CHARM learns semantic time-series embeddings via channel-aware JEPA training in an order-equivariant Transformer, achieving strong linear-probe performance on anomaly detection, classification, and forecasting.
unknown sensor
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Giving Sensors a Voice: Multimodal JEPA for Semantic Time-Series Embeddings
CHARM learns semantic time-series embeddings via channel-aware JEPA training in an order-equivariant Transformer, achieving strong linear-probe performance on anomaly detection, classification, and forecasting.