Towards Backward-Compatible Representation Learning

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.

arXiv commentAccepted to CVPR 2020 as oral, revised version

References in corpus (11)

• Distilling the Knowledge in a Neural Network
• In Defense of the Triplet Loss for Person Re-Identification
• Learning Transferable Features with Deep Adaptation Networks
• Deep Domain Confusion: Maximizing for Domain Invariance
• Deep Visual Domain Adaptation: A Survey
• Deep Learning Face Representation by Joint Identification-Verification
• Domain Separation Networks
• Slimmable Neural Networks
• AutoSlim: Towards One-Shot Architecture Search for Channel Numbers
• Convergent Learning: Do different neural networks learn the same representations?
• A Case for Backward Compatibility for Human-AI Teams