Qiang Qiu

Learning Diverse Textual Contexts for Robust Personalization of Text-to-Image Diffusion Models

Generate What Matters: Steering Diffusion Models for Targeted Data Generation to Improve Classification

In-Context Compositional Learning vis Sparse Coding Transformer

SSOLE: Rethinking Orthogonal Low-rank Embedding for Self-Supervised Learning

Tuning Timestep-Distilled Diffusion Model Using Pairwise Sample Optimization

Large Convolutional Model Tuning via Filter Subspace

Divide and Conform: Unleashing Spatial Filter Atoms for Unsupervised Target Transferability

Learning to Customize Text-to-Image Diffusion In Diverse Context

Posterior sampling via Langevin dynamics based on generative priors