Institution profile

Beijing Normal University

Academic institutionasia · cn
Official website
Research library528linked papers
Opportunities0open roles
Selected work

Representative Papers

GUIGuard: Toward a General Framework for Privacy-Preserving GUI Agents

Jan 26, 2026

This work addresses the critical privacy risks posed by GUI agents in automation, which often inadvertently leak sensitive information through uploaded interface screenshots, compounded by the lack of systematic approaches to identify and protect privacy across diverse interaction trajectories. To tackle this challenge, we propose GUIGuard—the first end-to-end privacy-preserving framework specifically designed for GUI agents—comprising three integrated stages: privacy identification, protection, and task execution under privacy constraints. We further introduce GUIGuard-Bench, a cross-platform benchmark encompassing 630 interaction trajectories and 13,830 region-level privacy-annotated screenshots. Experimental results reveal that existing agents exhibit alarmingly low privacy recognition accuracy (13.3% on Android and 1.4% on PC), whereas GUIGuard effectively masks sensitive content while preserving task semantics, demonstrating that robust privacy protection can be achieved without compromising task performance.

1 citationsRead paper

Non-Monotonicity in Fair Division of Graphs

Nov 05, 2025

This paper studies fair allocation of graph vertices among multiple agents, where each agent’s valuation of a vertex subset (bundle) is defined as its cut value—the number of edges with exactly one endpoint in the subset—yielding a canonical non-monotonic utility function (marginal utilities may be positive, negative, or zero). Fairness is formalized via EF1 (envy-freeness up to one good), and efficiency is captured by transfer stability (TS). The main contribution is a characterization of the compatibility between EF1 and TS: existence of EF1+TS allocations is non-monotonic in the number of agents—guaranteed for two agents, impossible in some instances with three agents, yet restored for four or more. This pattern is established for general graphs. Furthermore, the paper ensures existence and constructibility of EF1+TS allocations for any number of agents either by relaxing efficiency requirements or by restricting the underlying graph to be a forest.

1 citationsRead paper

LRScheduler: A Layer-aware and Resource-adaptive Container Scheduler in Edge Computing

Jun 04, 2025

To address the high deployment overhead and slow startup latency of containers in bandwidth-constrained edge computing environments, this paper proposes a dual-weight scheduling mechanism that is layer-aware and resource-adaptive. The method innovatively integrates container image layer sharing characteristics with real-time node resource load dynamics, establishing a scheduling framework based on image layer metadata analysis, multi-dimensional node scoring, and dynamically weighted load balancing—deeply embedded within the Kubernetes scheduler. Experimental evaluation demonstrates that, compared to the default Kubernetes scheduler, the approach reduces container image download volume by 37.2%, decreases average container startup time by 29.5%, and improves image layer sharing rate by 3.8×. This work is the first to jointly optimize image layer sharing benefits and real-time resource states, significantly enhancing container deployment efficiency and resource utilization in edge computing scenarios.

1 citationsRead paper

DNAD: Differentiable Neural Architecture Distillation

Apr 25, 2025

Addressing the challenge of balancing accuracy and computational cost in efficient neural network design, this paper proposes Differentiable Neural Architecture Distillation (DNAD), which jointly optimizes accuracy, parameter count, and FLOPs within a non-shared topology cell space. Methodologically, DNAD introduces a novel progressive shrinking mechanism for supernetworks to enable controllable compression of the architecture search space and, for the first time, integrates knowledge distillation into the differentiable search process to mitigate overfitting caused by single-level optimization in DARTS. Pareto-front optimization is employed to automatically generate a set of high-performance architectures achieving balanced multi-objective trade-offs. On ImageNet, the best-performing model achieves a 23.7% top-1 error rate with only 6.0M parameters and 598M FLOPs—substantially outperforming state-of-the-art DARTS variants. Moreover, DNAD yields diverse architectures with consistently lower error rates, fewer parameters, and reduced computational cost on both CIFAR-10 and ImageNet.

1 citationsRead paper

A Paradigm Shift in Neuroscience Driven by Big Data: State of art, Challenges, and Proof of Concept

Dec 08, 2022

Cognitive neuroscience is increasingly fragmented due to divergent methodological practices and research priorities. To address this, we propose “population neuroscience” as a new paradigm grounded in large-scale, multimodal, multi-site cohort data (e.g., UK Biobank), integrating rigorous experimental control, high-dimensional statistical modeling, standardized data harmonization, and interpretable machine learning. This paradigm shifts focus from traditional small-sample causal inference toward large-sample association modeling and mechanistic inference. It establishes, for the first time, a unified analytical framework spanning individual-difference characterization, population-level pattern identification, and within-subject dynamic benchmarking. Empirical validation demonstrates high reproducibility and substantially improved capacity for discovering neurobiological mechanisms underlying behavior.

1 citationsRead paper
Recent publications

Latest Papers

Vocabulary-Guided Gait Recognition

Sep 16, 2026

本文提出了一种词汇引导的步态识别方法Gait-World,通过视觉-语言模型理解步态概念,并设计了词汇关系映射器和步态细粒度检测器来解决模态异质性问题。

0 citationsRead paper