Matryoshka Representation Learning

arxiv:2205.13147

Original source · Paper JSON

Retrieval, embeddings, benchmarks · 2022-05-26

Cites (1) · Cited by (1)

Complete cleaned Markdown · Source Markdown including bibliography

Read in parts

1. Part 1
2. Part 2
3. Part 3
4. Part 4
5. Part 5

Contents

• Matryoshka Representation Learning
• Abstract
• 1 Introduction
• 2 Related Work
• Representation Learning.
• Efficient Classification and Retrieval.
• 3 Matryoshka ​ Representation ​ Learning {\rm Matryoshka~Representation~Learning}
• Adaptation to Learning Frameworks.
• 4 Applications
• 4.1 Representation Learning
• 4.2 Classification
• 4.2.1 Adaptive Classification
• 4.3 Retrieval
• 4.3.1 Adaptive Retrieval
• 5 Further Analysis and Ablations
• Robustness.
• Few-shot and Long-tail Learning.
• Disagreement across Dimensions.
• Superclass Accuracy.
• 5.1 Ablations
• 6 Discussion and Conclusions
• Acknowledgments
• Appendix A Code for Matryoshka ​ Representation ​ Learning {\rm Matryoshka~Representation~Learning} ( MRL {\rm MRL} )
• Appendix B Datasets
• ImageNet Robustness Datasets
• Appendix C Matryoshka ​ Representation ​ Learning {\rm Matryoshka~Representation~Learning} Model Training
• Appendix D Classification Results
• D.1 Adaptive Classification ( MRL {\rm MRL} –AC)
• D.2 JFT, ALIGN and BERT
• Appendix E Image Retrieval
• Appendix F Adaptive Retrieval
• Funnel Retrieval.
• Appendix G Few-shot and Sample Efficiency
• ImageNetV2.
• FLUID.
• Appendix H Robustness Experiments
• Appendix I In Practice Costs
• MRL {\rm MRL} models.
• Retrieval
• Appendix J Analysis of Model Disagreement
• Class Trends
• Discussion of Oracle Accuracy
• Grad-CAM Examples
• Superclass Performance
• Appendix K Ablation Studies
• K.1 MRL {\rm MRL} Training Paradigm
• Matryoshka ​ Representations {\rm Matryoshka~Representations} via Finetuning.
• Relative Importance.
• Matryoshka ​ Representations {\rm Matryoshka~Representations} at Arbitrary Granularities.
• Lower Dimensionality.
• K.2 Retrieval
• Adaptive Retrieval.
• Instructions for reporting errors