Optimized training on custom dataset

[yoloyash]

I understand that USI is used for training SOTA CNN models for ImageNet. I was wondering if there's any possibility if we could use it to train a model (ex: edgenext_xx_small_bn_hs) on our custom dataset?