profile photo

Agent Claws Distributed AATW
（多重影分身の術）🥷

Xiaohuan Pei (Terry)

profile photo

I am a PhD candidate in Computer Science at the University of Sydney, supervised by Prof. Chang Xu.

My research focuses on pretraining foundation models from scratch at the billion-parameter level, with emphasis on scalable training pipelines and efficiency-oriented inference.

My current research interests include:

  • (1) Stage-1 (Alignment), Stage-2 (SFT), and Stage-3 Paradigm for Foundation Model Training;
  • (2) Efficient Vision-Language Models;
  • (3) Foundation Models for Autonomous Driving.

Email: xiaohuan.pei [AT] sydney [DOT] edu.au
Alt: terrypei123 [AT] gmail [DOT] com

χ Chi Series ψ Psi Series ω Omega Series

News

  • [2026] Three papers accepted at ICLR 2026. 🎉🎉🎉
  • [2026] Visiting graduate researcher at UCLA, hosted by Prof. Cho-Jui Hsieh.
  • [2025] One paper accepted at AAAI 2025. 🎉
Earlier News
  • [2024] One paper accepted at ICLR 2024. 🎉
  • [2024] One paper accepted at ECCV 2024. 🎉
  • [2024] Guest Lecture on Artificial Intelligence at the University of Sydney.
  • [2022] ICDM Best Student Paper Award.
  • [2022] Two papers accepted at ICDM 2022.

Previous Work
ICLR 2026
VLA-ADP
Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulation
Xiaohuan Pei*, Yuxing Chen*, Siyu Xu, Yunke Wang, Yuheng Shi, Chang Xu
[Paper] [Code] [Web]
ICLR 2026
SD-RPN
Self-Distilled RoI Predictors for Fine-Grained MLLM Perception
Yuheng Shi, Xiaohuan Pei, Minjing Dong, Chang Xu
[Paper] [Code]
ICLR 2026
LFM
Light Future-aware Masking for Vision-Language Inference
Xiaohuan Pei, Tao Huang, Yanxiang Ma, Chang Xu
[Paper] [Code]
CSP
Cross-Self KV Cache Pruning for Efficient Vision-Language Inference
Xiaohuan Pei, Tao Huang, Chang Xu
[Paper] [Code] GitHub stars
AAAI 2025
EfficientVMamba
EfficientVMamba: Atrous Selective Scan for Light Weight Visual Mamba
Xiaohuan Pei, Tao Huang, Chang Xu
Proceedings of the AAAI Conference on Artificial Intelligence (AAAI), 2025
[Paper] [Code] [Tutorial] GitHub stars
ICLR 2024
Neural Architecture Retrieval
Neural Architecture Retrieval
Xiaohuan Pei, Yanxi Li, Minjing Dong, Chang Xu
The International Conference on Learning Representations (ICLR), 2024
[Paper] [Code]
ECCV 2024
LocalMamba
LocalMamba: Visual State Space Model with Windowed Selective Scan
Tao Huang, Xiaohuan Pei, Chang Xu
The European Conference on Computer Vision (ECCV), 2024
[Paper] [Code] GitHub stars

Selected Honors and Awards

ICDM Best Student Paper Award
IEEE
Full Scholarship Award (x2)
University of Sydney
Outstanding Graduate
National Second Prize
Mathematics Competition
Provincial Prize
C++ Programming Competition
  • ICDM Best Student Paper Award, IEEE
  • Full Scholarship Award (x2), University of Sydney
  • Outstanding Graduate
  • National Second Prize in Mathematics Competition
  • Provincial Prize in C++ Programming Competition
  • NCI Adapter Scheme Grant, National Computational Infrastructure, Australia

Publications

[C] Conference; [P] Preprint

Selected Publications

  • 2026 [ICLR'26] Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulation
    Xiaohuan Pei*, Yuxing Chen*, Siyu Xu, Yunke Wang, Yuheng Shi, Chang Xu
    [C] International Conference on Learning Representations (CORE Rank A*)
    [paper] [project]
  • [ICLR'26] Self-Distilled RoI Predictors for Fine-Grained MLLM Perception
    Yuheng Shi, Xiaohuan Pei, Minjing Dong, Chang Xu
    [C] International Conference on Learning Representations (CORE Rank A*)
    [paper] [code]
  • [ICLR'26] Light Future-aware Masking for Vision-Language Inference
    Xiaohuan Pei, Tao Huang, Yanxiang Ma, Chang Xu
    [C] International Conference on Learning Representations (CORE Rank A*)
    [paper]
  • 2025 [AAAI'25] EfficientVMamba: Atrous Selective Scan for Light Weight Visual Mamba
    Xiaohuan Pei, Tao Huang, Chang Xu
    [C] AAAI Conference on Artificial Intelligence (CORE Rank A*)
    [paper] [code]
  • 2024 [ICLR'24] Neural Architecture Retrieval
    Xiaohuan Pei, Yanxi Li, Minjing Dong, Chang Xu
    [C] International Conference on Learning Representations (CORE Rank A*)
    [paper] [code]
  • [ECCV'24] LocalMamba: Visual State Space Model with Windowed Selective Scan
    Tao Huang, Xiaohuan Pei, Chang Xu
    [C] European Conference on Computer Vision (CORE Rank A*)
    [paper] [code]

Other Publications

  • 2024 [Preprint] Cross-Self KV Cache Pruning for Efficient Vision-Language Inference
    Xiaohuan Pei, Tao Huang, Chang Xu
    [P] arXiv:2412.04652
    [paper] [code]
  • 2023 [Preprint] GPT Self-supervision for a Better Data Annotator
    Xiaohuan Pei, Yanxi Li, Chang Xu
    [P] arXiv:2306.04349
    [paper]
  • [Preprint] Text-driven Neural Architecture Embeddings and Retrieval
    Xiaohuan Pei, Yanxi Li, Minjing Dong, Chang Xu
    [P] Preprint
  • 2022 [ICDM'22] Contrastive Code-Comment Pre-training
    Xiaohuan Pei, Daochang Liu, Qian Luo, Chang Xu
    [C] IEEE International Conference on Data Mining (CORE Rank A*)
  • [ICDM'22] Self-attention Gated Cognitive Diagnosis for Faster Adaptive Educational Assessments
    Xiaohuan Pei, Shuo Yang, Jiajun Huang, Chang Xu
    [C] IEEE International Conference on Data Mining (CORE Rank A*)

Professional Services

  • Reviewer: TPAMI, ICML, NeurIPS, ICLR, CVPR, ICCV, KDD, ICDM

Teaching Experience

  • Guest Lecture, Artificial Intelligence, The University of Sydney, 2024
  • Tutor, COMP5329 Deep Learning, The University of Sydney, 2023, 2025

Close Collaborators