Institution profile

Tianjin University

Academic institutionasia · cn
Official website
Research library989linked papers
Opportunities0open roles
Selected work

Representative Papers

Prompt Injection attack against LLM-integrated Applications

Jun 08, 2023arXiv.org

Prompt injection attacks pose an increasingly severe security threat to large language model (LLM) integrated applications, yet existing black-box attack methods suffer from limited practical efficacy. Method: This paper proposes HouYi—the first real-world-oriented, three-stage black-box prompt injection framework comprising pre-prompt injection, context-aware segmentation, and malicious payload delivery. HouYi uniquely enables automated triggering of high-impact consequences—including arbitrary LLM misuse and application-level prompt stealing—via black-box fuzzing, context-aware prompt engineering, and web-injection-inspired modeling. Contribution/Results: Evaluated through real-world penetration testing across 36 mainstream LLM applications, HouYi uncovered 31 critical vulnerabilities, independently confirmed by ten vendors—including Notion—with impact on millions of users. The work significantly advances LLM security practice by bridging the gap between theoretical attack models and deployable, scalable exploitation techniques.

536 citations38 influentialRead paper

FedBCGD: Communication-Efficient Accelerated Block Coordinate Gradient Descent for Federated Learning

Oct 28, 2024ACM Multimedia

This work addresses the high communication overhead of large-scale models, such as Vision Transformers, in federated learning by proposing Federated Block Coordinate Gradient Descent (FedBCGD) and its accelerated variant, FedBCGD+. The method introduces, for the first time in federated learning, a block-wise parameter communication mechanism that uploads only a subset of parameter blocks per round, combined with stochastic variance reduction and client drift control strategies. Theoretical analysis shows that the communication complexity is reduced by a factor of 1/N compared to existing methods, where N denotes the number of blocks. Experimental results demonstrate that the proposed algorithms achieve faster convergence and higher communication efficiency than current state-of-the-art approaches.

37 citations2 influentialRead paper

The Box is in the Pen: Evaluating Commonsense Reasoning in Neural Machine Translation

Nov 01, 2020Findings

This work investigates neural machine translation (NMT) models’ capacity to resolve lexical and syntactic ambiguities via commonsense reasoning. To this end, we introduce CReM—the first commonsense reasoning evaluation benchmark tailored for NMT—comprising 1,200 triplets spanning seven categories of commonsense knowledge. We propose a dual-dimensional evaluation framework assessing both accuracy (where mainstream models achieve only 60.1%) and cross-context consistency (with merely 31% consistency), enabling the first systematic quantification of NMT’s commonsense reasoning capability. Extensive comparative experiments are conducted using models including BERT and GPT-2; statistical analysis and error attribution reveal that contextual modeling and effective commonsense integration remain critical bottlenecks. The CReM benchmark is publicly released to serve as a standardized evaluation tool for future research.

24 citations1 influentialRead paper

Consistency of Local and Global Flatness for Federated Learning

Oct 27, 2025ACM Multimedia

This work addresses the challenge in federated learning where multiple local updates under data heterogeneity often drive the global model toward sharp minima, degrading generalization. Existing sharpness-aware methods struggle to align local and global flatness. To this end, we propose FedNSAM, an algorithm that leverages global Nesterov momentum to guide local updates, constructing an estimated direction of global perturbation and performing extrapolation to harmonize local and global flatness. We introduce a novel “flatness distance” metric to quantify the inconsistency between local and global landscapes and establish a tighter convergence bound than FedSAM in our theoretical analysis. Empirical results demonstrate that FedNSAM significantly enhances both generalization performance and training efficiency across CNN and Transformer architectures, particularly in highly heterogeneous settings.

11 citationsRead paper

TGEA: An Error-Annotated Dataset and Benchmark Tasks for TextGeneration from Pretrained Language Models

Mar 06, 2025Annual Meeting of the Association for Computational Linguistics

This work addresses the challenge of diagnosing errors in text generated by pretrained language models (PLMs). We introduce TGEA, the first fine-grained error annotation dataset specifically designed for large language model (LLM) outputs, comprising 47K generated sentences and 12K human-annotated error instances spanning 24 linguistically and knowledge-based error categories. TGEA pioneers a multidimensional annotation schema—including error span, associated span, minimal correction, error type, and root-cause analysis—grounded in GPT-2–based controllable generation, crowdsourced annotation, and a linguistics- and commonsense-informed taxonomy. Leveraging this dataset, we establish a comprehensive PLM generation diagnostics benchmark, formalizing novel tasks: error detection, fine-grained classification, association identification, and attribution generation. TGEA substantially enhances the interpretability and diagnosability of PLM-generated errors, providing both foundational resources and standardized evaluation protocols for automated error detection and correction research.

8 citations2 influentialRead paper
Recent publications

Latest Papers