Papers by Xinbei Ma
EVA: Evolving Semantic Adversaries for Red-Teaming GUI Agents Against Environmental Injection Attacks (2026.findings-acl)
Copied to clipboard
Yijie Lu, Manman Zhao, Tianjie Ju, Zihe Yan, Xinbei Ma, Yuan Guo, Daizong Ding, Gongshen Liu, Zhuosheng Zhang
| Challenge: | Existing methods for red-teaming face a trade-off between requiring target-specific knowledge and incurring prohibitive computational costs. |
| Approach: | They propose a framework that evolves payloads exclusively on the semantic dimension via a discovery-deployment pipeline. |
| Outcome: | Experiments show that EVA outperforms baselines in terms of attack success rate while evolving benign seeds into successful attacks within 1.18 to 1.71 iterations. |
Query Rewriting in Retrieval-Augmented Large Language Models (2023.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies focus on adapting either the retriever or the reader, but this approach is more focused on adaptation of the query itself. |
| Approach: | They propose a new framework for retrieval-augmented Large Language Models . they propose rewrite-retrieve-read instead of retrieve-then-read . |
| Outcome: | The proposed framework improves performance on downstream tasks, open-domain QA and multiple-choice QA. |
Structural Characterization for Dialogue Disentanglement (2022.acl-long)
Copied to clipboard
| Challenge: | tangled multi-party dialogues lead to difficulties in understanding the dialogue history for both human and machine. |
| Approach: | They propose a model for disentangling multi-party dialogues using speaker property and reference dependency. |
| Outcome: | The proposed model achieves state-of-the-art on the Ubuntu IRC benchmark dataset and contributes to dialogue-related comprehension. |
PGPO: Enhancing Agent Reasoning via Pseudocode-style Planning Guided Preference Optimization (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing LLM agents generate verbose and inefficient natural language plans to guide reasoning, which restricts agents’ ability to generalize across similar tasks. |
| Approach: | They propose a pseudocode-style planning guide optimization method that captures the structural logic of reasoning and uses two planning-oriented rewards to enhance agent learning. |
| Outcome: | The proposed method outperforms existing LLM agents on representative agent benchmarks and outperformed the current leading baselines. |
ParaCook: On Time-Efficient Planning for Multi-Agent Systems (2026.findings-acl)
Copied to clipboard
Shiqi Zhang, Xinbei Ma, Yunqing Xu, Zouying Cao, Pengrui Lu, Haobo Yuan, Tiancheng Shen, Zhuosheng Zhang, Hai Zhao, Ming-Hsuan Yang
| Challenge: | Existing agent benchmarks focus on task completion while neglecting time efficiency in parallel and asynchronous operations. |
| Approach: | They propose a framework for large language models that allows agents to plan long-horizon tasks in a scalable way. |
| Outcome: | The proposed framework is based on the Overcooked game and can be used to evaluate time efficiency-aware multi-agent planning. |
CoCo-Agent: A Comprehensive Cognitive MLLM Agent for Smartphone GUI Automation (2024.findings-acl)
Copied to clipboard
| Challenge: | Current vital challenges for autonomous agents lie in two aspects: dependence on strong (M)LLMs and insufficient GUI environment modeling. |
| Approach: | They propose a comprehensive cognitive LLM agent with two novel approaches to improve GUI automation performance. |
| Outcome: | The proposed agent achieves state-of-the-art performance on AITW and META-GUI benchmarks. |
Social Welfare Function Leaderboard: On the Emergence of LLM Agents as the Welfare Dictator (2026.findings-acl)
Copied to clipboard
Zhengliang Shi, Ruotian Ma, Jen-tse Huang, Xinbei Ma, Xingyu Chen, Mengru Wang, Qu Yang, Yue Wang, Fanghua Ye, Ziyang Chen, Shanyi Wang, Cixing LI, Wenxuan Wang, Zhaopeng Tu, Xiaolong Li, Zhaochun Ren, Liefeng Bo
| Challenge: | Large language models (LLMs) are increasingly entrusted with high-stakes decisions that affect human welfare. |
| Approach: | They evaluate 20 state-of-the-art Large language models (LLMs) and 20 LLM dictators to create a social welfare function benchmark. |
| Outcome: | The proposed model creates dilemma between maximizing collective efficiency and ensuring distributive fairness. |
Agent-Dice: Disentangling Knowledge Updates via Geometric Consensus for Agent Continual Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Model (LLM)-based agents extend the utility of LLMs by interacting with dynamic environments. |
| Approach: | They propose a parameter fusion framework based on directional consensus evaluation that disentangles knowledge updates through a two-stage process. |
| Outcome: | The proposed framework disentangles knowledge updates through a two-stage process with minimal computational overhead and parameter updates. |
Dynamic Planning for LLM-based Graphical User Interface Automation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to planning for GUI tasks are limited due to long historical dialogues. |
| Approach: | They propose a novel approach to dynamic planning based on environmental feedback and execution history to guide action prediction in GUI tasks. |
| Outcome: | The proposed approach surpasses the strong GPT-4V baseline by +12.7% in accuracy. |
On the Robustness of Editing Large Language Models (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have exhibited impressive success and significant potential. |
| Approach: | They propose to modify the knowledge memory with minimum computational cost while preserving the performance on the retained knowledge. |
| Outcome: | The proposed methods avoid retraining to update the model parameters and have demonstrated promising performance and efficiency. |
LESA: Learnable LLM Layer Scaling-Up (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for depth scaling-up rely on empirical heuristic rules for layer duplication, resulting in poor initialization and slower convergence during continual pre-training. |
| Approach: | They propose a method for learning latent parameters between layers by concatenating parameters from each layer and applying Singular Value Decomposition. |
| Outcome: | Experiments show that LESA outperforms baseline models with less than half the cost of existing methods. |
Caution for the Environment: Multimodal LLM Agents are Susceptible to Environmental Distractions (2025.acl-long)
Copied to clipboard
| Challenge: | Experimental results show that multimodal GUI agents are susceptible to environmental distractions. |
| Approach: | They propose a scenario where both user and agent are benign and environment is not malicious . they implement an adversarial environment injection and analyze the approach to improve faithfulness . |
| Outcome: | The proposed approach improves faithfulness of multimodal large language model agents in a graphical user interface environment. |
PROM: A Phrase-level Copying Mechanism with Pre-training for Abstractive Summarization (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing summarization strategies are abstractive and extractive, but are hard to control. |
| Approach: | They propose a PhRase-level cOpying Mechanism that enhances attention on n-grams and calculates an auxiliary loss for the copying prediction. |
| Outcome: | Empirical studies show that PROM improves copying accuracy and faithfulness on benchmarks. |
MEGen: Generative Backdoor into Large Language Models via Model Editing (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for training large language models are limited to yes-or-no discriminative tasks, leading users to underestimate the potential risks. |
| Approach: | They propose an editing-based generative backdoor that expands the backdoor to generative tasks in a unified format of any text-to-any text. |
| Outcome: | The proposed model achieves high attack success rate by adjusting only a small set of local parameters with few-shot samples. |