Skip to content

Learning Intermediate Time Scales

Home / Open Questions / Learning Intermediate Time Scales

Background: Multi-scale time series forecasting models frequently utilize a binary division between global and local scales, failing to account for potentially important intermediate temporal patterns.

Question / Future Work: Architectures need to be designed to accommodate multiple intermediate time scales beyond the standard global/local dichotomy while maintaining computational efficiency and a low memory footprint.

Why It Matters: Capturing intermediate scales is essential for extremely long sequences where significant temporal variations occur between the extremes of global and local contexts.

Metadata & Links