Exploring Representations and Interventions in Time Series Foundation Models
This work addresses the need for more controlled and efficient time series analysis, offering incremental improvements in model optimization and intervention techniques for researchers and practitioners using time series foundation models.
The study tackled the problem of understanding internal representations and learned concepts in time series foundation models, revealing block-like redundancy that enables informed pruning for improved inference speed and demonstrating that latent space steering can manipulate features like periodicity and trends.
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.