Papers by Shang Zhu
Tracing the Roots: A Multi-Agent Framework for Uncovering Data Lineage in Post-Training LLMs (2026.acl-long)
Copied to clipboard
Yu Li, Xiaoran Shang, Qizhi Pei, Yun Zhu, Xin Gao, Honglin Lin, Zhanping Zhong, Zhuoshi Pan, Zheng Liu, Xiaoyang Wang, Conghui He, Dahua Lin, Feng Zhao, Lijun Wu
| Challenge: | High-quality post-training data is the primary engine driving LLM capabilities . datasets are often treated as isolated artifacts, overlooking their true developmental context . |
| Approach: | They propose a framework to reconstruct the evolutionary graph of dataset development using data lineage. |
| Outcome: | The proposed framework characterizes domain-specific structural patterns in Math-oriented datasets and general-domain corpora. |
ReasonIF: Large Reasoning Models Fail to Follow Instructions During Reasoning (2026.findings-acl)
Copied to clipboard
| Challenge: | Prior studies assess instruction adherence in the model’s main responses, but it is also critical for large reasoning models to follow user instructions throughout their reasoning process. |
| Approach: | They propose a systematic benchmark for assessing reasoning instruction following to assess the model's adherence to instructions. |
| Outcome: | The proposed benchmark reduces the risk of undesirable shortcuts, hallucinations, or reward hacking within reasoning traces. |
ChartVerse: Scaling Chart Reasoning via Reliable Programmatic Synthesis from Scratch (2026.acl-long)
Copied to clipboard
Zheng Liu, Honglin Lin, Xiaoyang Wang, Xin Gao, Yu Li, Mengzhang Cai, Yun Zhu, Zhanping Zhong, Qizhi Pei, Zhuoshi Pan, Xiaoran Shang, Conghui He, Bin Cui, Wentao Zhang, Lijun Wu
| Challenge: | Existing open-source vision language models lack high-quality training data for chart reasoning . current models are simplistic and repetitive, while associated QA pairs are prone to hallucinations . |
| Approach: | They propose a framework to synthesize complex charts and reliable reasoning data from scratch. |
| Outcome: | Experimental results show that ChartVerse-8B surpasses existing models in QA and difficulty . lack of high-quality training data hampers development of open-source models . |
MINER: Multi-Interest Matching Network for News Recommendation (2022.findings-acl)
Copied to clipboard
| Challenge: | Existing methods learn a single user embedding from user’s historical behaviors to represent the reading interest. |
| Approach: | They propose a poly attention scheme to learn multiple interest vectors for each user, which encodes the different aspects of user interest. |
| Outcome: | The proposed approach significantly outperforms existing state-of-the-art methods on the MIND news recommendation benchmark. |
Self-Error-Instruct: Generalizing from Errors for LLMs Mathematical Reasoning (2025.acl-long)
Copied to clipboard
Erxin Yu, Jing Li, Ming Liao, Qi Zhu, Boyang Xue, Minghui Xu, Baojun Wang, Lanqing Hong, Fei Mi, Lifeng Shang
| Challenge: | Existing approaches to learning from errors synthesize training data by extrapolating from isolated bad cases, thereby failing to generalize the extensive patterns inherent within these cases. |
| Approach: | They propose a framework that synthesizes more generalized training data from isolated bad cases by extrapolating from isolated cases. |
| Outcome: | The proposed framework synthesizes more generalized training data to address these model weaknesses. |
How Should We Enhance the Safety of Large Reasoning Models: An Empirical Study (2026.acl-long)
Copied to clipboard
Zhexin Zhang, Xian Qi Loye, Victor Shea-Jay Huang, Junxiao Yang, Qi Zhu, Shiyao Cui, Fei Mi, Lifeng Shang, Yingkang Wang, Hongning Wang, Minlie Huang
| Challenge: | Large Reasoning Models have achieved remarkable success on reasoning-intensive tasks, but their enhanced reasoning capabilities do not translate to improved safety performance. |
| Approach: | They propose to use supervised fine tuning to enhance the safety of Large Reasoning Models. |
| Outcome: | The proposed method improves the safety of large reasoning models on reasoning-intensive tasks. |
AutoConv: Automatically Generating Information-seeking Conversations with Large Language Models (2023.acl-short)
Copied to clipboard
Siheng Li, Cheng Yang, Yichun Yin, Xinyu Zhu, Zesen Cheng, Lifeng Shang, Xin Jiang, Qun Liu, Yujiu Yang
| Challenge: | Existing research on information-seeking conversations is stymied by the lack of training data. |
| Approach: | They propose to use autoconv for synthetic conversation generation to capture the characteristics of the information-seeking process and fine tune an LLM with a few human conversations to generate synthetic conversations with high quality. |
| Outcome: | The proposed model improves on two commonly-used datasets and alleviates the dependence on human annotation. |
More Tokens, Lower Precision: Towards the Optimal Token-Precision Trade-off in KV Cache Compression (2025.findings-emnlp)
Copied to clipboard
Jiebin Zhang, Dawei Zhu, Yifan Song, Wenhao Wu, Chuqiao Kuang, Xiaoguang Li, Lifeng Shang, Qun Liu, Sujian Li
| Challenge: | storing more tokens in the KV cache at lower precision can enhance the long-context performance of large language models. |
| Approach: | They propose a token-precision trade-off strategy to optimize KV cache compression . they also propose storing more tokens in the KV at lower precision . |
| Outcome: | The proposed method achieves an optimal point within the Information Bottleneck compared to standalone KV pruning or KV quantization. |
CompileAgent: Automated Real-World Repo-Level Compilation with Tool-Integrated LLM-based Agent System (2025.acl-long)
Copied to clipboard
Li Hu, Guoqiang Chen, Xiuwei Shang, Shaoyin Cheng, Benlong Wu, LiGangyang LiGangyang, Xu Zhu, Weiming Zhang, Nenghai Yu
| Challenge: | CompileAgent is the first LLM-based agent framework dedicated to repo-level compilation. |
| Approach: | They propose a LLM-based agent framework dedicated to repo-level compilation. |
| Outcome: | The proposed method significantly improves compilation success rate, ranging from 10% to 71%. |
BLADE: Benchmarking Language Model Agents for Data-Driven Science (2024.findings-emnlp)
Copied to clipboard
Ken Gu, Ruoxi Shang, Ruien Jiang, Keying Kuang, Richard-John Lin, Donghe Lyu, Yue Mao, Youran Pan, Teng Wu, Jiaqian Yu, Yikun Zhang, Tianmai Zhang, Lanyi Zhu, Mike Merrill, Jeffrey Heer, Tim Althoff
| Challenge: | Language model-based agents can be used to conduct and support data-driven science, but evaluating them on open-ended tasks is challenging due to multiple valid approaches, partially correct steps, and different ways to express the same decisions. |
| Approach: | They propose a benchmark to automatically evaluate agents’ multifaceted approaches to open-ended research questions. |
| Outcome: | BLADE evaluates agents’ multifaceted approaches to open-ended research questions using data from 12 datasets and research questions drawn from existing scientific literature. |
VEEF-Multi-LLM: Effective Vocabulary Expansion and Parameter Efficient Finetuning Towards Multilingual Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have a significant disadvantage for low-resource languages . VEEF-Multi-LLM-8B excels in multilingual instruction-following tasks . |
| Approach: | They propose a low-resource multilingual large language model that expands the vocabulary for multilingual support. |
| Outcome: | The proposed model outperforms existing models in multilingual instruction-following tasks, but lags behind English-centric models in some tasks. |
Bidirectional LMs are Better Knowledge Memorizers? A Benchmark for Real-world Knowledge Injection (2026.acl-long)
Copied to clipboard
Yuwei Zhang, Wenhao Yu, Shangbin Feng, Yifan Zhu, Letian Peng, Jayanth Srinivasa, Gaowen Liu, Jingbo Shang
| Challenge: | Existing knowledge injection benchmarks for large language models lack standardized testing grounds. |
| Approach: | They propose a knowledge injection benchmark that leverages recently-added and expert-curated facts from Wikipedia’s “Did You Know...” entries. |
| Outcome: | The proposed framework improves reliability accuracy by 29.1%. |
Planning, Creation, Usage: Benchmarking LLMs for Comprehensive Tool Utilization in Real-World Complex Scenarios (2024.findings-acl)
Copied to clipboard
Shijue Huang, Wanjun Zhong, Jianqiao Lu, Qi Zhu, Jiahui Gao, Weiwen Liu, Yutai Hou, Xingshan Zeng, Yasheng Wang, Lifeng Shang, Xin Jiang, Ruifeng Xu, Qun Liu
| Challenge: | Existing benchmarks focus on simple synthesized queries that do not reflect real-world complexity, thereby offering limited perspectives in evaluating tool utilization. |
| Approach: | They propose a benchmark to evaluate LLMs’ ability in tool utilization within real-world scenarios. |
| Outcome: | The proposed benchmark improves LLMs’ ability in tool utilization within real-world scenarios and eliminates the restriction of pre-defined toolset. |
Rethinking Expert Trajectory Utilization in LLM Post-training for Mathematical Reasoning (2026.acl-long)
Copied to clipboard
Bowen Ding, Yuhan Chen, Jiayang Lyu, Jiyao Yuan, Qi Zhu, Shuangshuang Tian, Dantong Zhu, Futing Wang, Heyuan Deng, Fei Mi, Lifeng Shang, Tao Lin
| Challenge: | evolving generic Large Language Models into specialized Large Reasoning Models requires effective post-training. |
| Approach: | They propose a plasticity-ceiling framework to harness expert trajectories . they establish the Sequential SFT-then-RL pipeline as the superior standard . |
| Outcome: | The proposed framework overcomes stability and premature convergence deficits in synchronized approaches. |
SafeScientist: Enhancing AI Scientist Safety for Risk-Aware Scientific Discovery (2025.emnlp-main)
Copied to clipboard
Kunlun Zhu, Jiaxun Zhang, Ziheng Qi, Nuoxing Shang, Zijia Liu, Peixuan Han, Yue Su, Haofei Yu, Jiaxuan You
| Challenge: | Recent advances in large language model (LLM) agents have significantly accelerated scientific discovery automation, yet raised critical ethical and safety concerns. |
| Approach: | They propose a framework to enhance safety and ethical responsibility in AI-driven scientific exploration. |
| Outcome: | The proposed framework significantly improves safety performance by 35% compared to traditional frameworks. |