Selected work

Representative Papers

DuoAD: Leveraging [CLS] Dual Characteristics for Training-Free Few-Shot Anomaly Detection

Jul 26, 2026

This work proposes a fully automatic, training-free few-shot anomaly detection framework that addresses the limitation of existing methods, which predominantly rely on local image patch features while overlooking the global contextual information embedded in the [CLS] token of Vision Transformers. The study is the first to reveal and exploit the dual nature of the [CLS] token: its global semantic invariance and its attention map’s ability to indicate spatial anomalies. By integrating a semantic consistency-driven automatic augmentation strategy with an attention-guided dynamic feature reweighting mechanism, the method achieves precise anomaly localization and scoring without manual hyperparameter tuning. Under single-sample settings on MVTec-AD, VisA, and Real-IAD, it attains Image-AUC scores of 97.7%, 93.2%, and 84.5%, respectively, demonstrating plug-and-play state-of-the-art performance across categories, backbone architectures, and datasets.

0 citationsRead paper

The Binding Effect: Analyzing How Multi-Dimensional Cues Form Gender Bias in Instruction TTS

Mar 21, 2026

This study addresses the limitations of existing gender bias evaluations in instructional text-to-speech (ITTS) systems, which often rely on univariate tests and fail to capture the combinatorial effects of social cues. The authors propose a multidimensional prompting framework that systematically integrates social status, occupational stereotypes, and role descriptors, revealing for the first time a binding effect among these dimensions in ITTS outputs. This binding effect indicates that bias arises from a deep coupling between semantic priors embedded in pretrained text encoders and the distributional properties of training data. Through analyses of open-source models, semantic probing, and diversity intervention experiments, the work demonstrates that generic diversity prompts are insufficient to mitigate such entrenched biases, underscoring the necessity of compositional analysis for diagnosing latent risks in synthetic speech and establishing a critical link between semantic priors in pretrained encoders and biased voice generation.

0 citationsRead paper

Feasibility-Guided Planning over Multi-Specialized Locomotion Policies

Feb 08, 2026

This work proposes a feasibility-guided path planning framework to address the challenge of coordinating multiple specialized locomotion strategies over unstructured terrain. The approach equips each terrain-specific strategy with a lightweight Feasibility-Net that predicts a feasibility tensor from local elevation maps and task vectors, thereby guiding classical planning algorithms to generate optimal paths consistent with the capabilities of the selected strategy. The framework supports plug-and-play integration of new strategies without retraining, while preserving both interpretability and strategy consistency. Experimental results in both simulation and real-world environments demonstrate that the method efficiently produces reliable paths and significantly enhances adaptability to complex and diverse terrains.

0 citationsRead paper

Dual-Process Scaffold Reasoning for Enhancing LLM Code Debugging

Nov 11, 2025

To address the high computational complexity and low efficiency of System 2 reasoning in large language model (LLM)-based code debugging, this paper proposes Scaffold-Reasoning—a dual-process theory–inspired scaffolding reasoning framework. It comprises three synergistic modules: Scaffold (structural guidance), Analytic (fine-grained error attribution analysis), and Integration (multi-stream logical fusion). By mapping cognitive psychology’s dual-system theory into a computationally tractable reasoning architecture, our approach balances reasoning depth and efficiency. Methodologically, it employs stepwise reference generation, precise error localization, and multi-path ensemble inference. Evaluated on the DebugBench benchmark, Scaffold-Reasoning achieves an 88.91% pass rate with an average solving time of only 5.36 seconds per problem—significantly outperforming state-of-the-art methods. These results empirically validate the effectiveness and practicality of cognition-aligned design for LLM-based code debugging.

0 citationsRead paper

How Bias Binds: Measuring Hidden Associations for Bias Control in Text-to-Image Compositions

Nov 10, 2025

In text-to-image generation, implicit bias amplification arises from semantic binding between objects and sensitive attributes within prompts; existing debiasing methods typically address single-object, isolated-attribute scenarios, neglecting the joint contextual effects of such bindings. This work introduces the “bias attachment score” — the first metric quantifying the activation strength of specific object–attribute bindings toward bias. Building upon this, we propose a context-aware, model-agnostic debiasing framework that explicitly decouples semantic bindings via prompt disentanglement and associative modeling—without requiring model retraining. Experiments demonstrate over 10% improvement in debiasing performance on composite prompt generation tasks. Our findings reveal the pivotal role of semantic binding in bias amplification and expose a fundamental limitation of current debiasing approaches in complex, context-rich settings.

0 citationsRead paper

Recent publications

Latest Papers

DuoAD: Leveraging [CLS] Dual Characteristics for Training-Free Few-Shot Anomaly Detection

Jul 26, 2026

This work proposes a fully automatic, training-free few-shot anomaly detection framework that addresses the limitation of existing methods, which predominantly rely on local image patch features while overlooking the global contextual information embedded in the [CLS] token of Vision Transformers. The study is the first to reveal and exploit the dual nature of the [CLS] token: its global semantic invariance and its attention map’s ability to indicate spatial anomalies. By integrating a semantic consistency-driven automatic augmentation strategy with an attention-guided dynamic feature reweighting mechanism, the method achieves precise anomaly localization and scoring without manual hyperparameter tuning. Under single-sample settings on MVTec-AD, VisA, and Real-IAD, it attains Image-AUC scores of 97.7%, 93.2%, and 84.5%, respectively, demonstrating plug-and-play state-of-the-art performance across categories, backbone architectures, and datasets.

0 citationsRead paper

The Binding Effect: Analyzing How Multi-Dimensional Cues Form Gender Bias in Instruction TTS

Mar 21, 2026

This study addresses the limitations of existing gender bias evaluations in instructional text-to-speech (ITTS) systems, which often rely on univariate tests and fail to capture the combinatorial effects of social cues. The authors propose a multidimensional prompting framework that systematically integrates social status, occupational stereotypes, and role descriptors, revealing for the first time a binding effect among these dimensions in ITTS outputs. This binding effect indicates that bias arises from a deep coupling between semantic priors embedded in pretrained text encoders and the distributional properties of training data. Through analyses of open-source models, semantic probing, and diversity intervention experiments, the work demonstrates that generic diversity prompts are insufficient to mitigate such entrenched biases, underscoring the necessity of compositional analysis for diagnosing latent risks in synthetic speech and establishing a critical link between semantic priors in pretrained encoders and biased voice generation.

0 citationsRead paper

Feasibility-Guided Planning over Multi-Specialized Locomotion Policies

Feb 08, 2026

This work proposes a feasibility-guided path planning framework to address the challenge of coordinating multiple specialized locomotion strategies over unstructured terrain. The approach equips each terrain-specific strategy with a lightweight Feasibility-Net that predicts a feasibility tensor from local elevation maps and task vectors, thereby guiding classical planning algorithms to generate optimal paths consistent with the capabilities of the selected strategy. The framework supports plug-and-play integration of new strategies without retraining, while preserving both interpretability and strategy consistency. Experimental results in both simulation and real-world environments demonstrate that the method efficiently produces reliable paths and significantly enhances adaptability to complex and diverse terrains.

0 citationsRead paper

Dual-Process Scaffold Reasoning for Enhancing LLM Code Debugging

Nov 11, 2025

To address the high computational complexity and low efficiency of System 2 reasoning in large language model (LLM)-based code debugging, this paper proposes Scaffold-Reasoning—a dual-process theory–inspired scaffolding reasoning framework. It comprises three synergistic modules: Scaffold (structural guidance), Analytic (fine-grained error attribution analysis), and Integration (multi-stream logical fusion). By mapping cognitive psychology’s dual-system theory into a computationally tractable reasoning architecture, our approach balances reasoning depth and efficiency. Methodologically, it employs stepwise reference generation, precise error localization, and multi-path ensemble inference. Evaluated on the DebugBench benchmark, Scaffold-Reasoning achieves an 88.91% pass rate with an average solving time of only 5.36 seconds per problem—significantly outperforming state-of-the-art methods. These results empirically validate the effectiveness and practicality of cognition-aligned design for LLM-based code debugging.

0 citationsRead paper

How Bias Binds: Measuring Hidden Associations for Bias Control in Text-to-Image Compositions

Nov 10, 2025

In text-to-image generation, implicit bias amplification arises from semantic binding between objects and sensitive attributes within prompts; existing debiasing methods typically address single-object, isolated-attribute scenarios, neglecting the joint contextual effects of such bindings. This work introduces the “bias attachment score” — the first metric quantifying the activation strength of specific object–attribute bindings toward bias. Building upon this, we propose a context-aware, model-agnostic debiasing framework that explicitly decouples semantic bindings via prompt disentanglement and associative modeling—without requiring model retraining. Experiments demonstrate over 10% improvement in debiasing performance on composite prompt generation tasks. Our findings reveal the pivotal role of semantic binding in bias amplification and expose a fundamental limitation of current debiasing approaches in complex, context-rich settings.

0 citationsRead paper