What augmentations are sensitive to hyper-parameters and why?
Ch Muhammad Awais, Imad Eddine Ibrahim Bekkouch
TL;DR
This paper investigates how data augmentations interact with neural network hyper-parameters in image classification. It proposes a LIME-based, regression-driven framework to quantify augmentation sensitivity via coefficients learned from ResNet50/101 models trained on FashionMNIST under four hyper-parameter settings, using nine augmentations from Albumentations. The study defines consistency, influence, and reliability to identify which augmentations are hyper-parameter sensitive (e.g., GaussianBlur, InvertImg, RandomRotate90) and which are robust (e.g., Transpose, Equalize, ShiftScaleRotate), providing practical guidance for augmentation selection under given hyper-parameters. The results offer a usable methodology for evaluating augmentation reliability and suggest directions for optimizing vector policy design and dataset scales in future work, with potential applicability to broader datasets and architectures. $ ext{reliability}_i = ext{consistency}_i imes ext{influence}_i $ and $ ext{sensitivity}_i = rac{1}{k} rac{}{} \
Abstract
We apply augmentations to our dataset to enhance the quality of our predictions and make our final models more resilient to noisy data and domain drifts. Yet the question remains, how are these augmentations going to perform with different hyper-parameters? In this study we evaluate the sensitivity of augmentations with regards to the model's hyper parameters along with their consistency and influence by performing a Local Surrogate (LIME) interpretation on the impact of hyper-parameters when different augmentations are applied to a machine learning model. We have utilized Linear regression coefficients for weighing each augmentation. Our research has proved that there are some augmentations which are highly sensitive to hyper-parameters and others which are more resilient and reliable.