MUGEN: Generating Unlearnable Graph Examples for Multiple Learning Tasks
本文提出MUGEN框架，通过共享GNN编码器和任务特定头生成不可学习的图数据，保护多种下游任务免受未授权学习，利用TASO和TAP方法增强防御效果。
0 citationsRead paper
Selected work
Representative Papers
A Non-Destructive Methodological Framework for Modernizing Legacy Clinical Reporting Systems for AI-Driven Pharmacoinformatics: A SAS Case Study
Legacy clinical reporting systems hinder AI integration and impede drug development and pharmacovigilance due to opaque outputs and the absence of machine-readable intermediate representations. This work proposes a non-intrusive, metadata-driven framework that bridges legacy components—without modifying their source code—through mapping layers, a typed intermediate representation (IR), and coordinator wrappers, thereby transforming their outputs into structured data suitable for large language models (LLMs) and enabling progressive replacement. Validated on 558 SAS components (373k lines of code), the approach achieves immediate AI readiness in coexistence mode, reduces proprietary code by 92% after optional integration, and demonstrates unit-level consistency exceeding 80% across 11 of 14 report types (mean: 82.7%, peak: 99.2%). Five reports achieved 100% compliance on the CDISCPilot01 benchmark, and the framework successfully enabled LLM-driven automated pharmacovigilance, table summarization, and trial configuration generation.
0 citationsRead paper
The Dynamic Gist-Based Memory Model (DGMM): A Memory-Centric Architecture for Artificial Intelligence
Current large language models implicitly encode experience within fixed parameters, limiting their capacity for persistent memory, temporal grounding, source traceability, and interpretability. To address these shortcomings, this work proposes the Dynamic Graph-based Memory Model (DGMM), which explicitly represents time-evolving episodic-semantic memory using a graph structure and incorporates a cue-conditioned recall mechanism to construct working memory. By treating memory as a first-class structural substrate for reasoning, the architecture enables continual evolution without retraining through additive memory growth and recall conditioned on contextual cues. The model supports episodic persistence, context-adaptive behavior, and localized surprise detection, thereby laying the foundation for building AI systems that are interpretable, temporally grounded, and context-aware.
0 citationsRead paper
How Frontier LLMs Adapt to Neurodivergence Context: A Measurement Framework for Surface vs. Structural Change in System-Prompted Responses
This study investigates whether large language models can substantively adapt their outputs in response to neurodiversity (ND)-related context provided in system prompts, distinguishing between superficial and structural adjustments. To this end, we introduce NDBench, a benchmark comprising 576 samples that integrates systematic prompt manipulation, adversarial masking, multidimensional human evaluation, and Krippendorff’s alpha reliability analysis. We propose the first measurement framework for differentiating levels of adaptation and find that merely stating an ND identity is insufficient to suppress harmful outputs. In contrast, explicit instructions significantly increase response length and structural coherence (p < 10⁻⁸) and reduce harmfulness by 36–44% along masking-augmented dimensions. The NDBench resource package is released publicly alongside this work.
0 citationsRead paper
Personalized electric vehicle energy consumption estimation framework that integrates driver behavior with map data
Existing electric vehicle energy consumption prediction methods struggle to simultaneously model individual driving behavior and road-environment context, limiting state-of-charge (SOC) estimation accuracy. This work proposes a high-fidelity SOC prediction framework that integrates map semantics, personalized driving patterns, and a physics-based energy model. By parsing route geometry and road attributes, the method generates reference speeds through rule-based logic, then employs a bidirectional LSTM to forecast individual speed profiles, which drive a PID-controlled vehicle dynamics simulator. A quasi-steady inverse energy model subsequently computes traction power, regenerative braking, and SOC evolution. For the first time, this approach tightly couples learned driving styles with map context and physical modeling, accurately capturing behaviors such as intersection deceleration, speed-limit compliance, and grade response across urban, highway, and hilly scenarios, thereby significantly improving the accuracy of both power demand and SOC trajectory predictions.
0 citationsRead paper
Recent publications
Latest Papers
MUGEN: Generating Unlearnable Graph Examples for Multiple Learning Tasks
本文提出MUGEN框架，通过共享GNN编码器和任务特定头生成不可学习的图数据，保护多种下游任务免受未授权学习，利用TASO和TAP方法增强防御效果。
0 citationsRead paper
A Non-Destructive Methodological Framework for Modernizing Legacy Clinical Reporting Systems for AI-Driven Pharmacoinformatics: A SAS Case Study
Legacy clinical reporting systems hinder AI integration and impede drug development and pharmacovigilance due to opaque outputs and the absence of machine-readable intermediate representations. This work proposes a non-intrusive, metadata-driven framework that bridges legacy components—without modifying their source code—through mapping layers, a typed intermediate representation (IR), and coordinator wrappers, thereby transforming their outputs into structured data suitable for large language models (LLMs) and enabling progressive replacement. Validated on 558 SAS components (373k lines of code), the approach achieves immediate AI readiness in coexistence mode, reduces proprietary code by 92% after optional integration, and demonstrates unit-level consistency exceeding 80% across 11 of 14 report types (mean: 82.7%, peak: 99.2%). Five reports achieved 100% compliance on the CDISCPilot01 benchmark, and the framework successfully enabled LLM-driven automated pharmacovigilance, table summarization, and trial configuration generation.
0 citationsRead paper
The Dynamic Gist-Based Memory Model (DGMM): A Memory-Centric Architecture for Artificial Intelligence
Current large language models implicitly encode experience within fixed parameters, limiting their capacity for persistent memory, temporal grounding, source traceability, and interpretability. To address these shortcomings, this work proposes the Dynamic Graph-based Memory Model (DGMM), which explicitly represents time-evolving episodic-semantic memory using a graph structure and incorporates a cue-conditioned recall mechanism to construct working memory. By treating memory as a first-class structural substrate for reasoning, the architecture enables continual evolution without retraining through additive memory growth and recall conditioned on contextual cues. The model supports episodic persistence, context-adaptive behavior, and localized surprise detection, thereby laying the foundation for building AI systems that are interpretable, temporally grounded, and context-aware.
0 citationsRead paper
How Frontier LLMs Adapt to Neurodivergence Context: A Measurement Framework for Surface vs. Structural Change in System-Prompted Responses
This study investigates whether large language models can substantively adapt their outputs in response to neurodiversity (ND)-related context provided in system prompts, distinguishing between superficial and structural adjustments. To this end, we introduce NDBench, a benchmark comprising 576 samples that integrates systematic prompt manipulation, adversarial masking, multidimensional human evaluation, and Krippendorff’s alpha reliability analysis. We propose the first measurement framework for differentiating levels of adaptation and find that merely stating an ND identity is insufficient to suppress harmful outputs. In contrast, explicit instructions significantly increase response length and structural coherence (p < 10⁻⁸) and reduce harmfulness by 36–44% along masking-augmented dimensions. The NDBench resource package is released publicly alongside this work.
0 citationsRead paper
Personalized electric vehicle energy consumption estimation framework that integrates driver behavior with map data
Existing electric vehicle energy consumption prediction methods struggle to simultaneously model individual driving behavior and road-environment context, limiting state-of-charge (SOC) estimation accuracy. This work proposes a high-fidelity SOC prediction framework that integrates map semantics, personalized driving patterns, and a physics-based energy model. By parsing route geometry and road attributes, the method generates reference speeds through rule-based logic, then employs a bidirectional LSTM to forecast individual speed profiles, which drive a PID-controlled vehicle dynamics simulator. A quasi-steady inverse energy model subsequently computes traction power, regenerative braking, and SOC evolution. For the first time, this approach tightly couples learned driving styles with map context and physical modeling, accurately capturing behaviors such as intersection deceleration, speed-limit compliance, and grade response across urban, highway, and hilly scenarios, thereby significantly improving the accuracy of both power demand and SOC trajectory predictions.
0 citationsRead paper