Selected work

Representative Papers

Constructing self-referential instances for the clique problem

Jan 27, 2026

This study investigates the intrinsic computational hardness of the clique problem in its critical region. By constructing pairs of graphs at the phase transition point of the Erdős–Rényi random graph model that share identical numbers of vertices, edges, and degree sequences yet exhibit opposite solution statuses—i.e., one contains a k-clique while the other does not—the authors employ degree-preserving symmetric transformations to reveal the indistinguishability of their solution spaces. This work presents the first family of self-referential instances that rigorously establishes the existence of an exact phase transition threshold for the clique problem. Theoretically, it demonstrates that near this threshold, any algorithm must almost surely explore the entire solution space to determine the existence of a solution, thereby explaining the unavoidable necessity of exhaustive search in this regime.

1 citationsRead paper

Placement Delivery Array for Cache-Aided MIMO Systems

Jan 15, 2026

This work addresses the challenge of simultaneously achieving the optimal sum degrees of freedom (sum-DoF) and low subpacketization in cache-aided MIMO systems. To this end, the authors propose a unified combinatorial structure—termed the MIMO Placement Delivery Array (MIMO-PDA)—to characterize uncoded cache placement and single-slot zero-forcing transmission strategies. Leveraging this framework, they derive the first tight upper bound on the achievable sum-DoF and introduce two novel construction methods. These constructions attain the optimal sum-DoF while significantly reducing subpacketization: one achieves linear reduction and the other exponential reduction under different system parameter regimes. Notably, the second construction operates under more relaxed parameter constraints than existing schemes, substantially lowering subpacketization complexity while still achieving the theoretical sum-DoF optimum.

1 citationsRead paper

A Low-Complexity Architecture for Multi-access Coded Caching Systems with Arbitrary User-cache Access Topology

Jan 15, 2026

This work proposes a general graph-based transmission design framework for multi-access coded caching systems under arbitrary user-cache access topologies, reformulating the coded multicast problem as a conflict graph coloring task. For the first time, graph neural networks are leveraged to achieve low-complexity near-optimal solutions, and the index coding duality bound is extended to arbitrary access structures. The proposed learning-based scheme achieves transmission loads close to those of the DSatur greedy algorithm and the theoretical duality bound, while substantially reducing computational overhead, thereby demonstrating strong scalability and adaptability in large-scale and heterogeneous network topologies.

1 citationsRead paper

An Out-Of-Distribution Membership Inference Attack Approach for Cross-Domain Graph Attacks

May 26, 2025

Graph Neural Networks (GNNs) are vulnerable to membership inference attacks (MIAs) due to their explicit exposure of topological structure; however, in realistic settings, attackers often lack access to auxiliary data drawn from the same distribution as the target model’s training data—rendering conventional MIA assumptions invalid. This work is the first to formulate cross-domain GNN-MIA as an out-of-distribution (OOD) problem. We propose GOOD-MIA, the first OOD-aware MIA framework for graph data: it constructs shadow subgraphs to characterize distributional discrepancies across domains, designs stable node representations coupled with environment-decoupling mechanisms to enhance cross-domain generalization, and employs risk extrapolation optimization to enable effective attack without requiring same-distribution auxiliary data. Evaluated on multiple domain-shifted graph datasets, GOOD-MIA achieves an average attack accuracy improvement of 12.7% over state-of-the-art methods.

1 citationsRead paper

Moirae: A Multimodal Agent Collaborative Framework for Dynamic Android Malware Detection

Aug 28, 2026

针对安卓恶意软件检测中的概念漂移问题，提出Moirae框架，通过动态收集多模态运行时证据并利用专门代理分析行为视图来解决。

0 citationsRead paper

Recent publications

Latest Papers

Moirae: A Multimodal Agent Collaborative Framework for Dynamic Android Malware Detection

Aug 28, 2026

针对安卓恶意软件检测中的概念漂移问题，提出Moirae框架，通过动态收集多模态运行时证据并利用专门代理分析行为视图来解决。

0 citationsRead paper

Does More Retrieved Evidence Help Visual Retrieval-Augmented Generation with Diffusion Language Models?

Aug 07, 2026

In visual retrieval-augmented generation, indiscriminately incorporating more retrieved evidence, while improving recall, often degrades the answer accuracy of diffusion language models due to semantic conflicts. This work identifies that mixing incompatible visual sources during parallel denoising leads to answer distortion and proposes a training-free Entropy-based Candidate Filter (ECF) to selectively introduce high-quality evidence. ECF leverages multi-granularity evidence units and confidence estimation from blank control blocks to assess retrieval candidates. Integrated with retrieval re-ranking and a training-free admission mechanism, the method achieves an average accuracy gain of 2.62 percentage points across three visual diffusion language models and five visual question answering benchmarks, significantly outperforming fixed top-k selection and other training-free strategies.

0 citationsRead paper

Audio-to-Score Transcription using Pre-trained Features, Data Augmentation, and the New SheetSage-A2S Dataset

Aug 06, 2026

This work addresses the limited focus of existing audio-to-score transcription research on classical music and the consequent lack of effective modeling for popular music. To bridge this gap, the authors introduce SheetSage-A2S, the first large-scale dataset specifically designed for popular music audio-to-score transcription. They propose an end-to-end transcription model that leverages MuQ-pretrained audio features, tailored data augmentation strategies, and a **kern-based score representation. The approach demonstrates substantially improved cross-genre generalization: it reduces the symbol error rate (SER) to 4.98% on the classical Quartets dataset—down from the previous best of 15.3%—and achieves an SER of 20.92% on SheetSage-A2S, thereby establishing the first benchmark for popular music audio-to-score transcription.

0 citationsRead paper

Self-Referential $K$-SAT and the Finite Analogue of Gödel's Incompleteness Theorem

Jul 01, 2026

This work uncovers an intrinsic undecidability in Boolean satisfiability (SAT) arising from self-reference and solution independence, establishing the first finite combinatorial analogue of Gödel’s incompleteness theorem. By constructing logarithmic-width K-SAT instances that simultaneously admit unsatisfiable and uniquely satisfiable formulas, and employing single-clause substitutions to generate locally indistinguishable SAT/UNSAT pairs, the study integrates algorithmic information theory, Shannon channel models, resolution proof complexity, and statistical physics approaches from random K-SAT. It demonstrates that any deductive system operating within a sublinear window requires proofs of exponential length. The paper reinterprets the Strong Exponential Time Hypothesis (SETH) as a manifestation of incompleteness in computational complexity and introduces a novel paradigm based on instance indistinguishability, revealing fundamental limitations of quantum computing and machine learning on such problems.

0 citationsRead paper

CRITIC-R1: Learning Structured Critics for Retrieval-Augmented Generation

May 28, 2026

This work addresses the susceptibility of existing retrieval-augmented generation (RAG) systems to hallucinations and reasoning errors, compounded by current critique mechanisms that provide coarse, unstructured, and overly intrusive feedback, leading to unreliable corrections. To overcome these limitations, the authors propose CRITIC-R1, a framework that formalizes RAG critique as an explicit, multi-dimensional error diagnosis task encompassing judgment, localization, analysis, and repair. Leveraging GRPO-based reinforcement learning guided by process-level supervision signals from external large language models, CRITIC-R1 trains a structured critique model augmented with a dual-reward mechanism—conservative judgment alignment and diagnostic quality alignment—to effectively curb over-intervention while enhancing feedback granularity and reliability. Experimental results demonstrate that this approach significantly outperforms strong RAG baselines across five question-answering benchmarks, consistently improving answer quality.

0 citationsRead paper