Papers by Shijie Zhou
MARS2: Scaling Multi-Agent Tree Search via Reinforcement Learning for Code Generation (2026.acl-long)
Copied to clipboard
Pengfei Li, Shijie Wang, Fangyuan Li, Yikun Fu, Kaifeng Liu, Kaiyan Zhang, Dazhi Zhang, Yuqiang Li, Biqing Qi, Bowen Zhou
| Challenge: | Existing approaches to reinforcement learning are decoupled from structured search due to limited trajectory diversity. |
| Approach: | They propose a unified RL framework that integrates multiple agents within a shared tree-structured search environment. |
| Outcome: | Experiments show that MARS2 improves performance across diverse model combinations and training settings. |
A High-Quality Text-Rich Image Instruction Tuning Dataset via Hybrid Instruction Generation (2025.coling-main)
Copied to clipboard
| Challenge: | Large multimodal models struggle with text-rich images because of inadequate training data. |
| Approach: | They propose to use annotations from human annotators to generate instruction data by a hybrid approach to generate text prompts for large language models. |
| Outcome: | The proposed model improves multimodal alignment for text-rich images by using human annotations and tailored text prompts for large language models. |
PaCoRe: Learning to Scale Test-Time Compute with Parallel Coordinated Reasoning (2026.acl-long)
Copied to clipboard
Jingcheng Hu, Yinmin Zhang, Shijie Shang, Xiaobo Yang, Yue Peng, Zhewei Huang, Hebin Zhou, Xin Wu, Jie Cheng, Fanqi Wan, Xiangwen Kong, Chengyuan Yao, Kaiwen Yan, Ailin Huang, Hongyu Zhou, Qi Han, Zheng Ge, Xiangyu Zhang, Heung-Yeung Shum
| Challenge: | Parallel Coordinated Reasoning (PaCoRe) overcomes a central limitation of contemporary language models: their inability to scale test-time compute (TTC) far beyond sequential reasoning under a fixed context window. |
| Approach: | They propose a training-and-inference framework to overcome a central limitation of language models: their inability to scale test-time compute (TTC) under a fixed context window. |
| Outcome: | The proposed model scales to multi-million-token effective TTC without exceeding context limits. |
Musical Score Understanding Benchmark: Evaluating Large Language Models’ Comprehension of Complete Musical Scores (2026.acl-long)
Copied to clipboard
Congren Dai, Yue Yang, Krinos Li, Huichi Zhou, Shijie Liang, Zhang Bo, Enyang Liu, Ge Jin, Hongran An, Haosen Zhang, Peiyuan Jing, KinHei Lee, Zhenxuan Zhang, Xiaobing Li, Maosong Sun
| Challenge: | Existing benchmarks for musical score understanding are narrow in scope, focusing on isolated fragments, short excerpts, or multiple-choice formulations, rather than supporting holistic reasoning over entire scores. |
| Approach: | They propose a benchmark for score-level musical understanding across textual and visual modalities. |
| Outcome: | The musical score understanding benchmark contains 1,800 question-answer pairs from works by Bach, Beethoven, Chopin, Debussy, and others. |
SwarmAgentic: Towards Fully Automated Agentic System Generation via Swarm Intelligence (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing agentic system generation frameworks lack autonomy, autonomy, and functionality . current frameworks are too rigid, limiting adaptability and scalability. |
| Approach: | They propose a framework that fully automates agentic system generation, optimization, and collaboration . they construct agents from scratch and jointly refine functionality and coordination . |
| Outcome: | The proposed framework outperforms ADAS on six real-world, open-ended, and exploratory tasks on the TravelPlanner benchmark. |
Unveiling Inherent Visual Grounding in Multimodal LLMs for Text-Rich Images (2026.findings-acl)
Copied to clipboard
Shijie Zhou, Jihyung Kil, Ming Li, Jiuxiang Gu, Curtis Wigington, Rajiv Jain, Changyou Chen, Ruiyi Zhang
| Challenge: | Existing multimodal large language model (MLLM) approaches struggle to align query tokens with visual–text patches, heavily relying on lengthy OCR inputs. |
| Approach: | They propose an OCR-free approach that leverages the MLLM's inherent multi-head attention for multi-patch grounding. |
| Outcome: | Empirical results show that the proposed approach outperforms existing approaches on challenging document grounding benchmarks. |
Reinforcement Learning for Large Language Models via Group Preference Reward Shaping (2025.emnlp-main)
Copied to clipboard
Huaisheng Zhu, Siyuan Xu, Hangfan Zhang, Teng Xiao, Zhimeng Guo, Shijie Zhou, Shuyue Hu, Vasant G. Honavar
| Challenge: | Existing methods for fine-tuning Large Language Models (LLMs) are expensive and sensitive to reward model quality. |
| Approach: | They propose a method that leverages preference-based comparisons rather than precise numerical rewards. |
| Outcome: | Experiments show that GPRS outperforms critic-model-free RL algorithms on RLHF and reasoning tasks. |
QPruner: Probabilistic Decision Quantization for Structured Pruning in Large Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Structured pruning can reduce model size but results in significant accuracy degradation . quantization and pruning increase the difficulty of fine-tuning, requiring a more refined quantization scheme. |
| Approach: | They propose a structured pruning framework followed by a layer-wise mixed-precision quantization scheme to reduce model memory consumption during fine-tuning and inference. |
| Outcome: | Experiments on benchmark datasets show that QPruner outperforms existing methods in memory savings while maintaining or improving model performance. |
ToW: Thoughts of Words Improve Reasoning in Large Language Models (2025.naacl-long)
Copied to clipboard
Zhikun Xu, Ming Shen, Jacob Dineen, Zhaonan Li, Xiao Ye, Shijie Lu, Aswin Rrv, Chitta Baral, Ben Zhou
| Challenge: | Unlike other data augmentation methods, thoughts of words (TOW) views next-word prediction as a core reasoning task and injects fine-grained thoughts into pre-training texts. |
| Approach: | They propose a training-time data-augmentation method called thoughts of words (TOW) that injects fine-grained thoughts directly into a next-word prediction task and teaches the model to understand how the observed next word is related to previous contexts. |
| Outcome: | The proposed method reduces model hallucination by 10% and improves reasoning performance by 7% to 9% on average. |
QA‐LIGN: Aligning LLMs through Constitutionally Decomposed QA (2025.findings-emnlp)
Copied to clipboard
Jacob Dineen, Aswin Rrv, Qin Liu, Zhikun Xu, Xiao Ye, Ming Shen, Zhaonan Li, Shijie Lu, Chitta Baral, Muhao Chen, Ben Zhou
| Challenge: | QA-LIGN decomposes monolithic rewards into interpretable principle-specific evaluations . scalar rewards obscure which objectives drive the training signal . |
| Approach: | a new method decomposes monolithic rewards into interpretable principle-specific evaluations . QA-LIGN reduces attack success rates by up to 68.7% while maintaining a 0.67% false refusal rate . |
| Outcome: | QA-LIGN reduces attack success rates by up to 68.7% while maintaining a 0.67% false refusal rate . the results outperform DPO and GRPO with state-of-the-art reward models given equivalent training . |
Bitnet.cpp: Efficient Edge Inference for Ternary LLMs (2025.acl-long)
Copied to clipboard
Jinheng Wang, Hansong Zhou, Ting Song, Shijie Cao, Yan Xia, Ting Cao, Jianyu Wei, Shuming Ma, Hongyu Wang, Furu Wei
| Challenge: | 1-bit large language models have spurred interest in ternary LLMs, but efficient edge inference is still scarce. |
| Approach: | They propose an inference system optimized for 1-bit large language models . they propose a new library that facilitates sub-2-bits-per-weight inference . |
| Outcome: | The proposed inference system achieves 6.25x speed increase over full-precision baselines and 2.32x over low-bit baselines. |
RankAdaptor: Hierarchical Rank Allocation for Efficient Fine-Tuning Pruned LLMs via Performance Model (2025.findings-naacl)
Copied to clipboard
| Challenge: | Current compression techniques entail structural pruning and a recovery phase that leverages the Low-Rank Adaptation algorithm. |
| Approach: | They propose a hierarchical rank allocation method that enables efficient fine-tuning of pruned LLMs according to layerwise specific recovery requirements. |
| Outcome: | The proposed algorithm outperforms state-of-the-art methods across pruning settings and LLM architectures with improvements ranging from 0.7% to 5.5%. |