On the ViT -L/16 architecture, our experiments show that a single adapter, 1.3% of the full model, is able to reach full fine-tuning accuracy on average across 11 challenging downstream classification tasks. Compared with other forms of parameter-efficient adaptation, the isolated nature of the InCA adaptation is computationally desirable for large-scale models. For instance, we adapt ViT -G/14 (1.8B+ parameters) quickly with 20+ adapters in parallel on a single V100 GPU (76% GPU memory reduction) and exhaustively identify its

Neural Information Processing Systems http://nips.cc/

Existing long-tailed recognition methods, aiming totrain class-balanced models from long-tailed data, generally assume the models would be evaluated on the uniform test class distribution.

Neural Information Processing Systems http://nips.cc/

Ifwecanset1 /w max η 1 / (20 w max) thentheiterationcomplexityof Theorem 1 reduces to O ( κeff( w ) logα 1). Themagnitudeofw max is, however, anunknownquantity.

Neural Information Processing Systems http://nips.cc/

Y et, under the strongly-aligned regime, KM suffers the saturation effect, while TKM can be continuously improved as the alignment becomes stronger. This further implies that TKM has a strong ability to capture the strong alignment and provide a theoretically guaranteed solution to eliminate the phenomena of saturation effect.

Optimal transport (OT) [30, 25] has become increasingly popular in the machine learning and optimizationcommunity.