The Evolution of Out-of-Distribution Robustness Throughout Fine-Tuning
arXiv:2106.15831
Abstract
Although machine learning models typically experience a drop in performance on out-of-distribution data, accuracies on in- versus out-of-distribution data are widely observed to follow a single linear trend when evaluated across a testbed of models. Models that are more accurate on the out-of-distribution data relative to this baseline exhibit "effective robustness" and are exceedingly rare. Identifying such models, and understanding their properties, is key to improving out-of-distribution performance. We conduct a thorough empirical investigation of effective robustness during fine-tuning and surprisingly find that models pre-trained on larger datasets exhibit effective robustness during training that vanishes at convergence. We study how properties of the data influence effective robustness, and we show that it increases with the larger size, more diversity, and higher example difficulty of the dataset. We also find that models that display effective robustness are able to correctly classify 10% of the examples that no other current testbed model gets correct. Finally, we discuss several strategies for scaling effective robustness to the high-accuracy regime to improve the out-of-distribution accuracy of state-of-the-art models.
arXiv comment27 pages, 25 figures
References in corpus (22)
- Very Deep Convolutional Networks for Large-Scale Image Recognition
- Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Overcoming catastrophic forgetting in neural networks
- Learning Transferable Visual Models From Natural Language Supervision
- Wide Residual Networks
- Wild Patterns: Ten Years After the Rise of Adversarial Machine Learning
- One weird trick for parallelizing convolutional neural networks
- Generalisation in humans and deep neural networks
- WILDS: A Benchmark of in-the-Wild Distribution Shifts
- Scalability in Perception for Autonomous Driving: Waymo Open Dataset
- Why do deep convolutional networks generalize so poorly to small image transformations?
- Do CIFAR-10 Classifiers Generalize to CIFAR-10?
- Micro-Batch Training with Batch-Channel Normalization and Weight Standardization
- An Empirical Study of Example Forgetting during Deep Neural Network Learning
- Cold Case: The Lost MNIST Digits
- Using Videos to Evaluate Image Model Robustness
- Scaling Laws for Transfer
- The Effect of Natural Distribution Shift on Question Answering Models
- The Deep Bootstrap Framework: Good Online Learners are Good Offline Generalizers
- Model Similarity Mitigates Test Set Overuse
- Why do classifier accuracies show linear trends under distribution shift?