School of Computing Science

Our paper titled 'Mind the Approximation: Fisher-Weighted SVD Compression for Vision Transformers' accepted to the 37th British Machine Vision Conference (BMVC) proposes Fisher Approximation tailored to Compressing ViTs with Fisher-weighted SVD and CoRS (Constrained Rank Search) for efficient Vision Transformer compression, achieving stronger accuracy–efficiency trade-offs than existing SVD-based methods.

 


First published: 17 August 2026