Latent Space Translation via Semantic Alignment

TL;DR

This paper introduces a method to translate latent representations between different neural networks using simple algebraic transformations, enabling effective model stitching and zero-shot multimodal tasks without additional training.

Contribution

It presents a novel, algebraic approach for translating latent spaces across neural models, facilitating encoder-decoder stitching and multimodal applications.

Findings

Transformations can be estimated with closed-form algebraic solutions.

The method enables zero-shot stitching of different modalities.

Effective across various architectures and tasks.

Abstract

While different neural models often exhibit latent spaces that are alike when exposed to semantically related data, this intrinsic similarity is not always immediately discernible. Towards a better understanding of this phenomenon, our work shows how representations learned from these neural modules can be translated between different pre-trained networks via simpler transformations than previously thought. An advantage of this approach is the ability to estimate these transformations using standard, well-understood algebraic procedures that have closed-form solutions. Our method directly estimates a transformation between two given latent spaces, thereby enabling effective stitching of encoders and decoders without additional training. We extensively validate the adaptability of this translation procedure in different experimental settings: across various trainings, domains, architectures (e.g., ResNet, CNN, ViT), and in multiple downstream tasks (classification, reconstruction). Notably, we show how it is possible to zero-shot stitch text encoders and vision decoders, or vice-versa, yielding surprisingly good classification performance in this multimodal setting.