Hi, I am Encheng Su.

Encheng Su^*, Pengze Li^*, Ziheng Hua^*, Qiucheng Yu, Jianyu Wu, Chen Tang, Yizhou Wang, Lintao Wang, Junjia Xiang, Xinzhu Ma, Xi Chen, Shixiang Tang, Houqiang Li, Xi Chen, Aoran Wang

^* Equal contribution.

BRIDGE uses dependency graphs to locate handoff failures in multi-agent workflows, selects repairs using calibrated evidence, and verifies each repair before releasing it. It improves task completion while limiting unnecessary interventions and token overhead.

Encheng Su^*, Xinping Liu^*, Yizhou Wang, Bozitao Zhong, Pengze Li, Jianyu Wu, Mingxiao Li, Xinzhu Ma, Aoran Wang, Liang Hong, Wanli Ouyang, Chen Tang, Houqiang Li, Shixiang Tang

^* Equal contribution.

PROTEAN benchmarks text-guided protein structure editing across 380 subtasks in six intervention families. It separately evaluates geometric compliance and biophysical realizability, revealing scope leakage and refolding failures as major bottlenecks for LLM-driven design agents.

Encheng Su, Jianyu Wu, Chen Tang, Lintao Wang, Pengze Li, Aoran Wang, Jinouwen Zhang, Yizhou Wang, Yuan Meng, Xinzhu Ma, Shixiang Tang, Houqiang Li

SciIF evaluates whether models can satisfy the explicit constraints that make scientific answers rigorous—not only produce a plausible final answer. It provides structured verification for conditions, units, assumptions, and required solution processes.

Encheng Su, Jianyu Wu, Jinouwen Zhang, Qiucheng Yu, Chen Tang, Pengze Li, Lintao Wang, Aoran Wang, Xinzhu Ma, Shixiang Tang, Yizhou Wang, Houqiang Li

S3Mem writes long interaction histories into structured scene-event units and retrieves compact evidence chains for question answering. The system targets accurate, token-efficient memory use in long-horizon agents.

Lintao Wang^*, Encheng Su^*, Jiaqi Liu, Pengze Li, Jiabei Xiao, Wenlong Zhang, Xinnan Dai, Xi Chen, Yuan Meng, Lei Bai, Wanli Ouyang, Shixiang Tang, Aoran Wang, Xinzhu Ma

^* Equal contribution.

PhysUniBench evaluates conceptual, mathematical, and diagram-based reasoning across undergraduate physics. It is designed to expose failures that simpler answer-only benchmarks miss.

Encheng Su, Hu Cao, Alois Knoll

BiSeg-SAM introduces a weakly supervised post-processing framework that refines SAM outputs for medical binary segmentation, reducing dependence on expensive pixel-level annotations. The work was presented as an oral paper at BIBM 2024.

Xuantuo Huang, Encheng Su, Aoran Wang, Pan Tan, Shengyong Xu, Yizhou Wang, Chen Tang, Shixiang Tang, Mingxiao Li

ADT encodes small molecules with one structural token per atom and proteins with one token per residue, jointly preserving backbone and side-chain geometry. A diffusion decoder recovers all-atom structures from these compact representations, supporting both reconstruction and autoregressive molecular generation.

Pengze Li, Jiaquan Zhang, Encheng Su, Yunbo Long, Zihang Zeng, Xinping Liu, Han Zhang, Junchi Yu, Shixiang Tang, Aoran Wang, Xi Chen

This framework improves an executable tool library and an LLM's tool-use policy through shared, verifiable fit-error feedback. Failure-driven tool development and training on successful trajectories complement each other to improve scientific equation discovery.

Yizhou Wang, Chen Tang, Han Deng, Jiabei Xiao, Jiaqi Liu, Jianyu Wu, Jun Yao, Pengze Li, Encheng Su, Lintao Wang, et al.

SciReasoner aligns language with heterogeneous scientific representations and trains deliberate reasoning across multiple scientific capability families and disciplines.

Ming Hu, Chenglong Ma, Wei Li, Wanghan Xu, Jiamin Wu, Jucheng Hu, Tianbin Li, Guohang Zhuang, Jiaqi Liu, Yingzhou Lu, …, Encheng Su, et al.

This data-centric survey organizes scientific LLMs, datasets, benchmarks, and domain-specific representations, then traces the field toward closed-loop autonomous research agents.

Chen Tang, Xinzhu Ma, Encheng Su, Xiufeng Song, Xiaohong Liu, Wei-Hong Li, Lei Bai, Wanli Ouyang, Xiangyu Yue

UniSTD is a unified Transformer framework that learns across ten spatio-temporal tasks in four disciplines, reducing dependence on task-specific architectures and training pipelines.

Chen Tang, Yizhou Wang, Jianyu Wu, Lintao Wang, Shixiang Tang, Pengze Li, Encheng Su, Jun Yao, Jiabei Xiao, Yuqi Shi, et al.

This work studies native structural reasoning for explaining structure-property relationships across biology, chemistry, and materials science with greater accuracy and transparency.

Songyan Xie, Jinghang Wen, Encheng Su, Qiucheng Yu

The paper demonstrates a stealthy black-box physical attack on near-infrared face recognition using infrared-absorbing ink patterns that remain difficult for people to perceive.

Pengze Li, Jiaquan Zhang, Yunbo Long, Xinping Liu, Wenjie Zhou, Encheng Su, Zihang Zeng, Jiaqi Liu, Jiyao Liu, Junchi Yu, Lihao Liu, Philip Torr, Shixiang Tang, Aoran Wang, Xi Chen

This work introduces visual-to-symbolic analytical inference: recovering executable symbolic solutions directly from scientific field visualizations and limited metadata.

Jianyu Wu, Yizhou Wang, Encheng Su, Chen Tang, Shixiang Tang

DAPD addresses the “privilege illusion” in on-policy self-distillation by aligning teacher and student behavior along two matched-information paths.

Zixian Su, Hongkai Zhang, Fan Gao, Encheng Su, Taiping Qu, Jingwei Guo, Nan Zhang, Hui Wang, Zhen Zhou, Kairui Bo, Yan Chen, Yue Ren, Shuai Li, Lei Xu, Henggui Zhang

CardioLens provides a leakage-resistant, multi-sequence cardiac MRI evaluation designed to reveal where strong benchmark results fail to translate into clinical usefulness. Accepted to the NeurIPS 2026 Evaluations and Datasets Track.

Yiheng Wang, Yixin Chen, Shuo Li, Yifan Zhou, Bo Liu, Hengjian Gao, Jiakang Yuan, Jia Bu, Wanghan Xu, Yuhao Zhou, …, Encheng Su, et al.

SciEvalKit is an extensible open-source platform for reproducible evaluation across scientific disciplines, model families, datasets, and capability categories.