Shiran Yuan

I'm a teenager who loves doing quirky stuff! Currently doing a PhD in Applied Math at Tsinghua University. I received my Master's degree in Electronic Engineering and Computer Science (EECS) from UC Berkeley in 2025. Before that I received dual B.S. degrees with Full Scholarship and Senior Award from Duke University (in Applied Math and Computational Sciences) and Duke Kunshan University (in Computer Science and Technology).

My research focuses on the intersection between Computer Vision, Graphics, Machine Learning, and Applied Math. I've received the Best Paper Award at the International Conference on 3D Vision (3DV 2024), and was elected a Full Member of Sigma Xi, the Scientific Research Honor Society.

Email / Scholar / Github

Research

	Next-Scale Autoregressive Models are Zero-Shot Single-Image Object View Synthesizers Shiran Yuan, Hao Zhao arXiv, 2025 code / arXiv The first object-centric novel view synthesis model to be based on autoregressive modeling. It is several times faster than current models, exhibits significantly higher synthesis quality, scales with both model size and dataset size, and can be trained from scratch without any 2D generative base models.
	SA-GS: Scale-Adaptive Gaussian Splatting for Training-Free Anti-Aliasing Xiaowei Song, Jv Zheng, Shiran Yuan, Huan-ang Gao, Jingwei Zhao, Xiang He, Weihao Gu, Hao Zhao CVM, 2025 (Journal Track) code / arXiv Antialiasing gaussian splats using supersampling and numerical integration to reduce artifacts from changing sampling frequency.
	Hybrid Spatial Representations for Species Distribution Modeling Shiran Yuan, Hao Zhao arXiv, 2024 code / arXiv A method combining implicit and explicit representations to achieve accurate species distribution models based on noisy presence-only data.
	A Generalizable Framework for Low-Rank Tensor Completion with Numerical Priors Shiran Yuan, Kaizhu Huang Pattern Recognition, 2024 code / arXiv A framework for low-rank tensor completion that leverages both numerical priors and inter-element priors, with applications to image completion.
	SlimmeRF: Slimmable Radiance Fields Shiran Yuan, Hao Zhao 3DV, 2024 (Best Paper Award) code / arXiv A single model can be slimmed at inference time to instantly achieve desired trade-offs between reconstruction quality and memory-efficiency.