Georgia Institute of Technology · GTISC Lab · Atlanta

Kejing Xia 夏可敬

I am a Computer Science Ph.D. student at Georgia Tech, advised by Prof. Wenke Lee in the GTISC Lab. My research centers on LLM post-training and agentic AI, with particular interests in reasoning, diffusion language models, and systems that make language models more useful in interactive environments.

Before joining the Ph.D. program, I earned my B.Eng. in Communication Engineering (Artificial Intelligence Track) from Wuhan University.

Kejing Xia

Research

Explicit and Implicit Reasoning

I study how language models reason and solve problems, spanning explicit reasoning steps, implicit representations, and the use of memory.

Efficient Language Models

I explore efficient and scalable language models, balancing computational cost, model capabilities, and the ability to handle long contexts.

Language Model Agents

I study language model agents that reason, act, and interact with their environments, aiming to make them more capable and reliable across tasks.

Industry research

Publications

Language models

2026

MetaState: Persistent Working Memory Enhances Reasoning in Discrete Diffusion Language Models

COLM · 2026

To preserve information between denoising steps, MetaState trains recurrent working-memory modules around a frozen diffusion language model, improving mathematical reasoning and code generation.

Paper: MetaState Code: MetaState Project: MetaState
BibTeX: MetaState
@misc{metastate2026,
  title = {MetaState: Persistent Working Memory Enhances Reasoning in Discrete Diffusion Language Models},
  author = {Kejing Xia and Mingzhe Li and Lixuan Wei and Zhenbang Du and Xiangchi Yuan and Dachuan Shi and Qirui Jin and Wenke Lee},
  year = {2026},
  eprint = {2603.01331},
  archivePrefix = {arXiv},
  primaryClass = {cs.CL},
  url = {https://arxiv.org/abs/2603.01331}
}

R²-dLLM: Accelerating Diffusion Large Language Models via Spatio-Temporal Redundancy Reduction

arXiv · 2026

To reduce redundant diffusion steps, R²-dLLM combines training-free spatial and temporal decoding rules with redundancy-aware supervised fine-tuning, improving the trade-off between computation and generation quality.

Paper: R²-dLLM Code: R²-dLLM Project: R²-dLLM
BibTeX: R²-dLLM
@misc{r2dllm2026,
  title = {{R$^2$-dLLM}: Accelerating Diffusion Large Language Models via Spatio-Temporal Redundancy Reduction},
  author = {Zhenbang Du and Kejing Xia and Xinrui Zhong and Yonggan Fu and Nicolai Oswald and Binfei Ji and Brucek Khailany and Pavlo Molchanov and Yingyan (Celine) Lin},
  year = {2026},
  eprint = {2604.18995},
  archivePrefix = {arXiv},
  primaryClass = {cs.CL},
  url = {https://arxiv.org/abs/2604.18995}
}

2025

D²GS: Dense Depth Regularization for LiDAR-free Urban Scene Reconstruction

NeurIPS · 2025

To reconstruct urban scenes without LiDAR, D²GS jointly refines dense depth and Gaussian geometry using diffusion priors, with progressive point pruning and road-specific geometric constraints.

Paper: D²GS Code: D²GS Project: D²GS
BibTeX: D²GS
@inproceedings{NEURIPS2025_524ff06d,
 author = {Xia, Kejing and Jia, Jidong and Jin, Ke and BAI, Yucai and Sun, Li and Tao, Dacheng and Zhang, Youjian},
 booktitle = {Advances in Neural Information Processing Systems},
 doi = {10.52202/085713-1905},
 editor = {D. Belgrave and C. Zhang and H. Lin and R. Pascanu and P. Koniusz and M. Ghassemi and N. Chen},
 pages = {56899--56927},
 publisher = {Curran Associates, Inc.},
 title = {D\^{}2GS: Dense Depth Regularization for LiDAR-free Urban Scene Reconstruction},
 url = {https://proceedings.neurips.cc/paper_files/paper/2025/file/524ff06d5375d76e93d8490e654a81bf-Paper-Conference.pdf},
 volume = {38, Main Conference},
 year = {2025}
}

EDITOR: Effective and Interpretable Prompt Inversion for Text-to-Image Diffusion Models

arXiv · 2025

To recover interpretable prompts from generated images, EDITOR combines caption-based initialization, latent-space optimization, and embedding-to-text conversion, balancing visual reconstruction with readable textual descriptions.

Paper: EDITOR Code: EDITOR Project: EDITOR
BibTeX: EDITOR
@misc{editor2025,
  title = {EDITOR: Effective and Interpretable Prompt Inversion for Text-to-Image Diffusion Models},
  author = {Mingzhe Li and Kejing Xia and Gehao Zhang and Zhenting Wang and Guanhong Tao and Siqi Pan and Juan Zhai and Shiqing Ma},
  year = {2025},
  eprint = {2506.03067},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV},
  url = {https://arxiv.org/abs/2506.03067}
}

LongMamba: Enhancing Mamba's Long-Context Capabilities via Training-Free Receptive Field Enlargement

ICLR · 2025

To counter memory decay at long context lengths, LongMamba identifies global state channels and filters unimportant tokens, enlarging receptive fields without additional training.

Paper: LongMamba Code: LongMamba Project: LongMamba
BibTeX: LongMamba
@inproceedings{ICLR2025_ab5d50d2,
 author = {Ye, Zhifan and Xia, Kejing and Fu, Yonggan and Dong, Xin and Hong, Jihoon and Yuan, Xiangchi and Diao, Shizhe and Kautz, Jan and Molchanov, Pavlo and Lin, Yingyan Celine},
 booktitle = {International Conference on Learning Representations},
 editor = {Y. Yue and A. Garg and N. Peng and F. Sha and R. Yu},
 pages = {68486--68504},
 title = {LongMamba: Enhancing Mamba\textquotesingle s Long-Context Capabilities via Training-Free Receptive Field Enlargement},
 url = {https://proceedings.iclr.cc/paper_files/paper/2025/file/ab5d50d269e52f8eed497062311ff173-Paper-Conference.pdf},
 volume = {2025},
 year = {2025}
}

LAMB: A Training-Free Method to Enhance the Long-Context Understanding of SSMs via Attention-Guided Token Filtering

ACL Main · 2025

By examining attention patterns in state space models, LAMB guides token filtering to preserve critical information, strengthening long-context understanding in pure and hybrid models without training.

Paper: LAMB Code: LAMB Project: LAMB
BibTeX: LAMB
@inproceedings{ye-etal-2025-lamb,
    title = "{LAMB}: A Training-Free Method to Enhance the Long-Context Understanding of {SSM}s via Attention-Guided Token Filtering",
    author = "Ye, Zhifan  and
      Wang, Zheng  and
      Xia, Kejing  and
      Hong, Jihoon  and
      Li, Leshu  and
      Whalen, Lexington  and
      Wan, Cheng  and
      Fu, Yonggan  and
      Lin, Yingyan Celine  and
      Kundu, Souvik",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.acl-short.96/",
    doi = "10.18653/v1/2025.acl-short.96",
    pages = "1200--1209",
    ISBN = "979-8-89176-252-7"
}

News

Earlier news
  • Our paper “A Spatio-temporal Event Transformer on Versatile Tasks for Human Behavior Analysis” was accepted at the IJCAI 2024 MiGA Workshop.
  • Our RoboCup China Open team won National Second Prize, placing fifth (top 9%). I served as team leader.
  • I received Hubei Province First Prize in the China Undergraduate Mathematical Contest in Modeling.

Experience

Research Experience

GTISC Lab, Georgia Tech

Ph.D. student

Jan. 2026 – Present

Developed persistent working-memory augmentation for reasoning across diffusion steps.

EIC Lab, Georgia Tech

Undergraduate to Ph.D. student

May 2024 – Feb. 2026

Studied efficient diffusion decoding and long-context memory retention in state space models, including redundancy-aware decoding, soft token interpolation, and token filtering.

Signal Processing Lab, Wuhan University

Undergraduate student

Oct. 2023 – Apr. 2024

Developed an event-and-RGB pipeline for BRS-E²NeRF, jointly addressing rolling-shutter distortion and motion blur for sharp 3D reconstruction and novel-view synthesis.

CMVS, University of Oulu

2023 – 2024

Research on event-based human behavior analysis using spatio-temporal transformers, presented at the IJCAI 2024 MiGA Workshop.

Industry Research

Zoom Communications, Inc. · AI Incubation

Research Intern

May 2026 – Aug. 2026

Native Action Merging for Efficient Computer-Use Agents: an execution-batching framework that merges consecutive GUI actions when intermediate observations are unnecessary, reducing repeated visual processing and model deliberation.

Bosch Research & Innovation Software Development Co., Ltd.

Research Intern

Jul. 2024 – May 2025

Co-optimized diffusion-based depth generation and 3D Gaussian Splatting for LiDAR-free urban reconstruction. Worked on road-specific geometric constraints and benchmarking on Waymo.

Education

Georgia Institute of Technology

Aug. 2025 – Present

Ph.D. in Computer Science · In progress

Wuhan University

Sept. 2021 – Jun. 2025

B.Eng. in Communication Engineering (Artificial Intelligence Track)

Earlier background

I studied at the No. 1 Middle School Affiliated to Central China Normal University.

Honors & Competitions

More honors & competitions
  • China Undergraduate Mathematical Contest in ModelingHubei Province First Prize
    2023
  • Student Innovation & Entrepreneurship ProjectNational Prize · All-in-Focus Seeing Through Occlusions with Event and Frame
    2023–2024
  • Wuhan University Merit Students
    2021–2022, 2023–2024
  • Wuhan University Outstanding Student
    2022–2023
  • Wuhan University Merit and Outstanding Academic Scholarships
    2021–2023