PAMNet: A New Neural Network for Multivariate Time Series Forecasting

A research paper on arXiv introduces PAMNet (Cycle-aware Phase-Amplitude Modulation Network), a novel neural network for multivariate time series forecasting. The paper, arXiv:2605.02938, addresses limitations in existing methods that either implicitly extract periodicity with high computational cost or overlook phase-amplitude coupling. PAMNet explicitly decomposes periodic patterns into phase and amplitude components using a dual-branch modulator with learnable embeddings. The phase branch captures phase-dependent mean shifts via cyclical embeddings, while the amplitude branch models intensity variations. A lightweight modulator with element-wise fusion combines these components efficiently. The approach aims to improve forecasting accuracy by leveraging reliable periodic patterns.