# Darwinian Model Upgrades: Model Evolving with Selective Compatibility

arXiv: https://arxiv.org/abs/2210.06954 · HTML: https://mlsanity.com/paper/2210.06954/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 21 of its 24 references that are in the map.
- First posted on arXiv 13 Oct 2022.

## Builds on (most cited first, 21 of 21)

- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://mlsanity.com/paper/2103.14030/) (2021)
- [Fine-tuning CNN Image Retrieval with No Human Annotation](https://mlsanity.com/paper/1711.02512/) (2017)
- [Google Landmarks Dataset v2 -- A Large-Scale Benchmark for Instance-Level Recognition and Retrieval](https://mlsanity.com/paper/2004.01804/) (2020)
- [Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking](https://mlsanity.com/paper/1803.11285/) (2018)
- [Towards Backward-Compatible Representation Learning](https://mlsanity.com/paper/2003.11942/) (2020)
- [Positive-Congruent Training: Towards Regression-Free Model Updates](https://mlsanity.com/paper/2011.09161/) (2020)
- [Learning Compatible Embeddings](https://mlsanity.com/paper/2108.01958/) (2021)
- [Unified Representation Learning for Cross Model Compatibility](https://mlsanity.com/paper/2008.04821/) (2020)
- [Asymmetric metric learning for knowledge transfer](https://mlsanity.com/paper/2006.16331/) (2020)
- [Compatibility-aware Heterogeneous Visual Search](https://mlsanity.com/paper/2105.06047/) (2021)
- [Towards Universal Backward-Compatible Representation Learning](https://mlsanity.com/paper/2203.01583/) (2022)
- [Forward Compatible Training for Large-Scale Embedding Retrieval Systems](https://mlsanity.com/paper/2112.02805/) (2021)
- [Large-scale Landmark Retrieval/Recognition under a Noisy and Diverse Dataset](https://mlsanity.com/paper/1906.04087/) (2019)
- [Learning Backward Compatible Embeddings](https://mlsanity.com/paper/2206.03040/) (2022)
- [Supporting large-scale image recognition with out-of-domain samples](https://mlsanity.com/paper/2010.01650/) (2020)
- [Privacy-Preserving Model Upgrades with Bidirectional Compatible Training in Image Retrieval](https://mlsanity.com/paper/2204.13919/) (2022)
- [Hot-Refresh Model Upgrades with Regression-Alleviating Compatible Training in Image Retrieval](https://mlsanity.com/paper/2201.09724/) (2022)
- [1st Place Solution to Google Landmark Retrieval 2020](https://mlsanity.com/paper/2009.05132/) (2020)
- [3rd Place Solution to "Google Landmark Retrieval 2020"](https://mlsanity.com/paper/2008.10480/) (2020)
- [2nd Place Solution to Google Landmark Retrieval 2021](https://mlsanity.com/paper/2110.04294/) (2021)

## Built on by (most cited first, 5 of 5)

- [TaCA: Upgrading Your Visual Foundation Model with Task-agnostic Compatible Adapter](https://mlsanity.com/paper/2306.12642/) (2023)
- [MixBCT: Towards Self-Adapting Backward-Compatible Training](https://mlsanity.com/paper/2308.06948/) (2023)
- [Structure Similarity Preservation Learning for Asymmetric Image Retrieval](https://mlsanity.com/paper/2403.00648/) (2024)
- [Prototype Perturbation for Relaxing Alignment Constraints in Backward-Compatible Learning](https://mlsanity.com/paper/2503.14824/) (2025)
- [Spherical Interpolation for Backward-Compatible Multimodal Representations](https://mlsanity.com/paper/2609.39836/) (2026)

## Abstract (arXiv)

The traditional model upgrading paradigm for retrieval requires recomputing all gallery embeddings before deploying the new model (dubbed as "backfilling"), which is quite expensive and time-consuming considering billions of instances in industrial applications. BCT presents the first step towards backward-compatible model upgrades to get rid of backfilling. It is workable but leaves the new model in a dilemma between new feature discriminativeness and new-to-old compatibility due to the undifferentiated compatibility constraints. In this work, we propose Darwinian Model Upgrades (DMU), which disentangle the inheritance and variation in the model evolving with selective backward compatibility and forward adaptation, respectively. The old-to-new heritable knowledge is measured by old feature discriminativeness, and the gallery features, especially those of poor quality, are evolved in a lightweight manner to become more adaptive in the new latent space. We demonstrate the superiority of DMU through comprehensive experiments on large-scale landmark retrieval and face recognition benchmarks. DMU effectively alleviates the new-to-new degradation and improves new-to-old compatibility, rendering a more proper model upgrading paradigm in large-scale retrieval systems.