Fetching the paper…

IA-RED$^2$: Interpretability-Aware Redundancy Reduction for Vision Transformers · Around