AnchorMoE: Interpretable Time Series Classification via Anchor-Routed MoE

2026-06-02

AI Summary

This work addresses the challenge of achieving transparency in multivariate time series classification, where discriminative signals are often sparse, heterogeneous, and easily obscured by noise. To this end, the authors propose an interpretable classification framework grounded in a Mixture-of-Experts architecture. By leveraging multi-view local representation learning and an anchor-based routing mechanism, the model decomposes predictions into an additive composition of input segments, thereby enabling ante-hoc interpretability. Geometric orthogonality constraints are introduced to minimize redundancy among expert representations and encourage specialization toward distinct discriminative patterns. Additionally, an uncertainty-aware reliability gating mechanism dynamically suppresses noisy inputs. Experimental results demonstrate that the proposed method achieves state-of-the-art classification performance while providing high-fidelity, traceable decision explanations across multiple real-world and synthetic datasets.

Abstract

Multivariate time series classification (MTSC) is pivotal in high-stakes domains, such as clinical diagnosis and industrial fault detection, where safe deployment necessitates transparent decision-making. However, isolating the temporal segments that drive model predictions is challenging because discriminative signals in real-world time series are typically sparse, heterogeneous, and heavily obscured by background noise. This paper, therefore, proposes AnchorMoE, an interpretable-by-construction classification framework. Built upon a Mixture-of-Experts (MoE) architecture, AnchorMoE encodes multi-view representations of local patches and routes them to specialized experts, ensuring that the final prediction is formulated as an exact additive decomposition over the input segments, facilitating ante-hoc transparency rather than relying on post-hoc estimations. To maintain the reliability of this decomposition under sparse signal distributions, we introduce a geometric orthogonality constraint that penalizes representational redundancy, compelling distinct experts to specialize in heterogeneous predictive patterns. Furthermore, an uncertainty-aware reliability gate is designed to dynamically calibrate the contribution of each segment, effectively suppressing residual background noise. Extensive experiments on real-world and synthetic benchmarks demonstrate that AnchorMoE achieves highly competitive classification performance while faithfully grounding its decisions in the raw time series.

Problem

• Multivariate time series classification
• Interpretability
• Sparse signals
• Background noise
• Temporal segments

Innovation

• Mixture-of-Experts
• Interpretable-by-construction
• Geometric orthogonality constraint
• Uncertainty-aware gating
• Additive decomposition