|
Zixiao Wang (王子骁) Hi there! I am Zixiao Wang, a third-year undergraduate student at Peking University, majoring in Artificial Intelligence in Zhi Class (an honors program directed by Prof. Baoquan Chen) at the School of Intelligence Science and Technology. Also, I am currently exchanging at School of Computer Science at Carnegie Mellon University. My research focuses on Large Language Models, Optimization and Multimodal Learning. I am particularly interested in applying efficient and effective optimization algorithms for pretraining and RL post training of large language models, as well as analyzing the theoretical properties behind the training process. I'm also interested in exploring the potential of multimodal learning, especially in the context of video understanding and generation. I am honored to be advised by Prof. Huishuai Zhang and Prof. Dongyan Zhao in Wangxuan Institute of Computer Technology. I am also very fortunate to be currently working closely with Prof. Jiawen Zhang from University of Wisconsin-Madison. Previously, I also had the privilege to work with Prof. Yang Liu. Email / GitHub / Google Scholar / Openreview / CV |
|
News2026-5 OLion: Approaching the Hadamard Ideal by Intersecting Spectral and L inf Implicit Biases has been accepted by ICML 2026! |
ResearchLarge Language Models; Optimization; Pretraining; Efficient Reinforcement Learning; Multimodal |
|
|
OLion: Approaching the Hadamard Ideal by Intersecting Spectral and L inf Implicit Biases
Zixiao Wang*, Yifei Shen*, Huishuai Zhang* When Muon meets Lion ICML 2026, 2026-5 Accepted Arxiv Preprint / Virtual Poster We introduce OLion, a memory-efficient optimizer that combines spectral control from Muon-style orthogonalized updates with ℓ∞-style coordinate control from Lion-style sign updates. |
|
|
Training-free Video Temporal Grounding Enhanced by Contextual Region Cue based Test-time Adaptation
Jiayi Gao, Minghang Zheng, Zixiao Wang, Yuxin Peng†, Yang Liu† https://openreview.net/pdf?id=eIhP3pDpWx We present a novel training-free video-level adapter to enhance video temporal grounding by importing in-domain context and region-wise visual cues. |
Experience |
|
|
University of Wisconsin-Madison
2026.5 - present Summertime Research Intern |
|
|
School of Computer Science, Carnegie Mellon University
2026.1 - 2026.5 Visiting Student |
|
|
Wangxuan Institute of Computer Technology Web Information Processing Lab
2025.5 - present Research Intern |
|
|
Zhi Class, Peking University
2023.9 - present Undergraduate Student |
Honors & Awards
2023-24 Leo KoGuan Scholarship (20000 CNY, top 100 students in PKU per year) |
Miscellaneous |
||
|
|
"Per Aspera Ad Astra."
|
|
Website template from Jon Barron. |