ALER-TI uses retrieval-augmented latent embedding alignment for time series imputation

Researchers propose ALER-TI, a retrieval-augmented framework for time series imputation that leverages historical patterns to supplement degraded local context. The core component, Latent Embedding Alignment (LEA), mitigates representation mismatch between corrupted queries and complete historical candidates by applying post-hoc masking in the latent space.

• ALER-TI aligns candidates with the query's missingness pattern while allowing historical embeddings to be pre-computed and cached for efficient retrieval.
• The framework is model-agnostic and integrates with various imputation backbones through a lightweight adaptation module.
• Experiments on six real-world datasets under different missing rates show ALER-TI consistently improves strong baseline models and enhances robustness across diverse imputation settings.

ALER-TI addresses the limitations of existing architectures that rely primarily on localized temporal context, which can be insufficient for non-stationary dynamics and weak correlations.