Outstanding Paper Award at the Korea Software Congress 2023 (KSC 2023)

- Big Data Processing and DB Laboratory -

The paper 'A GRU-based Time Series Forecasting Method Using Patching' by Kim Yun-young, graduate student in the Department of Computer Science & Engineering (supervisor: Jung Sung-won), received the Outstanding Paper Award at the Korea Software Congress 2023 (KSC 2023), held over three days from 20 to 22 December.

▲ From left: Kim Yun-young (master's, 4th semester), Jung Sung-won (supervisor)

▲ Certificate of the Outstanding Paper Award, Korea Software Congress 2023 (KSC 2023)

Time series forecasting is very important as an aid to decision-making in companies and in the field. Recently the transformer-based patch time series Transformer (PatchTST) and the MLP-based Long-term time series forecasting Linear (LTSF-Linear) have shown good performance in time series forecasting. PatchTST, however, takes a long time to train and infer, while LTSF-Linear, because of its structural simplicity, learns only a limited amount of the information contained in the training data.

To address this, the paper proposes patch time series GRU (PatchTSG), which applies a Gated Recurrent Unit (GRU) to patched data to reduce training time while capturing sufficient learnable information from time series data.

As a result, PatchTSG reduced training time by up to 82% and inference time by up to 46% compared with PatchTST.