Towards Context-Agnostic Learning Using Synthetic Data
This addresses the problem of data efficiency and robustness in visual classification for AI applications, though it appears incremental as it builds on existing synthetic data methods.
The paper tackles the problem of learning from a single synthetic example per class by proposing a context-agnostic setting where input domains are defined on product sets, deriving a risk bound with weak label dependence. It achieves robust performance on real-world image classification benchmarks, whereas direct training on real data yields brittle classifiers.
We propose a novel setting for learning, where the input domain is the image of a map defined on the product of two sets, one of which completely determines the labels. We derive a new risk bound for this setting that decomposes into a bias and an error term, and exhibits a surprisingly weak dependence on the true labels. Inspired by these results, we present an algorithm aimed at minimizing the bias term by exploiting the ability to sample from each set independently. We apply our setting to visual classification tasks, where our approach enables us to train classifiers on datasets that consist entirely of a single synthetic example of each class. On several standard benchmarks for real-world image classification, we achieve robust performance in the context-agnostic setting, with good generalization to real world domains, whereas training directly on real world data without our techniques yields classifiers that are brittle to perturbations of the background.