What is temporal embedding and how is it used in modern time series models?

Temporal embeddings convert time-related features (hour, day, season) into dense vector representations for neural models.

Temporal embeddings encode cyclical and categorical time features into continuous vector spaces. Instead of treating time components as raw integers, embeddings allow models to learn relationships like similarity between weekdays or seasonal effects. These embeddings are commonly used in Transformers and deep forecasting models to improve representation learning.