a premier international conference in multimodal interaction
GRoFA: Noise-Gated Adapters for Jointly Fair and Robust Face Embeddings
Amu Suemoto, Yutaka Arakawa, Tsunenori Mine
GRoFA uses noise-gated adapters to improve both demographic fairness and noise robustness in face embeddings. It adapts frozen BLIP, CLIP, and DINOv2 encoders with a small number of trainable parameters. The official repository includes training and evaluation code, experimental logs, and a supplementary appendix.
Presented: · Naples, Italy
15:00–16:30 · Poster Session #3 · Panel 17 (local time)
