In-Context Time Series Classification with Random Convolutional Features

2026-07-21

AI Summary

This work addresses the challenge of effectively integrating local shape, frequency characteristics, temporal shifts, and cross-channel interactions in time series classification by proposing a novel training-free approach. The method uniquely feeds random convolutional features generated by MultiRocket and Hydra into a pretrained tabular foundation model, leveraging in-context inference for classification without requiring task-specific fine-tuning. In contrast to conventional approaches that rely on linear classifiers, this strategy substantially enhances representational capacity. Empirical results demonstrate that the proposed method achieves superior average performance over the current state-of-the-art HIVE-COTE 2.0 on univariate datasets and remains highly competitive on multivariate benchmarks.

Abstract

Time series classification is central to domains like medical signal analysis, industrial monitoring, and sensor-based activity recognition, where class information manifests as localized shapes, specific frequencies, temporal shifts, or complex cross-channel interactions. Random convolutional transforms efficiently map these sequences to fixed-dimensional tabular features but are traditionally paired with simple linear classifiers. We investigate whether a pretrained tabular foundation model can more effectively harness these rich representations. We propose MASHT, a pipeline that marries MultiRocket and Hydra features with the power of in-context tabular foundation models. By leveraging a pretrained tabular foundation model, our approach completely bypasses task-specific model training, requiring only feature extraction and direct inference. Extensive experiments demonstrate that MASHT matches state-of-the-art time series classification baselines on univariate tasks, achieving a lower average rank than HIVE-COTE 2.0. On multivariate datasets, MASHT remains highly competitive with the strongest reference methods.

Problem

• time series classification
• in-context learning
• tabular foundation model
• random convolutional features

Innovation

• in-context learning
• random convolutional features
• tabular foundation model
• time series classification
• zero-shot inference