Search papers, labs, and topics across Lattice.
This paper introduces POEM, a novel phase-aware forecasting framework designed to tackle periodicity drift in time series data by employing latent feature rotation via the special orthogonal group $\mathrm{SO}(2)$. By learning a phase-correction coordinate and utilizing Directional Phase Increment Attention (DPIA) to incorporate historical phase increments, POEM effectively reduces phase-related variability in forecasts. Experimental results indicate that POEM not only performs competitively against existing methods but also enhances the regularity of latent trajectories, highlighting its potential for improved forecasting accuracy in dynamic environments.
POEM's innovative use of $\mathrm{SO}(2)$ rotations reveals how to effectively manage periodicity drift, transforming time series forecasting.
Deep learning has advanced time series forecasting, but periodicity drift, in which cycle timing and phase vary over time, remains a challenging problem. Existing methods predominantly model these sequences on fixed time grids, suffering from a limited ability to accommodate phase-related variation. To address this limitation, we propose \textbf{POEM}, a phase-aware forecasting framework based on latent feature rotation using the special orthogonal group in two dimensions, denoted by $\mathrm{SO}(2)$. POEM aims to reduce the phase-related variability by learning a phase-correction coordinate and applying an invertible $\mathrm{SO}(2)$-based rotation to paired latent features. To extrapolate this correction coordinate, Directional Phase Increment Attention (DPIA) retrieves historical phase increments from similar temporal contexts and integrates them into future phase corrections. Experiments demonstrate that POEM achieves competitive performance, while qualitative visualizations suggest that the learned phase-aware transformation makes latent trajectories more regular.
