Institution profile

University of North Carolina at Chapel Hill

Academic institutionnorthamerica · us
Official website
Research library744linked papers
Opportunities0open roles
Selected work

Representative Papers

Unlearning Sensitive Information in Multimodal LLMs: Benchmark and Attack-Defense Evaluation

May 01, 2025

Multimodal large language models (MLLMs) often inadvertently memorize sensitive information—such as personally identifiable data or harmful content—during training, and multimodal prompts can be exploited by adversaries to extract such knowledge. Yet, systematic evaluation of multimodal forgetting remains absent. Method: We introduce UnLOK-VQA, the first benchmark for targeted sensitive-knowledge forgetting in MLLMs, built upon high-quality, human-curated image–text pairs. We design both white-box and black-box multimodal extraction attacks and propose a white-box forgetting method grounded in hidden-state interpretability. Contribution/Results: We find that erasing answer-related hidden-state representations is the most effective defense, and model scale positively correlates with post-forgetting robustness. Experiments demonstrate that multimodal attacks significantly outperform unimodal ones, establishing a foundational framework for secure forgetting research in MLLMs.

3 citationsRead paper

Moments by Integrating the Moment-Generating Function

Oct 31, 2024

Conventional moment computation relies on analytic derivatives of probability density functions (PDFs) or moment-generating functions (MGFs), rendering it inapplicable to fractional, complex-order, and central/non-central moments when PDFs are unavailable or MGF derivatives are intractable. Method: This paper proposes the Complex-extended Moment Generating Function (CMGF) integration method—a novel framework that requires only integrability of the MGF over a contour in the complex plane. By leveraging analytic continuation and numerical contour integration, CMGF directly computes arbitrary real-order, complex-order, absolute, central, and non-central moments without invoking PDFs or MGF derivatives. Contribution/Results: As the first general-purpose MGF-based moment computation framework relying on integration rather than differentiation, CMGF is validated across three canonical scenarios where closed-form MGFs exist but PDFs or their derivatives do not. It significantly simplifies high-order and non-integer moment evaluation, enhancing efficiency in statistical inference and stochastic modeling.

2 citations1 influentialRead paper

GeoDynamics: A Geometric State-Space Neural Network for Understanding Brain Dynamics on Riemannian Manifolds

Jan 20, 2026

This work proposes a geometric state-space neural network that integrates state-space modeling with Riemannian geometry to capture the intrinsic dynamics of brain functional connectivity. Existing approaches often neglect the Riemannian manifold structure of symmetric positive-definite (SPD) functional connectivity matrices, thereby failing to characterize the geometric nature of the brain as a self-organizing system. By directly modeling the temporal evolution of functional connectivity on the SPD manifold, the proposed method establishes a manifold-aware recurrent framework that enables smooth and geometrically consistent tracking of high-dimensional brain state trajectories. Experiments demonstrate that the model accurately captures task-evoked brain state transitions and effectively identifies early biomarkers for Alzheimer’s disease, Parkinson’s disease, and autism spectrum disorder. Furthermore, it exhibits strong generalization performance across multiple action recognition benchmarks.

2 citationsRead paper

Learning Covariance-Based Multi-Scale Representation of Neuroimaging Measures for Alzheimer Classification

Apr 18, 2023IEEE International Symposium on Biomedical Imaging

To address over-parameterization and uncertainty in deep networks caused by limited samples in early-stage Alzheimer’s disease (AD) neuroimaging classification, this paper proposes a covariance-driven multi-scale scale-space representation framework. Methodologically, it innovatively integrates covariance modeling and scale-space theory into the neural network backbone, enabling compact representation of high-dimensional brain images and dual-space disentanglement of features and tasks. The framework incorporates scale-space convolution, covariance-based feature extraction, and multi-scale fusion, while supporting gradient-weighted, individualized localization of AD-affected brain regions. Evaluated on the ADNI dataset, the model achieves significantly improved classification accuracy, accelerated convergence, and a substantial reduction in parameter count. Crucially, it retains strong interpretability—precisely identifying AD-specific atrophic brain regions such as the hippocampus and entorhinal cortex.

2 citationsRead paper

Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing

Feb 03, 2026

This work proposes Parallel-Probe, a novel approach to parallel inference that addresses the high computational cost and the underutilization of global branch dynamics in existing methods. By introducing a 2D probing mechanism that periodically samples intermediate outputs from all branches, the study uncovers key patterns—including non-monotonic width-depth trade-offs, heterogeneous branch lengths, and the early stabilization of global consensus. Leveraging these insights, the authors design a training-free online controller that dynamically adjusts inference depth via a consensus-based early-exit strategy and modulates width through bias-driven branch pruning. Experiments across three benchmarks and multiple models demonstrate that Parallel-Probe significantly outperforms standard majority voting, reducing sequential token consumption by up to 35.8% and total token cost by up to 25.8%, while maintaining competitive accuracy.

1 citationsRead paper
Recent publications

Latest Papers

Designing Signals for Deterrence

Sep 13, 2026

本文提出了一种新的信号设计方法,通过预先承诺根据实际决心采取成本行动,以在有限观众成本下提高防御方收益并降低战争可能性。

0 citationsRead paper