Personalized Federated Sparse Adaptation of Time-Series Foundation Models

2026-08-05

AI Summary

This work addresses the challenge in federated building energy forecasting, where data privacy, geographical dispersion, and highly non-IID distributions hinder the balance between personalization and generalization. To this end, we propose a personalized federated sparse adaptation framework that, following pretraining of a time series foundation model (TSFM), introduces a heterogeneous temporal mixture-of-experts (MoE) adapter. This adapter employs a client-aware and backbone-aware sequence-level sparse routing mechanism to dynamically select relevant experts for each 168-hour context, enabling personalized modeling of multidimensional temporal patterns—including periodicity, long-range interactions, and local variations. Experiments across 50 buildings and three TSFM backbones demonstrate that our method significantly outperforms both global and local MoE baselines. Routing analysis further reveals expert specialization and concentration at the client level, confirming that effective adaptation must jointly account for client-specific and backbone-specific characteristics.

Abstract

Federated adaptation of time-series foundation models (TSFMs) is attractive for building energy forecasting because meter data are private, distributed, and highly non-IID. However, a single parameter-sharing strategy is unlikely to serve all pretrained TSFMs or building clients: fully shared adapters can suppress building-specific temporal behavior, while fully local adaptation discards cross-building transfer. We propose a personalized federated sparse adaptation framework with a heterogeneous temporal mixture-of-experts (MoE) adapter placed after the pretrained TSFM representation. A sequence-level router maps each 168-hour context window to a top-$k$ subset of experts specialized for periodicity, long-range interactions, local variation, trend-residual structure, and multi-resolution behavior. We compare global FL, local training, and personalized FL variants with globally shared or client-private expert banks. Across 50 buildings and three TSFM backbones, personalization consistently outperforms Global FL-MoE and Local MoE, while the best sparse-adaptation strategy varies by backbone and metric. Routing behavior further reveals client-level expert specialization, expert concentration, and near-uniform routing across backbones, showing that federated TSFM adaptation should be both client-aware and backbone-aware.

Problem

• federated learning
• time-series foundation models
• personalization
• sparsity
• non-IID data

Innovation

• personalized federated learning
• sparse adaptation
• time-series foundation models
• mixture-of-experts
• heterogeneous temporal modeling