Papers by Kaiyan Zhang
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. |
ReviewRL: Towards Automated Scientific Review with RL (2025.emnlp-main)
Copied to clipboard
Sihang Zeng, Kai Tian, Kaiyan Zhang, Yuru Wang, Junqi Gao, Runze Liu, Sa Yang, Jingxuan Li, Xinwei Long, Jiaheng Ma, Biqing Qi, Bowen Zhou
| Challenge: | Existing automated review systems struggle with factual accuracy, rating consistency, and analytical depth. |
| Approach: | They propose a framework for generating comprehensive and factually grounded scientific paper reviews using supervised fine-tuning and reinforcement learning. |
| Outcome: | The proposed framework outperforms existing methods on ICLR 2025 papers. |
CoGenesis: A Framework Collaborating Large and Small Language Models for Secure Context-Aware Instruction Following (2024.acl-long)
Copied to clipboard
| Challenge: | Large-scale language models (LLMs) are increasingly exposed to private data and are becoming more and more prevalent. |
| Approach: | They propose a collaborative generation framework that integrates large and small language models to address privacy concerns logically. |
| Outcome: | The proposed framework combines large and small models to address privacy concerns logically. |
SciSketch: An Open-source Framework for Automated Schematic Diagram Generation in Scientific Papers (2025.emnlp-demos)
Copied to clipboard
| Challenge: | SCISKETCH is an open-source framework that supports two automated workflows for schematic diagram generation using foundation models. |
| Approach: | They propose an open-source framework that supports two automated workflows for schematic diagram generation using foundation models. |
| Outcome: | The open-source framework outperforms several state-of-the-art foundation models in generating schematic diagrams for scientific papers. |
PaD: Program-aided Distillation Can Teach Small Models Reasoning Better than Chain-of-thought Fine-tuning (2024.naacl-long)
Copied to clipboard
| Challenge: | Large language models excel in various tasks, but their huge size and inaccessibility of parameters present challenges for practical deployment. |
| Approach: | They propose to use CoT data to distill task-specific ability from large language models to smaller models . they use reasoning programs to suppress errors in distilled data and improve distillation quality . |
| Outcome: | The proposed model outperforms LLMs on arithmetic reasoning, symbolic reasoning, and general ability. |
Fusing Highly Specialized Language Models for Comprehensive Expertise (2025.acl-long)
Copied to clipboard
Ning Ding, Yulin Chen, Ganqu Cui, Xingtai Lv, Weilin Zhao, Kaiyan Zhang, Ruobing Xie, Bowen Zhou, Zhiyuan Liu, Maosong Sun
| Challenge: | Existing models that focus on language, programming code, and mathematical symbols are not able to achieve mastery of all three domains simultaneously. |
| Approach: | They propose to fuse highly-specialized models that are already sufficiently trained on different domains to achieve a highly-specific model. |
| Outcome: | The proposed model could achieve mastery of the three crucial domains simultaneously. |
Bypassing Neural Evaluations for Fast Audio Editing via Adaptive Trajectory Extrapolation (2026.findings-acl)
Copied to clipboard
Xiaoqian Liu, Zhengkun Ge, Jianjin Wang, Haoran Zhang, Yuan Ge, Kaiyan Chang, Chen Xu, Tong Xiao, Zhengtao Yu, Linfeng Zhang, JingBo Zhu
| Challenge: | Recent advances in audio diffusion models have significantly improved text-to-audio editing via inversion techniques, but these models typically rely on dense, fixed-step sampling trajectories to maintain structural integrity. |
| Approach: | They propose a model-agnostic Adaptive Trajectory Extrapolation framework that accelerates inversion-based editing process by dynamically evaluating only the most critical generative phases. |
| Outcome: | The proposed framework achieves a 3.9 speedup with negligible loss in fidelity. |
Intuitive Fine-Tuning: Towards Simplifying Alignment into a Single Process (2025.acl-long)
Copied to clipboard
| Challenge: | Supervised Fine-Tuning (SFT) and Preference Optimization (PO) are key processes for aligning Language Models with human preferences post pre-training. |
| Approach: | They propose to combine Supervised Fine-Tuning and Preference Optimization (PO) with two sub-processes defined at token level within the Markov Decision Process (MDP) |
| Outcome: | The proposed process performs comparably or even superiorly to SFT and some typical PO methods across several tasks, particularly those requires generation, reasoning, and fact-following abilities. |
BoB: BERT Over BERT for Training Persona-based Dialogue Models from Limited Personalized Data (2021.acl-long)
Copied to clipboard
| Challenge: | Existing persona-based dialogue models use crowd-sourced data, such as the PersonaChat . however, the cost of such datasets is limited, and the model is not robust. |
| Approach: | They propose to disentangle persona-based dialogue generation into two sub-tasks by using a BERT-over-BERT model. |
| Outcome: | The proposed model outperforms baselines in response quality and persona consistency under different limited data settings. |
Scalable Efficient Training of Large Language Models with Low-dimensional Projected Attention (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing studies have found that low-rank pre-training often compromises effectiveness. |
| Approach: | They propose to apply low-dimensional module only to the attention layer to improve both effectiveness and efficiency. |
| Outcome: | The proposed model saves 12.4% time while improving test perplexity and on downstream tasks compared with vanilla Transformer. |
Can AI Be a Good Peer Reviewer? A Survey of Peer Review Process, Evaluation, and the Future (2026.acl-long)
Copied to clipboard
Sihong Wu, Owen Jiang, Yilun Zhao, Tiansheng Hu, Yiling Ma, Kaiyan Zhang, Manasi Patwardhan, Arman Cohan
| Challenge: | Recent advances in large language models (LLMs) motivated methods that assist or automate different stages of peer review pipeline. |
| Approach: | They synthesize techniques to enhance peer review generation and after-review tasks aligned to reviews. |
| Outcome: | The proposed methods improve the peer review process by fine-tuning strategies, agent-based systems, and emerging paradigms. |
SMR: State Memory Replay for Long Sequence Modeling (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing state space models (SSMs) address non-uniform sampling, but their recursive structures impede efficient SSM computation via convolution. |
| Approach: | They propose a plug-and-play mechanism to solve the Non-Stable State problem by adjusting input sequences with early memories. |
| Outcome: | The proposed method overcomes the non-uniform sample processing problem . it can achieve Sampling Step Adaptation (SSA) by adjusting input sequences with early memories. |
CRaSh: Clustering, Removing, and Sharing Enhance Fine-tuning without Full Large Language Model (2023.emnlp-main)
Copied to clipboard
| Challenge: | Instruction tuning is an effective way of aligning large language models with private instruction data. |
| Approach: | They propose a training-free strategy to derive improved emulators from LLMs by using Offsite-Tuning (OFT) they propose CRaSh, which transfers transformer blocks between centralized LLM and downstream emulators . |
| Outcome: | The proposed technique boosts performance of large language models with billions of parameters. |