Towards Backward-Compatible Representation Learning

TL;DR

This paper introduces a backward-compatible training framework that allows new visual feature embeddings to be directly compared with older ones, facilitating efficient updates in large-scale visual search systems.

Contribution

The paper presents a novel training method enabling different neural network embeddings to be compatible, reducing the need for re-computing features during model updates.

Findings

Models trained with BCT maintain accuracy while achieving backward compatibility.

BCT enables backfill-free updates in large-scale visual search systems.

Experimental results on face recognition demonstrate effectiveness.

Abstract

We propose a way to learn visual features that are compatible with previously computed ones even when they have different dimensions and are learned via different neural network architectures and loss functions. Compatible means that, if such features are used to compare images, then "new" features can be compared directly to "old" features, so they can be used interchangeably. This enables visual search systems to bypass computing new features for all previously seen images when updating the embedding models, a process known as backfilling. Backward compatibility is critical to quickly deploy new embedding models that leverage ever-growing large-scale training datasets and improvements in deep learning architectures and training methods. We propose a framework to train embedding models, called backward-compatible training (BCT), as a first step towards backward compatible representation learning. In experiments on learning embeddings for face recognition, models trained with BCT successfully achieve backward compatibility without sacrificing accuracy, thus enabling backfill-free model updates of visual embeddings.