L-Drive: Beyond a Single Mapping-Latent Context Drives Time Series Forecasting

In three linesL-Drive introduces a change-aware forecasting framework for multivariate time-series that uses latent context to characterize evolving dynamics and gating mechanisms to modulate representations. Patch-shared relative positional basis functions strengthen structural modeling and reduce overfitting, improving accuracy at regime transitions.

Summary generated by Claude — human-verified