Common Variable Learning and Invariant Representation Learning using Siamese Neural Networks

TL;DR

This paper explores how Siamese neural networks can be used to learn invariant representations of data from multiple sensors, effectively capturing common variability without relying on models or labels.

Contribution

It provides a theoretical interpretation of Siamese networks as learning equivalence classes and demonstrates their effectiveness in capturing shared variability in data.

Findings

Siamese networks can learn invariant representations of common variability.

The approach is effective in exploratory, data-driven scenarios without labels.

Empirical results confirm the ability to learn shared features across sensors.

Abstract

We consider the statistical problem of learning common source of variability in data which are synchronously captured by multiple sensors, and demonstrate that Siamese neural networks can be naturally applied to this problem. This approach is useful in particular in exploratory, data-driven applications, where neither a model nor label information is available. In recent years, many researchers have successfully applied Siamese neural networks to obtain an embedding of data which corresponds to a "semantic similarity". We present an interpretation of this "semantic similarity" as learning of equivalence classes. We discuss properties of the embedding obtained by Siamese networks and provide empirical results that demonstrate the ability of Siamese networks to learn common variability.