From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication
This work addresses the problem of efficiently merging and reusing neural modules for researchers and practitioners, though it appears incremental as it builds on existing invariance concepts.
The paper tackles the challenge of connecting representations from different neural networks by introducing a method to incorporate invariances into latent spaces, enabling zero-shot stitching across models. It demonstrates consistent improvements in latent similarity and downstream performance across vision, text, and graph modalities using twelve pretrained models and nine benchmarks.
It has been observed that representations learned by distinct neural networks conceal structural similarities when the models are trained under similar inductive biases. From a geometric perspective, identifying the classes of transformations and the related invariances that connect these representations is fundamental to unlocking applications, such as merging, stitching, and reusing different neural modules. However, estimating task-specific transformations a priori can be challenging and expensive due to several factors (e.g., weights initialization, training hyperparameters, or data modality). To this end, we introduce a versatile method to directly incorporate a set of invariances into the representations, constructing a product space of invariant components on top of the latent representations without requiring prior knowledge about the optimal invariance to infuse. We validate our solution on classification and reconstruction tasks, observing consistent latent similarity and downstream performance improvements in a zero-shot stitching setting. The experimental analysis comprises three modalities (vision, text, and graphs), twelve pretrained foundational models, nine benchmarks, and several architectures trained from scratch.