MTM: A Multi-Scale Token Mixing Transformer for Irregular Multivariate Time Series Classification

2025-09-22

AI Summary

To address the classification of irregular multivariate time series (IMTS), where asynchronous channel observations impede effective modeling, this paper proposes the Multi-scale Token Mixing Transformer (MTM). MTM mitigates asynchrony via multi-scale downsampling, employs masked concatenation pooling to preserve critical temporal structures, and introduces cross-channel token mixing alongside channel-wise attention to explicitly enhance dynamic inter-channel interactions. Evaluated on multiple real-world IMTS benchmarks, MTM achieves significant performance gains over state-of-the-art methods, with up to a 3.8% improvement in AUPRC—setting new SOTA results. Its core contribution lies in the first integration of multi-scale token mixing with masked pooling for IMTS modeling, effectively resolving the challenge of collaborative learning across asynchronously observed channels.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Computer Vision: Multi-modal Vision
• Multiagent Systems: Multiagent Learning

Application Category

• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

Irregular multivariate time series (IMTS) is characterized by the lack of synchronized observations across its different channels. In this paper, we point out that this channel-wise asynchrony can lead to poor channel-wise modeling of existing deep learning methods. To overcome this limitation, we propose MTM, a multi-scale token mixing transformer for the classification of IMTS. We find that the channel-wise asynchrony can be alleviated by down-sampling the time series to coarser timescales, and propose to incorporate a masked concat pooling in MTM that gradually down-samples IMTS to enhance the channel-wise attention modules. Meanwhile, we propose a novel channel-wise token mixing mechanism which proactively chooses important tokens from one channel and mixes them with other channels, to further boost the channel-wise learning of our model. Through extensive experiments on real-world datasets and comparison with state-of-the-art methods, we demonstrate that MTM consistently achieves the best performance on all the benchmarks, with improvements of up to 3.8% in AUPRC for classification.

Problem

• Classifying irregular multivariate time series with unsynchronized channel observations
• Addressing poor channel-wise modeling caused by asynchrony in existing methods
• Enhancing channel-wise attention and token mixing for improved classification performance

Innovation

• Multi-scale token mixing transformer for irregular time series
• Masked concat pooling gradually down-samples time series
• Channel-wise token mixing mechanism boosts inter-channel learning