Grounding inductive biases in natural images:invariance stems from variations in data

• URL: http://arxiv.org/abs/2106.05121v1
• Date: Wed, 9 Jun 2021 14:58:57 GMT
• Title: Grounding inductive biases in natural images:invariance stems from variations in data
• Authors: Diane Bouchacourt, Mark Ibrahim, Ari S. Morcos
• Abstract summary: We study the factors of variation in a real dataset, ImageNet. We show standard augmentation relies on a precise combination of translation and scale. We find that the main factors of variation in ImageNet mostly relate to appearance.
• Score: 20.432568247732206
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: 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. 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 inductive biases. We show the training data itself is the main source of invariance, and that data augmentation only further increases the learned invariances. Interestingly, the invariances brought from the training process 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.

Related papers

• Understanding the Role of Invariance in Transfer Learning [9.220104991339104]
  Transfer learning is a powerful technique for knowledge-sharing between different tasks. Recent work has found that the representations of models with certain invariances, such as to adversarial input perturbations, achieve higher performance on downstream tasks.
  arXiv  Detail & Related papers  (2024-07-05T07:53:52Z)
• Revisiting Data Augmentation for Rotational Invariance in Convolutional Neural Networks [0.29127054707887967]
  We investigate how best to include rotational invariance in a CNN for image classification. Our experiments show that networks trained with data augmentation alone can classify rotated images nearly as well as in the normal unrotated case.
  arXiv  Detail & Related papers  (2023-10-12T15:53:24Z)
• What Affects Learned Equivariance in Deep Image Recognition Models? [10.590129221143222]
  We find evidence for a correlation between learned translation equivariance and validation accuracy on ImageNet. Data augmentation, reduced model capacity and inductive bias in the form of convolutions induce higher learned equivariance in neural networks.
  arXiv  Detail & Related papers  (2023-04-05T17:54:25Z)
• The Lie Derivative for Measuring Learned Equivariance [84.29366874540217]
  We study the equivariance properties of hundreds of pretrained models, spanning CNNs, transformers, and Mixer architectures. We find that many violations of equivariance can be linked to spatial aliasing in ubiquitous network layers, such as pointwise non-linearities. For example, transformers can be more equivariant than convolutional neural networks after training.
  arXiv  Detail & Related papers  (2022-10-06T15:20:55Z)
• On the Strong Correlation Between Model Invariance and Generalization [54.812786542023325]
  Generalization captures a model's ability to classify unseen data. Invariance measures consistency of model predictions on transformations of the data. From a dataset-centric view, we find a certain model's accuracy and invariance linearly correlated on different test sets.
  arXiv  Detail & Related papers  (2022-07-14T17:08:25Z)
• Deep invariant networks with differentiable augmentation layers [87.22033101185201]
  Methods for learning data augmentation policies require held-out data and are based on bilevel optimization problems. We show that our approach is easier and faster to train than modern automatic data augmentation techniques.
  arXiv  Detail & Related papers  (2022-02-04T14:12:31Z)
• Quantised Transforming Auto-Encoders: Achieving Equivariance to Arbitrary Transformations in Deep Networks [23.673155102696338]
  Convolutional Neural Networks (CNNs) are equivariant to image translation. We propose an auto-encoder architecture whose embedding obeys an arbitrary set of equivariance relations simultaneously. We demonstrate results of successful re-rendering of transformed versions of input images on several datasets.
  arXiv  Detail & Related papers  (2021-11-25T02:26:38Z)
• Why Do Self-Supervised Models Transfer? Investigating the Impact of Invariance on Downstream Tasks [79.13089902898848]
  Self-supervised learning is a powerful paradigm for representation learning on unlabelled images. We show that different tasks in computer vision require features to encode different (in)variances.
  arXiv  Detail & Related papers  (2021-11-22T18:16:35Z)
• Learning Invariances in Neural Networks [51.20867785006147]
  We show how to parameterize a distribution over augmentations and optimize the training loss simultaneously with respect to the network parameters and augmentation parameters. We can recover the correct set and extent of invariances on image classification, regression, segmentation, and molecular property prediction from a large space of augmentations.
  arXiv  Detail & Related papers  (2020-10-22T17:18:48Z)
• On Robustness and Transferability of Convolutional Neural Networks [147.71743081671508]
  Modern deep convolutional networks (CNNs) are often criticized for not generalizing under distributional shifts. We study the interplay between out-of-distribution and transfer performance of modern image classification CNNs for the first time. We find that increasing both the training set and model sizes significantly improve the distributional shift robustness.
  arXiv  Detail & Related papers  (2020-07-16T18:39:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.