Grounding inductive biases in natural images:invariance stems from variations in data
This work addresses the problem of achieving predictable model behavior on unseen data for researchers and practitioners in computer vision, but it is incremental as it builds on prior studies of inductive biases and invariance.
The study investigated how data, augmentations, and model architectures contribute to invariance in machine learning models, using ImageNet to analyze factors of variation. It found that training data is the primary source of invariance, with standard augmentations like translation recapturing most performance improvements, and that invariances learned align with ImageNet's class-specific appearance factors.
To perform well on unseen and potentially out-of-distribution samples, it is desirable for machine learning models to have a predictable response with respect to transformations affecting the factors of variation of the input. Here, we study the relative importance of several types of inductive biases towards such predictable behavior: the choice of data, their augmentations, and model architectures. Invariance is commonly achieved through hand-engineered data augmentation, but do standard data augmentations address transformations that explain variations in real data? While prior work has focused on synthetic data, we attempt here to characterize the factors of variation in a real dataset, ImageNet, and study the invariance of both standard residual networks and the recently proposed vision transformer with respect to changes in these factors. We show standard augmentation relies on a precise combination of translation and scale, with translation recapturing most of the performance improvement -- despite the (approximate) translation invariance built in to convolutional architectures, such as residual networks. In fact, we found that scale and translation invariance was similar across residual networks and vision transformer models despite their markedly different architectural inductive biases. We show the training data itself is the main source of invariance, and that data augmentation only further increases the learned invariances. Notably, the invariances learned during training align with the ImageNet factors of variation we found. Finally, we find that the main factors of variation in ImageNet mostly relate to appearance and are specific to each class.