Papers by Zicheng Liu
CriticBench: Benchmarking LLMs for Critique-Correct Reasoning (2024.findings-acl)
Copied to clipboard
| Challenge: | CriticBench is a benchmark designed to assess LLMs’ abilities to critique and refine their reasoning across a variety of tasks. |
| Approach: | They propose a benchmark to assess LLMs' ability to critique and correct reasoning across a variety of tasks. |
| Outcome: | The proposed benchmark examines the performance of 17 large language models in generation, critique, and correction reasoning. |
Self-Taught Agentic Long Context Understanding (2025.acl-long)
Copied to clipboard
Yufan Zhuang, Xiaodong Yu, Jialian Wu, Ximeng Sun, Ze Wang, Jiang Liu, Yusheng Su, Jingbo Shang, Zicheng Liu, Emad Barsoum
| Challenge: | Extensive experiments across seven long-context tasks demonstrate that AgenticLU significantly outperforms state-of-the-art prompting methods and specialized long-consumer LLMs. |
| Approach: | They propose a framework to enhance an LLM's understanding of long-context questions by integrating targeted self-clarification with contextual grounding within an agentic workflow. |
| Outcome: | The proposed framework outperforms state-of-the-art prompting methods and specialized long-context LLMs in seven long-constitut tasks. |
Agent Laboratory: Using LLM Agents as Research Assistants (2025.findings-emnlp)
Copied to clipboard
Samuel Schmidgall, Yusheng Su, Ze Wang, Ximeng Sun, Jialian Wu, Xiaodong Yu, Jiang Liu, Michael Moor, Zicheng Liu, Emad Barsoum
| Challenge: | Agent Laboratory is an autonomous LLM-based framework that can complete the entire research process. |
| Approach: | Agent Laboratory is an autonomous LLM-based framework that can complete the entire research process. |
| Outcome: | Agent Laboratory is an autonomous LLM-based framework that can complete the entire research process. |
Stabilizing Efficient Reasoning with Step-Level Advantage Selection (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models generate long and verbose reasoning traces at inference time . short context post-training alone induces substantial reasoning compression . |
| Approach: | They propose a step-level advantage selection approach that reduces reasoning length by over 30% . they propose to use GRPO without any length-aware objective to train models in a shorter context window . |
| Outcome: | The proposed approach reduces average reasoning length by over 30% while improving Pass@1 accuracy by 3.79 points over the strongest length-aware baseline. |
Beyond Logits: Aligning Feature Dynamics for Effective Knowledge Distillation (2025.acl-long)
Copied to clipboard
Guoqiang Gong, Jiaxing Wang, Jin Xu, Deping Xiang, Zicheng Zhang, Leqi Shen, Yifeng Zhang, JunhuaShu JunhuaShu, ZhaolongXing ZhaolongXing, Zhen Chen, Pengzhang Liu, Ke Zhang
| Challenge: | Knowledge distillation (KD) compresses large language models into lightweight versions called student models. |
| Approach: | They propose to align the entire feature dynamics between teacher and student models by using two additional loss terms to achieve this. |
| Outcome: | The proposed method matches the entire feature dynamics between teacher and student models rather than just the final states. |
Taming LLMs with Gradient Grouping (2025.acl-long)
Copied to clipboard
| Challenge: | a new study presents scaling with gradient grouping (SGG) the adaptive learning rate scaling approach is based on per-parameter statistics, which incurs memory overhead. |
| Approach: | They propose an optimizer wrapper that improves adaptive learning rate estimation by dynamic grouping and group-specific scaling. |
| Outcome: | The proposed algorithm improves learning rate estimation on diverse models with different model sizes and batch sizes. |
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 . |
AdaptEvolve: Improving Efficiency of Evolutionary AI Agents through Adaptive Model Selection (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing routing strategies rely on static heuristics or external controllers to optimize performance. |
| Approach: | They propose a framework that leverages intrinsic generation confidence to estimate solvability. |
| Outcome: | Empirical results show that confidence-driven selection yields favorable Pareto frontier . computational cost of state-of-the-art large language models remains a key barrier to scalable deployment . |
TTT-Bench: A Benchmark for Evaluating Reasoning Ability with Simple and Novel Tic-Tac-Toe-style Games (2025.emnlp-main)
Copied to clipboard
| Challenge: | Recent advances in large reasoning models (LRMs) have driven significant breakthroughs across various reasoning tasks including deductive, arithmetic, commonsense, relational, and symbolic reasoning. |
| Approach: | They propose a programmatic approach to evaluate basic strategic, spatial, and logical reasoning abilities in large reasoning models through four two-player Tic-Tac-Toe-style games that humans can effortlessly solve from a young age. |
| Outcome: | The proposed model performs 41% lower on TTT-Bench than MATH 500 and AIME 2024 models, while the larger models perform better on longer reasoning traces. |
DRIFT: Transferring Reasoning Priors for Efficient MLLM Fine-Tuning (2026.findings-acl)
Copied to clipboard
Chao Huang, Zeliang Zhang, Jiang Liu, Ximeng Sun, Jialian Wu, Xiaodong Yu, Ze Wang, Chenliang Xu, Emad Barsoum, Zicheng Liu
| Challenge: | Multimodal large language models (MLLMs) have made rapid progress in perception and alignment, but their reasoning ability often lags behind strong text-only LLMs. |
| Approach: | They propose a method that transfers reasoning knowledge in the gradient space while preserving multimodal alignment. |
| Outcome: | Experiments on multimodal reasoning benchmarks show that DRIFT outperforms naive merging and standard SFT. |
Redundancy Principles for MLLMs Benchmarks (2025.acl-long)
Copied to clipboard
Zicheng Zhang, Xiangyu Zhao, Xinyu Fang, Chunyi Li, Xiaohong Liu, Xiongkuo Min, Haodong Duan, Kai Chen, Guangtao Zhai
| Challenge: | Rapid growth of Multi-modality Large Language Models has led to significant redundancy among benchmarks. |
| Approach: | They propose a framework to improve MLLM benchmark design by identifying redundancy at three levels: dimension, instance, and cross-benchmark redundancies. |
| Outcome: | The proposed framework streamlines evaluations and enhances reliability. |
NUWA-XL: Diffusion over Diffusion for eXtremely Long Video Generation (2023.acl-long)
Copied to clipboard
Shengming Yin, Chenfei Wu, Huan Yang, Jianfeng Wang, Xiaodong Wang, Minheng Ni, Zhengyuan Yang, Linjie Li, Shuguang Liu, Fan Yang, Jianlong Fu, Ming Gong, Lijuan Wang, Zicheng Liu, Houqiang Li, Nan Duan
| Challenge: | Existing work generates long videos segment by segment sequentially, which is inefficient. |
| Approach: | They propose a Diffusion over Difference architecture for eXtremely Long video generation. |
| Outcome: | The proposed architecture reduces the average inference time from 7.55min to 26s (94.26%) and generates high-quality long videos with both global and local coherence. |
TaDA: Training-free recipe for Decoding with Adaptive KV Cache Compression and Mean-centering (2025.acl-industry)
Copied to clipboard
| Challenge: | key-value caches in large language models consume memory, posing a major challenge for scalable deployment. |
| Approach: | They propose a training-free recipe for KV cache compression with quantization precision that adapts to error sensitivity across layers and a mean centering to eliminate separate outlier handling. |
| Outcome: | The proposed technique reduces the KV cache memory footprint to 27% of the original 16-bit baseline while achieving comparable accuracy. |