Publications

arXiv preprint arXiv:2611.12865

Shortcuts and Boundaries: Verifiable RL Misalignment and Contextual Safety Learning

Verifiable reward–context conflicts reveal how agentic RL learns shortcuts and how execution feedback supports useful, transferable contextual safety.

arXiv preprint

SEAD: A State-Based Perspective on Attack and Defense in Tool-Using Agents

A state-based framework for attacks and defenses in tool-using agents, pairing execution-guided trajectory search with state-aware pre-execution defense.

Qwen Technical Report

Verifiable Hidden Dynamics Play: Generating Agentic RL Environments from Solved Mechanisms

Generate diverse, stateful agentic RL environments with verifiable rewards by deriving both dynamics and evaluation from solved mechanisms.

Qwen Technical Report

E-Commerce Bench: Evaluating LLM Agents on Long-Horizon Autonomous Business Operation

A reproducible 365-day benchmark for agents operating online stores under negotiation, delayed outcomes, market shocks, and persistent memory demands.

ICML 2026 AIWILD Workshop

One Turn Too Late: Response-Aware Defense Against Hidden Malicious Intent in Multi-Turn Dialogue

A response-aware monitor trained to intervene at the first harm-enabling turn while preserving benign exploratory dialogue.

NeurIPS 2026

How Far Are VLMs from Privacy Awareness in the Physical World?

An interactive audiovisual benchmark showing that current VLMs struggle to turn privacy cues into situated behavior.

arXiv preprint

The Trojan Knowledge: Bypassing Commercial LLM Guardrails via Harmless Prompt Weaving and Adaptive Tree Search

Introduce CKA-Agent, a framework that reformulates jailbreaking as an adaptive tree search over the target LLM’s correlated knowledge, achieving 96-99% attack success rates against state-of-the-art commercial LLMs.

Preprint

Measuring Physical-World Privacy Awareness of Large Language Models: An Evaluation Benchmark

A comprehensive evaluation benchmark for assessing privacy awareness of large language models in physical environments, revealing significant gaps when privacy is grounded in real-world contexts across four evaluation tiers.

In SIGIR 2025

Exploring ℓ0 Sparsification for Inference-free Sparse Retrievers

In IEEE Transactions on Neural Networks and Learning Systems

Cross-View Approximation on Grassmann Manifold for Multiview Clustering

Propose a cross-view approximation on Grassman manifold (CAGM) model to address inconsistencies within multiview adjacency matrices, feature matrices, and cross-view combinations from the two sources.

In The Web Conference 2024

FinReport: Explainable Stock Earnings Forecasting via News Factor Analyzing Model

Propose an explainable stock earning framework via news factor analyzing model. Formalizing news into semantic graph and learn embedding of it as a more expressible factor. Stock earning foresting and explainable earning module are built via aggregating and utilizing news factor and numeric stock factor, achieving SOTA performance on A-stock dataset. A detailed report will be generated with the power of LLM.