Latent Structural Similarity Networks for Unsupervised Discovery in Multivariate Time Series

TL;DR

This paper introduces a task-agnostic method for discovering relationships in multivariate time series by constructing a latent similarity network, enabling analysis without assuming linearity or stationarity.

Contribution

It presents a novel unsupervised framework that creates an interpretable relational graph from multivariate time series without relying on specific downstream tasks.

Findings

Latent similarity networks reveal coherent structures in cryptocurrency data.

The approach uncovers classical econometric relationships as external validation.

Framework is flexible and does not assume linearity or stationarity.

Abstract

This paper proposes a task-agnostic discovery layer for multivariate time series that constructs a relational hypothesis graph over entities without assuming linearity, stationarity, or a downstream objective. The method learns window-level sequence representations using an unsupervised sequence-to-sequence autoencoder, aggregates these representations into entity-level embeddings, and induces a sparse similarity network by thresholding a latent-space similarity measure. This network is intended as an analyzable abstraction that compresses the pairwise search space and exposes candidate relationships for further investigation, rather than as a model optimized for prediction, trading, or any decision rule. The framework is demonstrated on a challenging real-world dataset of hourly cryptocurrency returns, illustrating how latent similarity induces coherent network structure; a classical econometric relation is also reported as an external diagnostic lens to contextualize discovered edges.