NPMixer: Hierarchical Neighboring Patch Mixing for Time Series Forecasting
Jung Min Choi, Vijaya Krishna Yalavarthi, Lars Schmidt-Thieme · 2026
Read NPMixer: Hierarchical Neighboring Patch Mixing for Time Series Forecasting by Jung Min Choi, Vijaya Krishna Yalavarthi, Lars Schmidt-Thieme (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-07476.txt.
Description
Jung Min Choi, Vijaya Krishna Yalavarthi, and Lars Schmidt-Thieme forecast multivariate time series with a hierarchical patch-mixing network. Forecasting must capture both local patterns and dependencies across variables at several scales. NPMixer opens with a learnable wavelet transform that splits signals into trend and detail parts, adapting its filters to the data.
Mixer blocks then learn patterns within and across non-overlapping neighboring patches, widening the view step by step. A channel-mixing encoder handles the high-frequency detail while the global trend stays stable. Across seven benchmark datasets and twenty-eight setups, the model beat leading rivals in most configurations on squared error.
Catalog record
| CREATOR | Jung Min Choi, Vijaya Krishna Yalavarthi, Lars Schmidt-Thieme |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.07476