Papers by Ben Wu
MSEarth: A Multimodal Benchmark for Earth Science Phenomenon Discovery with MLLMs (2026.acl-long)
Copied to clipboard
Xiangyu Zhao, Wanghan Xu, Bo Liu, Yuhao Zhou, Fenghua Ling, Ben Fei, Xiaoyu Yue, Lei Bai, Wenlong Zhang, Xiao-Ming Wu
| Challenge: | Existing datasets often rely on synthetic data or figure-caption pairs, failing to capture the depth and complexity of geoscientific reasoning. |
| Approach: | They propose a multimodal scientific dataset and benchmark curated from open-access publications. |
| Outcome: | MSEarth features over 289K figures with captions enriched by contextual discussions and reasoning from original papers. |
Imbalanced Gradients in RL Post-Training of Multi-Task LLMs (2026.findings-eacl)
Copied to clipboard
Runzhe Wu, Ankur Samanta, Ayush Jain, Scott Fujimoto, Jeongyeol Kwon, Ben Kretzu, Youliang Yu, Kaveh Hassani, Boris Vidolov, Yonathan Efroni
| Challenge: | Large-gradient tasks can achieve similar or even much lower learning gains than small-grading ones. |
| Approach: | They show that large-gradient tasks can achieve lower learning gains than small-grading ones . large-grade tasks can accomplish similar or even lower learning gain than small grade ones if they are large . |
| Outcome: | The proposed approach fails when certain tasks produce larger gradients . Large-gradient tasks can achieve lower learning gains than small-gradent ones . |
Reliable Use of Lemmas via Eligibility Reasoning and Section-Aware Reinforcement Learning (2026.acl-short)
Copied to clipboard
| Challenge: | Recent large language models (LLMs) perform strongly on mathematical benchmarks but often import conclusions without validating assumptions. |
| Approach: | They propose a model that encodes a lemma specification and trains with reinforcement learning and section-aware loss masking to assign penalty to the section responsible for errors. |
| Outcome: | The proposed model performs well on benchmarks but often misapplyes lemmas . the model is able to encode the specification and train with reinforcement learning . |
OS-Genesis: Automating GUI Agent Trajectory Construction via Reverse Task Synthesis (2025.acl-long)
Copied to clipboard
Qiushi Sun, Kanzhi Cheng, Zichen Ding, Chuanyang Jin, Yian Wang, Fangzhi Xu, Zhenyu Wu, Chengyou Jia, Liheng Chen, Zhoumianze Liu, Ben Kao, Guohao Li, Junxian He, Yu Qiao, Zhiyong Wu
| Challenge: | Graphical User Interface (GUI) agents powered by Vision-Language Models (VLMs) have demonstrated human-like computer control capability. |
| Approach: | They propose a GUI data synthesis pipeline that reverse engineers GUI trajectory construction process by executing pre-defined tasks. |
| Outcome: | The proposed GUI data synthesis pipeline overcomes the bottlenecks of previous methods that rely on pre-defined tasks and limited data diversity. |
Good for Misconceived Reasons: An Empirical Revisiting on the Need for Visual Context in Multimodal Machine Translation (2021.acl-long)
Copied to clipboard
| Challenge: | Recent studies report improvements when equipping models with multimodal information, but it remains unclear whether such improvements actually come from the multimodal part. |
| Approach: | They propose to extend conventional text-only translation models with multimodal information by extending them with visual input. |
| Outcome: | The proposed models replicate similar gains as recently developed multimodal-integrated systems achieved, but learn to ignore multimodal information. |
Prompt Leakage effect and mitigation strategies for multi-turn LLM Applications (2024.emnlp-industry)
Copied to clipboard
| Challenge: | Prompt leakage poses a compelling security and privacy threat in LLM applications. |
| Approach: | They propose a model which leverages the LLM sycophancy effect and a threat model which fine tunes an open-source model to defend against prompt leakage attempts. |
| Outcome: | The proposed model elevates the average attack success rate (ASR) from 17.7% to 86.2% in a multi-turn setting. |
Perturbed Masking: Parameter-free Probing for Analyzing and Interpreting BERT (2020.acl-main)
Copied to clipboard
| Challenge: | Recent pre-trained language models achieve state-of-the-art performance for downstream NLP tasks. |
| Approach: | They propose a parameter-free probing technique for analyzing pre-trained language models . their method does not require direct supervision from probing tasks . |
| Outcome: | The proposed method improves on linguistically-uninformed baselines on pre-trained language models. |
A Meta-framework for Spatiotemporal Quantity Extraction from Text (2022.acl-long)
Copied to clipboard
| Challenge: | a meta-framework for news events that extracts quantities from text is proposed . a previous work on news events focused on extracting event mentions, attributes, and relationships . |
| Approach: | They propose a meta-framework for solving the NLP problem of spatiotemporal quantity extraction . they demonstrate the framework is general and extensible, and shareable crowdsourcing pipeline and baseline models are used . |
| Outcome: | The proposed framework is general and extensible, the authors say . it can extract quantity from news streams, quickly respond to emergencies, investigate incidents . |
OS-Sentinel: Towards Safety-Enhanced Mobile GUI Agents via Hybrid Validation in Realistic Workflows (2026.acl-long)
Copied to clipboard
Qiushi Sun, Mukai Li, Zhoumianze Liu, Zhihui Xie, Fangzhi Xu, Zhangyue Yin, Kanzhi Cheng, Zehao Li, Zichen Ding, Qi Liu, Zhiyong Wu, Zhuosheng Zhang, Ben Kao, Lingpeng Kong
| Challenge: | Existing methods for detecting unsafe mobile GUI agents are underexplored. |
| Approach: | They propose a mobile agent safety detection framework that integrates a formal verifier and a VLM-based contextual judge to detect system-level violations. |
| Outcome: | The proposed framework achieves 10%–30% improvements over existing approaches across multiple metrics. |
Lexical Knowledge Internalization for Neural Dialog Generation (2022.acl-long)
Copied to clipboard
| Challenge: | Existing knowledge-grounded dialog models ignore the knowledge that resides in people's minds during a conversation. |
| Approach: | They propose to integrate lexical knowledge internally into the model's parameters instead of further conditioning them on external knowledge . they adopt contrastive learning approach and use a dictionary-based token-level lexicon retriever that requires only weak supervision. |
| Outcome: | The proposed model can relate J.K Rowling to Khalsa Aid with the knowledge retrieved from Wikipedia. |
Don’t waste a single annotation: improving single-label classifiers through soft labels (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for annotating data are limited by ambiguity and lack of context in data samples. |
| Approach: | They challenge the traditional approach of annotating data by only providing a single label for each sample and annotator disagreement is discarded . instead, they use additional annotation information such as confidence, secondary label and disagreement to generate soft labels. |
| Outcome: | The proposed method improves model performance and calibration on the hard label test set. |