TimEE: End-to-end Time Series Classification via In-Context Learning

2026-07-08

AI Summary

This work proposes TimEE, an end-to-end foundation model for time series classification based on in-context learning. Unlike conventional two-stage approaches that decouple feature encoding from classification, require dataset-specific training, and cannot leverage label information during inference, TimEE directly outputs class probabilities in a single forward pass without any downstream fine-tuning. Remarkably, TimEE is pretrained exclusively on synthetic data and achieves state-of-the-art performance on the UCR benchmark, ranking first in ROC AUC and third in accuracy—surpassing both existing supervised models and foundation models. Its key innovation lies in integrating the Prior-Data Fitted Network (PFN) framework with structured distributional shifts, enabling meta-training on synthetic tasks and establishing a novel paradigm of end-to-end in-context learning grounded in synthetic priors.

Abstract

Time series classification (TSC) is dominated by a two-stage paradigm: train a feature encoder -- either from scratch on the target dataset or via pretraining on large corpora -- and then fit a task-specific classifier on top. While effective, this decoupling optimizes representation learning independently of the classification objective, requires per-dataset training, and prevents the model from exploiting label information during inference. We introduce TimEE, a 4.5M-parameter foundation model for end-to-end TSC via in-context learning. Given a labeled support set and a query time series, TimEE directly outputs a predicted class distribution in a single forward pass with no per-dataset training required. Following the prior-data fitted network (PFN) framework, TimEE is meta-trained exclusively on synthetic TSC tasks, where each task contains time series with distinct class identities arising from structured distributional shifts in the generative process. Despite seeing no real time series during pre-training, TimEE ranks first in ROC AUC (and third on accuracy) on the UCR benchmark among all compared methods, which include both foundation models and supervised deep learning baselines. To our knowledge, TimEE is the first purely synthetic-pretrained model to reach state-of-the-art performance on the UCR benchmark. These results establish end-to-end ICL with synthetic priors as a compelling, largely unexplored direction for TSC, with scaling, prior design, and richer generation mechanisms as natural avenues for improvement. Code is publicly available at http://github.com/automl/timee.

Problem

• time series classification
• in-context learning
• foundation model
• synthetic pretraining
• end-to-end learning

Innovation

• in-context learning
• time series classification
• foundation model
• synthetic pretraining
• end-to-end learning