Discrete Prototypical Memories for Federated Time Series Foundation Models

TL;DR

This paper introduces eDPM, a federated learning framework for time series foundation models that uses discrete prototypical memories to improve semantic alignment and personalization.

Contribution

It proposes a novel federated framework leveraging discrete prototypical memories to better model heterogeneous time-series data across domains.

Findings

eDPM demonstrates improved performance over existing methods.

The framework effectively balances shared and personalized knowledge.

Experiments validate the efficiency and effectiveness of the approach.

Abstract

Leveraging Large Language Models (LLMs) as federated learning (FL)-based time series foundation models offers a promising way to transfer the generalization capabilities of LLMs to time series data while preserving access to private data. However, the semantic misalignment between time-series data and the text-centric latent space of existing LLMs often leads to degraded performance. Meanwhile, the parameter-sharing mechanism in existing FL methods model heterogeneous cross-domain time-series data into a unified continuous latent space, which contradicts the fact that time-series semantics frequently manifest as discrete and recurring regimes. To address these limitations, we propose \textsc{FeDPM}, a federated framework for time-series foundation models based on discrete prototypical memories. Specifically, we learn local prototypical memory priors for intra-domain time-series data. We then align cross-domain memories to promote a unified discrete latent space and introduce a domain-specific memory update mechanism to balance shared and personalized prototypical knowledge. Extensive experiments demonstrate the efficiency and effectiveness of \textsc{FeDPM}. The code is publicly available at https://anonymous.4open.science/r/FedUnit-64D1.