Exploring Representations and Interventions in Time Series Foundation Models

TL;DR

This paper analyzes the internal representations of Time Series Foundation Models, revealing redundancy and how interventions can manipulate learned concepts like periodicity and trends for improved control and efficiency.

Contribution

It provides the first detailed analysis of representation structure in TSFMs and demonstrates how latent space steering can modify learned concepts.

Findings

Redundancy in model layer representations enables pruning.

Steering interventions can add or modify features like periodicity.

Representation analysis aids in optimizing model performance.

Abstract

Time series foundation models (TSFMs) promise to be powerful tools for a wide range of applications. However, their internal representations and learned concepts are still not well understood. In this study, we investigate the structure and redundancy of representations across various TSFMs, examining the self-similarity of model layers within and across different model sizes. This analysis reveals block-like redundancy in the representations, which can be utilized for informed pruning to improve inference speed and efficiency. Additionally, we explore the concepts learned by these models - such as periodicity and trends - and how these can be manipulated through latent space steering to influence model behavior. Our experiments show that steering interventions can introduce new features, e.g., adding periodicity or trends to signals that initially lacked them. These findings underscore the value of representational analysis for optimizing models and demonstrate how conceptual steering offers new possibilities for more controlled and efficient time series analysis with TSFMs.