Multi-scale information is crucial for multivariate time series modeling. However, most existing time series multi-scale analysis methods treat all variables in the same manner, making them unsuitable for Irregular Multivariate Time Series (IMTS), where variables have distinct origin scales/sampling rates. To fill this gap, we propose Hi-Patch, a hierarchical patch graph network. Hi-Patch encodes each observation as a node, represents and captures local temporal and inter-variable dependencies of densely sampled variables through an intra-patch graph layer, and obtains patch-level nodes through aggregation. These nodes are then updated and re-aggregated through a stack of inter-patch graph layers, where several scale-specific graph networks progressively extract more global temporal and inter-variable features of both sparsely and densely sampled variables under specific scales. The output of the last layer is fed into task-specific decoders to adapt to different downstream tasks. Experiments on 8 datasets demonstrate that Hi-Patch outperforms state-of-the-art models in IMTS forecasting and classification tasks.

In many fields like healthcare or environmental monitoring, data is collected over time at different speeds—for example, heart rate every second vs. lab results once a day. Most AI models struggle with this kind of irregular data. We introduce Hi-Patch, a new method that respects these differences. It groups data into small units and uses a layered graph approach to find both local and global patterns, even when variables are sampled unevenly. Hi-Patch outperforms leading models on eight real-world datasets, improving forecasting and classification in complex time-based data.