Multivariate time series encode structural patterns that unfold across multiple temporal scales, yet most forecasting models treat learned representations as transient byproducts of prediction, leaving the organizational geometry of these patterns underexploited. We introduce M2Patch, a CNN-based forecasting architecture that maps channel-independent multivariate observations into a structured latent space through two complementary differentiable constraints.
M2Patch employs multi-scale patching to decompose the input into overlapping temporal granularities. Depthwise separable convolutions with progressive dilation extract scale-specific features in linear time. Per-scale learned projections compress these features into a compact latent representation.
The latent space is organized by an intra-scale smoothness constraint that enforces temporal continuity between adjacent patches, and an inter-scale alignment constraint, realized through learnable cross-scale mappings, that restores cross-granularity interaction within the channel-independent design, ensuring that all scales encode mutually consistent representations of the underlying dynamics.
Experiments on ten real-world benchmarks show that M2Patch achieves 57 best and 34 second-best results across 40 forecasting settings, matching or exceeding representative baselines on most benchmarks while maintaining linear computational complexity and robustness to patch-level input corruption.
Blogger's Review: The M2Patch model breaks through traditional time series forecasting limitations with its innovative multi-scale structured latent space modeling approach, enhancing prediction accuracy while providing new directions for future research. Its performance in practical applications is particularly impressive, showcasing remarkable computational efficiency and robustness.