|
My name is Zhongzhi Li, and I am a Ph.D student in AI4Math and LLM Reasoning at CASIA. I pursuing my Ph.D in Artificial Intelligence from the Institute of Automation, Chinese Academy of Sciences (CASIA), under the supervision of IEEE/CAAI/CAA/IAPR Fellow, Chenglin Liu My past research focused on LLM Reasoning in Math, Code and Neural-Symbolic Systems, Brain-Inspired System-2 Intelligence, and Continual Learning in LLM Reasoning. Currently, I am exploring System-2 Reasoning in (Multimodal) LLMs, LLM Agents, and LRM4Science, with applications in fields like Math, Code, Medicine, and beyond. I closely collaborate with Zhijiang Guo in HKUST, Duzhen Zhang in MBZUAI, Xiao Liang in UCLA, Xiao Liu in MSRA. I am looking for cooperation. Contact me if you are interested in the above topics. Email / Google Scholar / GitHub / Twitter / LinkedIn |
|
|
|
- [2025.05] 📣 Welcome to follow our latest work on Synthetic Reinforced LLM Reasoning: SwS
- [2025.05] 📣 Welcome to follow our latest work on Efficient System-2 LLM Reasoning: TLDR
- [2025.05] 📣 Welcome to follow our latest survey on: LLM Reasoning Safety
- [2025.05] 📣 Welcome to follow our latest work on Reinforced Multimodal LLM Reasoning: PeRL
- [2025.05] 🎉 Two papers: BranchLoRA, and LongDocURL are accepted by ACL2025. See you in Vienna!
- [2025.03] 🎉 One paper: MV-MATH is accepted by CVPR2025.
- [2025.02] 📣 Welcome to follow our latest survey on LLM Reasoning.
- [2025.02] 📣 Welcome to follow our latest techniques report on R1-like model RedStar.
- [2024.11] 🎉 One papers: CMMaTH are accepted by COLING2025. See you in Abu Dhabi!
- [2024.05] 🎉 Two paper: LANS and GeoEval are accepted by ACL2024. See you in Bangkok!
|
|
- Brain-inspired Intelligence, System-2 Reasoning: [Preprint-SwS], [Preprint-TLDR], [Preprint-ManipLVM-R1] [Preprint-PeRL], [Preprint-SoliGeo], [LLM Reasoning Safety Survey], [System-2 Reasoning Survey], [Tech Report]
- Continual Learning in LLMs Reasoning & Planning: [ACL2025]. [Preprint-LifelongAgentBench].
- Reasoning in Multimodal Learning & MLLM: [ACL 2024] [ACL 2024] [ACL 2025] [CVPR 2025] [Preprint-PeRL]
|
* denotes equal contribution and # denotes the corresponding author. |
Conference Paper
|
Zhong-Zhi Li*, Du-Zhen Zhang*, Ming-Liang Zhang, Jiaxin Zhang, Zengyan Liu, Yuxuan Yao, Haotian Xu, Junhao Zheng, Xiuyi Chen, Yingying Zhang, Fei Yin, Jiahua Dong, Zhijiang Guo#, Le Song#, Cheng-Lin Liu# Preprint Under Review. Paper / Code DetailsAchieving human-level intelligence requires refining the transition from the fast, intuitive System 1 to the slower, more deliberate System 2 reasoning. While System 1 excels in quick, heuristic decisions, System 2 relies on logical reasoning for more accurate judgments and reduced biases. Foundational Large Language Models (LLMs) excel at fast decision-making but lack the depth for complex reasoning, as they have not yet fully embraced the step-by-step analysis characteristic of true System 2 thinking. Recently, reasoning LLMs like OpenAI's o1/o3 and DeepSeek's R1 have demonstrated expert-level performance in fields such as mathematics and coding, closely mimicking the deliberate reasoning of System 2 and showcasing human-like cognitive abilities. This survey begins with a brief overview of the progress in foundational LLMs and the early development of System 2 technologies, exploring how their combination has paved the way for reasoning LLMs. Next, we discuss how to construct reasoning LLMs, analyzing their features, the core methods enabling advanced reasoning, and the evolution of various reasoning LLMs. Additionally, we provide an overview of reasoning benchmarks, offering an in-depth comparison of the performance of representative reasoning LLMs. Finally, we explore promising directions for advancing reasoning LLMs and maintain a real-time GitHub Repository to track the latest developments. We hope this survey will serve as a valuable resource to inspire innovation and drive progress in this rapidly evolving field. |
|
Xiao Liang*, Zhong-Zhi Li*, Yeyun Gong, Yang Wang, Hengyuan Zhang, Yelong Shen, Ying Nian Wu, Weizhu Chen Preprint Under Review. Paper / Code DetailsReinforcement Learning with Verifiable Rewards (RLVR) has proven effective for training large language models (LLMs) on complex reasoning tasks, such as mathematical problem solving. A prerequisite for the scalability of RLVR is a high-quality problem set with precise and verifiable answers. However, the scarcity of well-crafted human-labeled math problems and limited-verification answers in existing distillation-oriented synthetic datasets limit their effectiveness in RL. Additionally, most problem synthesis strategies indiscriminately expand the problem set without considering the model’s capabilities, leading to low efficiency in generating useful questions. To mitigate this issue, we introduce a Self-aware Weakness-driven problem Synthesis framework (SwS) that systematically identifies model deficiencies and leverages them for problem augmentation. Specifically, we define weaknesses as questions that the model consistently fails to learn through its iterative sampling during RL training. We then extract the core concepts from these failure cases and synthesize new problems to strengthen the model’s weak areas in subsequent augmented training, enabling it to focus on and gradually overcome its weaknesses. Without relying on external knowledge distillation, our framework enables robust generalization by empowering the model to self-identify and address its weaknesses in RL, yielding average performance gains of 10.0% and 7.7% on 7B and 32B models across eight mainstream reasoning benchmarks. |
|
Zhong-Zhi Li*, Xiao Liang*, Zihao Tang, Lei Ji, Peijie Wang, Haotian Xu, Xing W, Haizhen Huang, Weiwei Deng, Yeyun Gong, Zhijiang Guo#, Xiao Liu#, Fei Yin, Cheng-Lin Liu# Preprint Under Review. Paper / Code DetailsLarge Language Models (LLMs) have recently achieved remarkable progress by leveraging Reinforcement Learning and extended Chain-of-Thought (CoT) techniques. However, the challenge of performing efficient language reasoning--especially during inference with extremely long outputs--has drawn increasing attention from the research community. In this work, we propose a dynamic ratio-based training pipeline that does not rely on sophisticated data annotations or interpolation between multiple models. We continuously balance the weights between the model's System-1 and System-2 data to eliminate redundant reasoning processes while preserving the model's reasoning capability. We validate our approach across models on DeepSeek-R1-Distill-7B and DeepSeek-R1-Distill-14B and on a diverse set of benchmarks with varying difficulty levels. Our method significantly reduces the number of output tokens by nearly 40% while maintaining the accuracy of the reasoning. Our code and data will be available soon. |
|
Zirui Song, Guangxian Ouyang, Mingzhe Li, Yuheng Ji, Chenxi Wang, Zixiang Xu, Zeyu Zhang, Xiaoqing Zhang, Qian Jiang, Zhenhao Chen, Zhong-Zhi Li, Rui Yan, Xiuying Chen Preprint Under Review. Paper / Code DetailsLarge Vision-Language Models (LVLMs) have recently advanced robotic manipulation by leveraging vision for scene perception and language for instruction following. However, existing methods rely heavily on costly human-annotated training datasets, which limits their generalization and causes them to struggle in out-of-domain (OOD) scenarios, reducing real-world adaptability. To address these challenges, we propose ManipLVM-R1, a novel reinforcement learning framework that replaces traditional supervision with Reinforcement Learning using Verifiable Rewards (RLVR). By directly optimizing for task-aligned outcomes, our method enhances generalization and physical reasoning while removing the dependence on costly annotations. Specifically, we design two rule-based reward functions targeting key robotic manipulation subtasks: an Affordance Perception Reward to enhance localization of interaction regions, and a Trajectory Match Reward to ensure the physical plausibility of action paths. These rewards provide immediate feedback and impose spatial-logical constraints, encouraging the model to go beyond shallow pattern matching and instead learn deeper, more systematic reasoning about physical interactions. |
|
Junhao Zheng, Xidi Cai, Qiuke Li, Duzhen Zhang, ZhongZhi Li, Yingying Zhang, Le Song, Qianli Ma Preprint Under Review. Paper / Code DetailsLifelong learning is essential for intelligent agents operating in dynamic environments. Current large language model (LLM)-based agents, however, remain stateless and unable to accumulate or transfer knowledge over time. Existing benchmarks treat agents as static systems and fail to evaluate lifelong learning capabilities. We present LifelongAgentBench, the first unified benchmark designed to systematically assess the lifelong learning ability of LLM agents. It provides skill-grounded, interdependent tasks across three interactive environments, Database, Operating System, and Knowledge Graph, with automatic label verification, reproducibility, and modular extensibility. Extensive experiments reveal that conventional experience replay has limited effectiveness for LLM agents due to irrelevant information and context length constraints. We further introduce a group self-consistency mechanism that significantly improves lifelong learning performance. We hope LifelongAgentBench will advance the development of adaptive, memory-capable LLM agents. |
|
Duzhen Zhang*, Yong Ren*, Zhong-Zhi Li, Yahan Yu, Jiahua Dong, Chenxing Li, Zhilong Ji, Jinfeng Bai ACL2025 Paper / Code DetailsMultimodal Continual Instruction Tuning (MCIT) aims to finetune Multimodal Large Language Models (MLLMs) to continually align with human intent across sequential tasks. Existing approaches often rely on the Mixture-of-Experts (MoE) LoRA framework to preserve previous instruction alignments. However, these methods are prone to Catastrophic Forgetting (CF), as they aggregate all LoRA blocks via simple summation, which compromises performance over time. In this paper, we identify a critical parameter inefficiency in the MoELoRA framework within the MCIT context. Based on this insight, we propose BranchLoRA, an asymmetric framework to enhance both efficiency and performance. To mitigate CF, we introduce a flexible tuning-freezing mechanism within BranchLoRA, enabling branches to specialize in intra-task knowledge while fostering inter-task collaboration. Moreover, we incrementally incorporate task-specific routers to ensure an optimal branch distribution over time, rather than favoring the most recent task. To streamline inference, we introduce a task selector that automatically routes test inputs to the appropriate router without requiring task identity. Extensive experiments with the latest MCIT benchmark demonstrate that BranchLoRA significantly outperforms MoELoRA. Additionally, BranchLoRA consistently shows its superiority when scaled to different MLLM backbones and sizes. |
|
Zhong-Zhi Li*, Ming-Liang Zhang*, Fei Yin, Cheng-Lin Liu ACL2024 Paper / Code DetailsGeometry problem solving (GPS) is a challenging mathematical reasoning task requiring multi-modal understanding, fusion, and reasoning. Existing neural solvers take GPS as a vision-language task but are short in the representation of geometry diagrams that carry rich and complex layout information. In this paper, we propose a layout-aware neural solver named LANS, integrated with two new modules: multimodal layout-aware pre-trained language module (MLA-PLM) and layout-aware fusion attention (LA-FA). MLA-PLM adopts structural-semantic pre-training (SSP) to implement global relationship modeling, and point-match pre-training (PMP) to achieve alignment between visual points and textual points. LA-FA employs a layout-aware attention mask to realize point-guided cross-modal fusion for further boosting layout awareness of LANS. Extensive experiments on datasets Geometry3K and PGPS9K validate the effectiveness of the layout-aware modules and superior problem-solving performance of our LANS solver, over existing symbolic and neural solvers. |
|
Zhong-Zhi Li*, Jiaxin Zhang*, Mingliang Zhang, Fei Yin, Chenglin Liu, Yashar Moshfeghi ACL2024 Paper / Code DetailsRecent advancements in large language models (LLMs) and multi-modal models (MMs) have demonstrated their remarkable capabilities in problem-solving. Yet, their proficiency in tackling geometry math problems, which necessitates an integrated understanding of both textual and visual information, has not been thoroughly evaluated. To address this gap, we introduce the GeoEval benchmark, a comprehensive collection that includes a main subset of 2,000 problems, a 750 problems subset focusing on backward reasoning, an augmented subset of 2,000 problems, and a hard subset of 300 problems. This benchmark facilitates a deeper investigation into the performance of LLMs and MMs in solving geometry math problems. Our evaluation of ten LLMs and MMs across these varied subsets reveals that the WizardMath model excels, achieving a 55.67\% accuracy rate on the main subset but only a 6.00\% accuracy on the hard subset. This highlights the critical need for testing models against datasets on which they have not been pre-trained. Additionally, our findings indicate that GPT-series models perform more effectively on problems they have rephrased, suggesting a promising method for enhancing model capabilities. |
|
Zhongzhi Li, Ming-Liang Zhang, Fei Yin, Zhi-Long Ji, Jin-Feng Bai, Zhen-Ru Pan, Fan-Hu Zeng, Jian Xu, Jia-Xin Zhang, Cheng-Lin Liu COLING2025 Paper / Code DetailsDue to the rapid advancements in multimodal large language models, evaluating their multimodal mathematical capabilities continues to receive wide attention. Despite the datasets like MathVista proposed benchmarks for assessing mathematical capabilities in multimodal scenarios, there is still a lack of corresponding evaluation tools and datasets for fine-grained assessment in the context of K12 education in Chinese language. To systematically evaluate the capability of multimodal large models in solving Chinese multimodal mathematical problems, we propose a Chinese Multi-modal Math Skill Evaluation Benchmark, named CMMaTH, contraining 23k multimodal K12 math related questions, forming the largest Chinese multimodal mathematical problem benchmark to date. CMMaTH questions from elementary to high school levels, provide increased diversity in problem types, solution objectives, visual elements, detailed knowledge points, and standard solution annotations. We have constructed an open-source tool GradeGPT integrated with the CMMaTH dataset, facilitating stable, rapid, and cost-free model evaluation. Our data and code are available. |
|
Main Contirbuter Tech Report Paper / Code DetailsRecent advancements in large language models (LLMs) and multi-modal models (MMs) have demonstrated their remarkable capabilities in problem-solving. Yet, their proficiency in tackling geometry math problems, which necessitates an integrated understanding of both textual and visual information, has not been thoroughly evaluated. To address this gap, we introduce the GeoEval benchmark, a comprehensive collection that includes a main subset of 2,000 problems, a 750 problems subset focusing on backward reasoning, an augmented subset of 2,000 problems, and a hard subset of 300 problems. This benchmark facilitates a deeper investigation into the performance of LLMs and MMs in solving geometry math problems. Our evaluation of ten LLMs and MMs across these varied subsets reveals that the WizardMath model excels, achieving a 55.67\% accuracy rate on the main subset but only a 6.00\% accuracy on the hard subset. This highlights the critical need for testing models against datasets on which they have not been pre-trained. Additionally, our findings indicate that GPT-series models perform more effectively on problems they have rephrased, suggesting a promising method for enhancing model capabilities. |
|
Ming-Liang Zhang, Zhong-Zhi Li, Fei Yin, Liang Lin, Cheng-Lin Liu Preprint Under Review. Paper / Code DetailsGeometry problem solving (GPS) requires capacities of multi-modal understanding, multi-hop reasoning and theorem knowledge application. In this paper, we propose a neural-symbolic model for plane geometry problem solving (PGPS), named PGPSNet-v2, with three key steps: modal fusion, reasoning process and knowledge verification. In modal fusion, we leverage textual clauses to express fine-grained structural and semantic content of geometry diagram, and fuse diagram with textual problem efficiently through structural-semantic pre-training. For reasoning, we design an explicable solution program to describe the geometric reasoning process, and employ a self-limited decoder to generate solution program autoregressively. To reduce solution errors, a multi-level theorem verifier is proposed to eliminate solutions that do not match geometric principles, alleviating the hallucination of the neural model. We also construct a large-scale geometry problem dataset called PGPS9K, containing fine-grained annotations of textual clauses, solution program and involved knowledge tuples. Extensive experiments on datasets Geometry3K and PGPS9K show that our PGPSNet solver outperforms existing symbolic and neural solvers in GPS performance, while maintaining good explainability and reliability, and the solver components (fusion, reasoning, verification) are all justified effective. |
|
Cheng Wang, Yue Liu, Baolong Bi, Duzhen Zhang, Zhongzhi Li, Junfeng Fang, Bryan Hooi Preprint Under Review. Paper / Code DetailsLarge Reasoning Models (LRMs) have exhibited extraordinary prowess in tasks like mathematics and coding, leveraging their advanced reasoning capabilities. Nevertheless, as these capabilities progress, significant concerns regarding their vulnerabilities and safety have arisen, which can pose challenges to their deployment and application in real-world settings. This paper presents a comprehensive survey of LRMs, meticulously exploring and summarizing the newly emerged safety risks, attacks, and defense strategies. By organizing these elements into a detailed taxonomy, this work aims to offer a clear and structured understanding of the current safety landscape of LRMs, facilitating future research and development to enhance the security and reliability of these powerful models. |
|
Conference Reviewers: CVPR2022, ACL2025, KDD2025 EMNLP2025, ICCV2025 Journal Reviewers: Pattern Recognition, IEEE TMM |
Website template
Last updated: May 2025