Institution profile

Hong Kong Polytechnic University

Academic institutionasia · hk
Official website
Research library2,102linked papers
Opportunities0open roles
Selected work

Representative Papers

CheatAgent: Attacking LLM-Empowered Recommender Systems via LLM Agent

Aug 24, 2024Knowledge Discovery and Data Mining

This work investigates security vulnerabilities in LLM-augmented recommender systems (RecSys) under black-box settings—a previously unexplored threat scenario. Method: We propose the first adversarial attack framework leveraging an LLM-based agent, which identifies high-impact text insertion positions via position-aware mechanisms, generates semantically preserved and minimally perturbed adversarial texts, and iteratively refines prompts using black-box query feedback. Crucially, the LLM itself serves as the attack agent, enabling efficient, stealthy, and cross-model transferable attacks. Contribution/Results: Extensive experiments on three real-world datasets demonstrate that our framework significantly outperforms conventional reinforcement learning–based attacks: it achieves higher attack success rates, requires fewer API queries, and induces smaller textual perturbations. These results empirically expose previously underappreciated, substantive security and privacy risks inherent in LLM-RecSys architectures.

14 citations1 influentialRead paper

Data-Driven Merton's Strategies via Policy Randomization

Dec 19, 2023

This paper addresses the Merton expected utility maximization problem in an incomplete market with fully unknown dynamics: the market comprises a stock and latent state-factor processes, while investors observe only prices and instantaneous volatility—rendering factor dynamics and market parameters unidentifiable. We propose a novel continuous-time reinforcement learning (RL) framework that, for the first time, employs Gaussian policy randomization as an analytical tool—not merely for exploration—and rigorously prove that the randomized policy’s mean coincides with the optimal control of the original problem, thereby bridging RL and classical portfolio theory. We design online and offline actor-critic algorithms integrating policy improvement theorems with randomized policy analysis. Under stochastic volatility, our method substantially outperforms conventional parametric interpolation approaches. Both simulation studies and empirical tests confirm its robustness and capacity to generate alpha.

10 citations1 influentialRead paper

FedConv: A Learning-on-Model Paradigm for Heterogeneous Federated Clients

Jun 03, 2024ACM SIGMOBILE International Conference on Mobile Systems, Applications, and Services

To address the uneven computational burden imposed by client-side resource heterogeneity in federated learning (FL), this paper proposes FedConv. It trains lightweight submodels directly in compressed convolutional form, eliminating decompression overhead. FedConv introduces the novel “learning-on-model” paradigm—the first approach enabling end-to-end training of compressed submodels. It further designs a transposed-convolution-based expansion mechanism to unify aggregation of heterogeneous submodels while preserving personalized parameters. Complemented by joint optimization on the server using a small public dataset, FedConv achieves an average accuracy improvement of 35.2% across six benchmark datasets, while reducing computational cost by 33.1% and communication cost by 24.8%, significantly outperforming existing FL methods.

7 citationsRead paper

Agent-as-a-Judge

Jan 08, 2026arXiv.org

Traditional LLM-as-a-Judge approaches are limited in evaluating complex, multi-step tasks due to inherent biases, shallow reasoning, and a lack of real-world validation capabilities. This work proposes a paradigm shift toward Agent-as-a-Judge, establishing a unified, verifiable, and fine-grained evaluation framework for intelligent agents. Through a systematic review, the study integrates key technical dimensions—including planning, tool usage, multi-agent collaboration, and persistent memory—and presents the first comprehensive taxonomy of evaluation benchmarks spanning both general and domain-specific scenarios. Furthermore, it outlines a roadmap for future research, clearly identifying current challenges and promising directions in agent-based evaluation methodologies.

4 citationsRead paper

AgencyBench: Benchmarking the Frontiers of Autonomous Agents in 1M-Token Real-World Contexts

Jan 16, 2026

This work addresses the limitations of existing agent benchmarks, which often focus on isolated capabilities and struggle to evaluate long-horizon, high-complexity real-world tasks due to reliance on manual feedback that hinders scalability. The authors propose the first comprehensive, automated benchmark tailored to everyday AI usage scenarios, encompassing 32 real-world settings and 138 tasks—each requiring an average of 90 tool invocations and processing over one million tokens. The framework employs user-simulation agents for iterative feedback, Docker-based sandboxing for visual and functional rule validation, and a standardized task interface enabling unified closed-loop evaluation of both open- and closed-source models. Experimental results demonstrate a significant performance gap favoring closed-source models (48.4% vs. 32.1%) and highlight the critical role of co-optimizing models with agent frameworks to enhance overall effectiveness.

3 citationsRead paper
Recent publications

Latest Papers