Giving Sensors a Voice: Multimodal JEPA for Semantic Time-Series Embeddings

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Dutta, Utsav, Pastrana, Gerardo, Pakazad, Sina Khoshfetrat, Ohlsson, Henrik
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911733702983680
author Dutta, Utsav
Pastrana, Gerardo
Pakazad, Sina Khoshfetrat
Ohlsson, Henrik
author_facet Dutta, Utsav
Pastrana, Gerardo
Pakazad, Sina Khoshfetrat
Ohlsson, Henrik
contents Transformer-based architectures have advanced sequence modeling in language and vision, yet general-purpose representation learning for heterogeneous multivariate time series remains underexplored. We introduce CHARM (Channel-Aware Representation Model), which incorporates channel-level textual descriptions into a Transformer encoder equivariant to channel order. CHARM is trained with a Joint Embedding Predictive Architecture (JEPA) and a novel loss promoting informative, temporally stable embeddings; latent-space prediction encourages robustness to sensor noise while description-aware gating provides interpretability through learned inter-channel relationships. Across anomaly detection, classification, and short- and long-term forecasting, the learned embeddings achieve strong performance using only a linear probe. Performance is driven primarily by the JEPA objective and conditioning architecture, with text descriptions serving as channel identifiers for cross-dataset generalization.
format Preprint
id arxiv_https___arxiv_org_abs_2605_31580
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Giving Sensors a Voice: Multimodal JEPA for Semantic Time-Series Embeddings
Dutta, Utsav
Pastrana, Gerardo
Pakazad, Sina Khoshfetrat
Ohlsson, Henrik
Machine Learning
I.2.6; I.5.1
Transformer-based architectures have advanced sequence modeling in language and vision, yet general-purpose representation learning for heterogeneous multivariate time series remains underexplored. We introduce CHARM (Channel-Aware Representation Model), which incorporates channel-level textual descriptions into a Transformer encoder equivariant to channel order. CHARM is trained with a Joint Embedding Predictive Architecture (JEPA) and a novel loss promoting informative, temporally stable embeddings; latent-space prediction encourages robustness to sensor noise while description-aware gating provides interpretability through learned inter-channel relationships. Across anomaly detection, classification, and short- and long-term forecasting, the learned embeddings achieve strong performance using only a linear probe. Performance is driven primarily by the JEPA objective and conditioning architecture, with text descriptions serving as channel identifiers for cross-dataset generalization.
title Giving Sensors a Voice: Multimodal JEPA for Semantic Time-Series Embeddings
topic Machine Learning
I.2.6; I.5.1
url https://arxiv.org/abs/2605.31580