New POEM framework tackles time series forecasting with phase-aware rotation

Researchers have developed POEM, a novel framework designed to improve time series forecasting by addressing the challenge of periodicity drift. This drift occurs when the timing and phase of cycles within a time series vary over time, a problem that current methods struggle to handle due to their reliance on fixed time grids. POEM utilizes latent feature rotation based on the special orthogonal group SO(2) to learn a phase-correction coordinate and apply an invertible rotation to latent features, thereby reducing phase-related variability. The framework incorporates Directional Phase Increment Attention (DPIA) to learn from historical phase increments and integrate them into future corrections, showing competitive performance in experiments. AI

IMPACT This research could lead to more accurate forecasting models for systems with drifting periodicities, impacting fields like finance and meteorology.

RANK_REASON The cluster contains an academic paper detailing a new method for time series forecasting. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Directional Phase Increment Attention
• Hugging Face

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →