Synthetic Data for Model Selection
This work addresses the challenge of limited real data for model selection in machine learning, offering a practical solution with potential domain-specific applications, though it is incremental as it builds on existing synthetic data generation techniques.
The paper tackles the problem of model selection in image classification when real data is scarce, by using synthetic data to replace the validation set, allowing for larger training datasets and introducing a calibration method that improves synthetic error estimation, resulting in significant enhancements in model selection utility.
Recent breakthroughs in synthetic data generation approaches made it possible to produce highly photorealistic images which are hardly distinguishable from real ones. Furthermore, synthetic generation pipelines have the potential to generate an unlimited number of images. The combination of high photorealism and scale turn synthetic data into a promising candidate for improving various machine learning (ML) pipelines. Thus far, a large body of research in this field has focused on using synthetic images for training, by augmenting and enlarging training data. In contrast to using synthetic data for training, in this work we explore whether synthetic data can be beneficial for model selection. Considering the task of image classification, we demonstrate that when data is scarce, synthetic data can be used to replace the held out validation set, thus allowing to train on a larger dataset. We also introduce a novel method to calibrate the synthetic error estimation to fit that of the real domain. We show that such calibration significantly improves the usefulness of synthetic data for model selection.