E-Stitchup: Data Augmentation for Pre-Trained Embeddings

TL;DR

This paper introduces data augmentation techniques for pre-trained embeddings that enhance classification accuracy, calibration, and out-of-distribution detection, benefiting production deep learning systems.

Contribution

It proposes novel augmentation methods inspired by Mixup for embeddings, combining label softening to improve model performance and interpretability.

Findings

Significant increase in classification accuracy

Reduced training time for models

Improved confidence calibration and out-of-distribution detection

Abstract

In this work, we propose data augmentation methods for embeddings from pre-trained deep learning models that take a weighted combination of a pair of input embeddings, as inspired by Mixup, and combine such augmentation with extra label softening. These methods are shown to significantly increase classification accuracy, reduce training time, and improve confidence calibration of a downstream model that is trained with them. As a result of such improved confidence calibration, the model output can be more intuitively interpreted and used to accurately identify out-of-distribution data by applying an appropriate confidence threshold to model predictions. The identified out-of-distribution data can then be prioritized for labeling, thus focusing labeling effort on data that is more likely to boost model performance. These findings, we believe, lay a solid foundation for improving the classification performance and calibration of models that use pre-trained embeddings as input and provide several benefits that prove extremely useful in a production-level deep learning system.