Mask the Redundancy: Evolving Masking Representation Learning for Multivariate Time-Series Clustering

TL;DR

The paper targets temporal redundancy in multivariate time-series clustering by introducing EMTC, a framework that learns to mask redundant timestamps through an Importance-aware Variate-wise Masking (IVM) module and learns robust representations via Multi-Endogenous Views (MEV). The model integrates Consistency and Reconstruction Learning (CRL) with Clustering-guided MEV Contrastive Learning (CMC) to align representation learning with clustering objectives, achieving state-of-the-art results across 15 datasets with a reported average F1 improvement of 4.85% over strong baselines. Extensive ablations, significance testing, and efficiency analyses corroborate the effectiveness and practicality of the evolving-masking approach for unsupervised MTS analysis. The work highlights the potential of dynamic, task-aware masking combined with multi-view learning to address redundancy while preserving discriminative temporal patterns.

Abstract

Multivariate Time-Series (MTS) clustering discovers intrinsic grouping patterns of temporal data samples. Although time-series provide rich discriminative information, they also contain substantial redundancy, such as steady-state machine operation records and zero-output periods of solar power generation. Such redundancy diminishes the attention given to discriminative timestamps in representation learning, thus leading to performance bottlenecks in MTS clustering. Masking has been widely adopted to enhance the MTS representation, where temporal reconstruction tasks are designed to capture critical information from MTS. However, most existing masking strategies appear to be standalone preprocessing steps, isolated from the learning process, which hinders dynamic adaptation to the importance of clustering-critical timestamps. Accordingly, this paper proposes the Evolving-masked MTS Clustering (EMTC) method, whose model architecture comprises Importance-aware Variate-wise Masking (IVM) and Multi-Endogenous Views (MEV) generation modules. IVM adaptively guides the model in learning more discriminative representations for clustering, while the reconstruction and cluster-guided contrastive learning pathways enhance and connect the representation learning to clustering tasks. Extensive experiments on 15 benchmark datasets demonstrate the superiority of EMTC over eight SOTA methods, where the EMTC achieves an average improvement of 4.85% in F1-Score over the strongest baselines.

Figures (7)

• Figure 1: (a) Static Masking vs. (b) Evolving Masking (ours). Static masking is a fixed MTS preprocessing strategy to enhance the downstream representation learning. In contrast, our evolving masking dynamically adapts to the representation learning and clustering objective through attention-based timestamp weighting, simultaneously surpassing the MTS redundancy and enhancing the cluster discrimination of representations.
• Figure 2: Overview of the proposed EMTC framework.
• Figure 3: BD test (a), (b), (c) and (d) based on the average ranks in ACC, F1, NMI and ARI presented in Table \ref{['tab:comparison_all_metrics']}.
• Figure 4: Clustering performance of EMTC+Evolving Masking vs. EMTC+Static Masking.
• Figure 5: Loss function values during EMTC training.