Hyperspherical simplex method keeps AI retrieval systems aligned

Hyperspherical Simplex Representations for Backward-Compatible Retrieval

Computer Vision and Pattern RecognitionMachine Learning

Summary

Updating AI models can change how they understand data, making older and newer versions hard to compare. The authors propose a new way to represent AI model outputs using geometric shapes called hyperspherical simplices, which naturally line up across different model versions. This method helps systems compare old and new features without extra training or complex adjustments. Their approach is supported by a theoretical framework and works well in practice on standard tests.

What this means in practice

  • For machine learning engineers: Maintain retrieval system accuracy when updating models without costly re-indexing by using hyperspherical simplex representations.
  • For search infrastructure teams: Reduce compute and downtime when refreshing search databases by enabling direct comparison of features from independently trained models.

Authors

Simone Ricci, Niccolò Biondi, Federico Pernici, Alberto Del Bimbo

Abstract

Training modern AI models has become increasingly expensive, and updating deployed models can alter their internal feature representations, affecting applications built upon them. In retrieval systems, such updates often require complete gallery re-indexing, a computationally expensive process at scale. Existing backward-compatibility methods enable direct comparison of representations from updated and earlier models, avoiding re-indexing. However, they typically require compatibility-specific training, learned mapping functions, or architectural modifications, and often depend explicitly on the previous model. We propose hyperspherical simplex representations derived from softmax outputs and logits for backward-compatible retrieval across independently trained models. They leverage the simplex structure induced by softmax cross-entropy training with one-hot labels as a common geometric reference across model updates. Since this reference depends on the number of classes, adding new ones expands the simplex, misaligning existing class prototypes with their earlier counterparts. We therefore derive a deterministic linear transformation to restore their alignment. Our theoretical analysis of the formal backward-compatibility definition identifies two geometric conditions under which its inequalities hold in expectation, assuming von Mises--Fisher class distributions with increasing concentration across updates. Aligned regular-simplex representations satisfy these conditions, providing a theoretical basis for the proposed representations under the stated assumptions. Without using additional compatibility-specific losses, learned mappings, or architectural modifications, the proposed representations achieve competitive results on standard compatibility benchmarks, including expanded class sets, more expressive architectures, and independently trained models from public repositories.