Alternative Architectures Have Variable Effect On Augmentation-Induced Bias | HackerNoon

Overall, the experiments suggest that the mechanism of augmentation-induced bias is largely model-agnostic, indicating that varying architectures maintain similar performance dynamics under data augmentation.

The EfficientNetV2S architecture offered insights into those dynamics, showing that while performance effects vary, the pattern of class-specific performance degradation remains similar across different model complexities.

Read at Hackernoon