Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Long-term series forecasting with Query Selector -- efficient model of sparse attention

About

Various modifications of TRANSFORMER were recently used to solve time-series forecasting problem. We propose Query Selector - an efficient, deterministic algorithm for sparse attention matrix. Experiments show it achieves state-of-the art results on ETT, Helpdesk and BPI'12 datasets.

Jacek Klimek, Jakub Klimek, Witold Kraskiewicz, Mateusz Topolewski• 2021

Related benchmarks

TaskDatasetResultRank
Multivariate Time-series ForecastingETTh2 (test)
MSE0.4124
171
Multivariate Time-series ForecastingETTh1 (test)
MSE0.4226
134
Multivariate Time-series ForecastingETTm1 (test)
MSE0.3351
67
Univariate Time Series ForecastingETTh1 (test)
MSE0.0436
39
Univariate Time Series ForecastingETTm1 (test)
MSE0.0139
15
Activity PredictionHelpdesk (test)
Accuracy74.3
2
Activity PredictionBPI'12 (test)
Accuracy79
2
Showing 7 of 7 rows

Other info

Code

Follow for update