The research field of time series classification deals with this task. A pioneering algorithm in this field is ROCKET, known for its high accuracy with low computational cost. We have developed a GPU implementation of the algorithm, achieving an 11-fold increase in computational efficiency compared to the previous CPU implementation.
The code is available here: https://github.com/oleeven/CUROCKET.
And the publication can be found here: https://doi.org/10.48550/arXiv.2601.17091
This work is part of the research project #HoleListic, funded by the Federal Ministry for Economic Affairs and Energy under the Federal Aviation Research Programme VI-3. 