Conference

One paper accepted to ICLR 2026.

We propose SiNGER, a nullspace-guided LoRA framework that suppresses artifacts in Vision Transformer distillation while preserving informative representations, achieving state-of-the-art student performance.

The paper introduces A2XP, a novel approach for domain generalization, which optimizes prompts for each source domain and combines them for robust performance across domains, while also preserving privacy and integrity of the network architecture. Experimental results demonstrate its state-of-the-art performance in computer vision tasks and its potential for broader applications.

One paper accepted to CVPR 2024.