TSRM: A Lightweight Temporal Feature Encoding Architecture for Time Series Forecasting and Imputation
Robert Leppich, Michael Stenger, Daniel Grillmeyer, Vanessa Borst, Samuel Kounev · 2025
Read TSRM: A Lightweight Temporal Feature Encoding Architecture for Time Series Forecasting and Imputation by Robert Leppich, Michael Stenger, Daniel Grillmeyer, Vanessa Borst, Samuel Kounev (2025) 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-2504-18878.txt.
Description
Robert Leppich and co-authors introduce in 2025 a light architecture for forecasting and filling gaps in multivariate time series. Their model builds representation layers based on convolutions, each tasked with learning a different temporal pattern. An attention layer then extracts features and a merge layer combines them, in a layout inspired by the transformer encoder with self attention at its core.
Across seven established benchmark data sets, the model beat leading approaches on most tasks for both forecasting and gap filling. It also used far fewer learnable parameters than those rivals. Source code is released. Readers who work with sensor, finance, or other time series data get a cheaper model to try.
Catalog record
| CREATOR | Robert Leppich, Michael Stenger, Daniel Grillmeyer, Vanessa Borst, Samuel Kounev |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 67 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-18878