Xuan Qi 齐轩

I am a Ph.D. student at the Italian Institute of Technology (IIT) and the University of Genoa, working with the AI for Good (AIGO) research group under the supervision of Prof. Vittorio Murino. My research interests include generative modeling, representation geometry, and transfer learning, centered on the understanding, transfer, and control of learned representations.

Xuan Qi

News

“AffineLens: Capturing the Continuous Piecewise Affine Functions of Neural Networks” has been accepted in Machine Learning.
“Revisiting Diffusion Fine-Tuning for Unsupervised Domain Adaptation” has been accepted at NeurIPS 2026.
“VT-DUDA: Visual Token Conditioning for Diffusion-guided Unsupervised Domain Adaptation” was accepted in Transactions on Machine Learning Research (TMLR).
“Expressivity Saturation: Reduced Affine Region Usage Under Increasing Task Complexity” was accepted in Transactions on Machine Learning Research (TMLR).
Participated in the ELIAS–ELLIS–VISMAC Winter School 2025.

Selected Publications (*) Equal contribution. (†) Corresponding author.

Publications →
NeurIPS 2026

Revisiting Diffusion Fine-Tuning for Unsupervised Domain Adaptation

Xuan Qi, Yi Wei, Daniele Berardini, Vito Paolo Pastore, Vittorio Murino
Machine Learning 2026

AffineLens: Capturing the Continuous Piecewise Affine Functions of Neural Networks

Yi Wei (*), Xuan Qi (*), Furao Shen, Jian Zhao, Vittorio Murino, Cigdem Beyan
TMLR 2026

VT-DUDA: Visual Token Conditioning for Diffusion-guided Unsupervised Domain Adaptation

Xuan Qi (†), Daniele Berardini, Dario Serez, Vito Paolo Pastore, Vittorio Murino
TMLR 2026

Expressivity Saturation: Reduced Affine Region Usage Under Increasing Task Complexity

Xuan Qi (†), Yi Wei, Fanqi Yu, Manuel Lecha

Generative Modeling

Research on controllable and conditional visual generation leveraging Generative Foundation Models (GenFMs).

Representation Geometry

Research on understanding the geometric structure of learned representations and their connections to model expressivity and behavior.

Transfer Learning

Research on parameter-efficient fine-tuning (PEFT) for transfer and generalization across domains and distributions.

Academic Service