Zehao Xiao

Zehao Xiao

Ph.D. Student

University van Amsterdam

VIS Lab , AIM Lab

About me

I am currently a last-year Ph.D. student in VIS Lab & AIM Lab at University van Amsterdam, supervised by Cees Snoek. I am also a member of ELLIS society. Before joining University van Amsterdam in 2020, I got my Master’s and Bachelor’s degree from Beihang University.

My current research focuses on: domain generalization, test-time adaptation, and meta-learning.

News

  • [Nov 2024] Our survey paper on test-time adaptation: Beyond Model Adaptation at Test Time: A Survey is released.
  • [Nov 2024] One paper accepted to WACV 2025.
  • [Apr 2024] One paper accepted to NeurIPS2024.
  • [Apr 2024] One paper accepted to CoLLAs2024.
  • [Feb 2024] One paper accepted to CVPR2024.
  • [Oct 2023] One paper accepted to NeurIPS2023.
  • [Feb 2023] One paper accepted to ICLR2023.
  • [Sep 2022] One paper accepted to NeurIPS2022.
  • [Apr 2022] One paper accepted to ICLR2022.

Publications

Beyond Model Adaptation at Test Time: A Survey
Preprint
PaperList
None
GeneralizeFormer: Layer-Adaptive Model Generation across Test-Time Distribution Shifts
WACV, 2025
Code
None
GO4Align: Group Optimization for Multi-Task Alignment
NeurIPS, 2024
Code
None
Probabilistic Test-Time Generalization by Variational Neighbor-Labeling
CoLLAs, 2024
We propose variational neighbor labels with meta-generalization that introduces pseudo labels as latent variables and meta-learn them with variational inference and neighboring information for test-time domain generalization.
Code
None
Any-Shift Prompting for Generalization over Distributions
CVPR, 2024
We propose any-shift prompting, a general probabilistic inference framework that considers both the training and test task information for prompt generation, to improve the generalization on downstream tasks across various distribution shifts by utilizing the original generalization ability.
Code
None
ProtoDiff: Learning to Learn Prototypical Networks by Task-Guided Diffusion
NeurIPS, 2023
We introduce ProtoDiff, a novel framework that leverages a task-guided diffusion model during the meta-training phase to gradually generate prototypes, thereby providing efficient class representations.
Code
None
Energy-Based Test Sample Adaptation for Domain Generalization
ICLR, 2023
We propose a discriminative energy-based model to adapt target samples to the source domain distributions for domain generalization.
Code
None
Association Graph Learning for Multi-Task Classification with Category Shifts
NeurIPS, 2022
We propose a new MTL setting which suffers from category shifts from training to test data.
Code
None
Learning to Generalize across Domains on Single Test Samples
ICLR, 2022
We leverage a meta-learning paradigm to learn our model to acquire the ability of adaptation with single samples at training time so as to further adapt itself to each single test sample at test time.
Code
None
A Bit More Bayesian: Domain-Invariant Learning with Uncertainty
ICML, 2021
To better explore domain invariant learning, we introduce weight uncertainty to the model by leveraging variational Bayesian inference.
Code
None
Relational Attention Network for Crowd Counting
ICCV, 2019
We propose a Relational Attention Network with a self-attention mechanism for capturing interdependence of pixels in crowd counting and density estimation.
None
Crowd Counting and Density Estimation by Trellis Encoder-Decoder Networks
CVPR, 2019
We propose a trellis encoder-decoder network for crowd counting, which focuses on generating high-quality density estimation maps.
For the other publications please refer to my Google Scholar.
None

Contact