-detector adversarial attack strategy. High-fidelity identity-preserving face generation is achieved through integration of FLUX.1-dev and PuLID, while an ensemble of PGD attacks—incorporating DiffJPEG-in-loop and momentum/diversity/expectation-over-transformation (MI/DI/EoT) techniques—simultaneously evades 12 state-of-the-art detectors. On the detection front, the authors construct a maximum-probability ensemble model combining SigLIP+DINOv2 and GenD-DINOv3, and for the first time systematically validate the efficacy of logit-difference signals under median filtering in distinguishing diverse adversarial sources, revealing a detection failure at JPEG compression quality threshold Q70. Experiments show 90% and 57.6% evasion rates against organizer- and participant-submitted detectors (score: 0.4170), while the detector achieves 99.4% accuracy on baseline data and AUROC scores of 0.81–0.98 across four adversarial sources after purification (final score: 0.6986).
0 citationsRead paper
Mechanical Conscience: A Mathematical Framework for Dependability of Machine Intelligenc
This work addresses the challenge that locally optimal decisions by individual agents in distributed collaborative intelligent systems often lead to globally non-compliant trajectories under uncertainty. To mitigate this, the authors propose a “mechanical conscience” framework that employs an optimization- and control-theoretic supervisory filter to minimally correct baseline policy actions, ensuring trajectories remain within normatively admissible regions while explicitly modeling epistemic uncertainty. The framework introduces novel metrics—conscience score, mechanical guilt, and resonant reliability—and establishes a trajectory-level theory of normative admissibility, enabling interpretable governance in high-uncertainty, multi-agent settings. Experimental results demonstrate that the approach effectively enforces trajectory compliance in both single- and multi-agent systems, suppresses emergent risks, and outperforms conventional controllers.
0 citationsRead paper
Enhancing Automotive Security with a Hybrid Approach towards Universal Intrusion Detection System
To address the challenges of cross-vehicle model transferability and model degradation caused by firmware updates in automotive intrusion detection systems (IDS), this paper proposes a universal, vehicle-agnostic IDS capable of autonomously adapting to firmware evolution. Methodologically, multi-source in-vehicle signals are first transformed into the frequency domain via wavelet transform to construct a dual-frequency dataset; subsequently, interpretable statistical features are extracted by integrating Pearson correlation analysis with an independent rule system, while deep learning models jointly perform end-to-end feature learning and classification. This work introduces the first hybrid detection framework that synergistically combines statistical robustness with deep representation capability. Evaluated on four real-world vehicle datasets, the proposed method achieves significantly higher accuracy than eight state-of-the-art vehicle-specific IDSs, three general-purpose baselines, and five conventional approaches, demonstrating strong generalization and cross-platform deployability.
0 citationsRead paper
Recent publications
Latest Papers
WSPolypNet: Weakly Supervised Polyp Localization in Colonoscopy Videos
为解决结肠镜视频中息肉定位的密集标注成本问题，提出WSPolypNet框架，仅使用视频级标签通过3D卷积神经网络生成类激活图，并结合多视角策略和MedSAM2细化定位，有效减少空间标注需求。
0 citationsRead paper
Multi-Backbone Self-Supervised Ensembles for Audio Deepfake Detection and a Cross-Track Analysis of Generation-Detection Asymmetry
This work addresses the challenges of poor generalization and performance asymmetry between generation and detection in audio deepfake detection. We propose a multi-backbone self-supervised ensemble framework that integrates representations from WavLM-Large, Wav2Vec2-XLS-R-300M, ECAPA-TDNN, and x-vector, and conducts adversarial exploration using state-of-the-art TTS systems such as F5-TTS and GLM-TTTS. For the first time, we systematically reveal a pronounced asymmetry between generation and detection capabilities, formulate the “architectural insurance” hypothesis, and validate its causal mechanism through a pre-registered falsification experiment. On the ImageCLEF 2026 task, our detector achieves an overall score of 0.9522 (100% accuracy on fake audio, 88.75% on real), while generated speech attains remarkably low WER/CER of 4.99%/2.07%, evading 56.2%–61.4% of detectors. We further report an 11.25% false positive rate on organizer-provided real data, highlighting a critical open issue.
0 citationsRead paper
Adversarial Deepfake Generation and an Investigation of Purification-Based Adversarial Detection
This work addresses the dual challenges of deepfake image generation and detection by proposing a two-stage hot-start, adaptive-weighted multi-detector adversarial attack strategy. High-fidelity identity-preserving face generation is achieved through integration of FLUX.1-dev and PuLID, while an ensemble of PGD attacks—incorporating DiffJPEG-in-loop and momentum/diversity/expectation-over-transformation (MI/DI/EoT) techniques—simultaneously evades 12 state-of-the-art detectors. On the detection front, the authors construct a maximum-probability ensemble model combining SigLIP+DINOv2 and GenD-DINOv3, and for the first time systematically validate the efficacy of logit-difference signals under median filtering in distinguishing diverse adversarial sources, revealing a detection failure at JPEG compression quality threshold Q70. Experiments show 90% and 57.6% evasion rates against organizer- and participant-submitted detectors (score: 0.4170), while the detector achieves 99.4% accuracy on baseline data and AUROC scores of 0.81–0.98 across four adversarial sources after purification (final score: 0.6986).
0 citationsRead paper
Mechanical Conscience: A Mathematical Framework for Dependability of Machine Intelligenc
This work addresses the challenge that locally optimal decisions by individual agents in distributed collaborative intelligent systems often lead to globally non-compliant trajectories under uncertainty. To mitigate this, the authors propose a “mechanical conscience” framework that employs an optimization- and control-theoretic supervisory filter to minimally correct baseline policy actions, ensuring trajectories remain within normatively admissible regions while explicitly modeling epistemic uncertainty. The framework introduces novel metrics—conscience score, mechanical guilt, and resonant reliability—and establishes a trajectory-level theory of normative admissibility, enabling interpretable governance in high-uncertainty, multi-agent settings. Experimental results demonstrate that the approach effectively enforces trajectory compliance in both single- and multi-agent systems, suppresses emergent risks, and outperforms conventional controllers.
0 citationsRead paper
Enhancing Automotive Security with a Hybrid Approach towards Universal Intrusion Detection System
To address the challenges of cross-vehicle model transferability and model degradation caused by firmware updates in automotive intrusion detection systems (IDS), this paper proposes a universal, vehicle-agnostic IDS capable of autonomously adapting to firmware evolution. Methodologically, multi-source in-vehicle signals are first transformed into the frequency domain via wavelet transform to construct a dual-frequency dataset; subsequently, interpretable statistical features are extracted by integrating Pearson correlation analysis with an independent rule system, while deep learning models jointly perform end-to-end feature learning and classification. This work introduces the first hybrid detection framework that synergistically combines statistical robustness with deep representation capability. Evaluated on four real-world vehicle datasets, the proposed method achieves significantly higher accuracy than eight state-of-the-art vehicle-specific IDSs, three general-purpose baselines, and five conventional approaches, demonstrating strong generalization and cross-platform deployability.
0 citationsRead paper