HELIX: Hybrid Encoding with Learnable Identity and Cross-dimensional Synthesis for Time Series Imputation

2026-05-04

AI Summary

Existing time series imputation methods struggle to model the complex dependencies between heterogeneous spatial locations and semantic variables, and often lack a unified feature representation mechanism. To address this, this work proposes learnable feature identity embeddings as persistent anchors and introduces a hybrid temporal-feature attention mechanism that captures arbitrary cross-dimensional dependencies in an end-to-end manner. This approach jointly models semantic and physical relationships without requiring predefined graph structures. Furthermore, a hierarchical alignment strategy is employed to significantly enhance imputation consistency and accuracy. Evaluated across five public datasets under 21 distinct settings, the proposed method consistently outperforms 16 baseline models, achieving state-of-the-art performance.

Abstract

Time series imputation benefits from leveraging cross-feature correlations, yet existing attention-based methods re-discover feature relationships at each layer, lacking persistent anchors to maintain consistent representations. To address this, we propose HELIX, which assigns each feature a learnable feature identity, a persistent embedding that captures intrinsic semantic properties throughout the network. Unlike graph-based methods that rely on predefined topology and assume homogeneous spatial relationships, HELIX learns arbitrary feature dependencies end-to-end from temporal co-variation, naturally handling datasets where features mix spatial locations with semantic variables. Integrated with hybrid temporal-feature attention, HELIX achieves the state-of-the-art performance, surpassing all 16 baselines on 5 public datasets across 21 experimental settings in our evaluation. Furthermore, our mechanistic analysis reveals that HELIX aligns learned feature identities and dependencies with latent physical and semantic structure progressively across layers, demonstrating that it more effectively translates cross-feature structure into imputation accuracy.

Problem

• time series imputation
• cross-feature correlations
• attention-based methods
• feature representation
• missing data

Innovation

• learnable feature identity
• cross-dimensional synthesis
• hybrid attention
• time series imputation
• end-to-end dependency learning