Transformer-Based Time Series Classification for the OpenPack Challenge 2022

Tomoki Uchiyama

Abstract: This report describes a solution for the OpenPack Challenge 2022 developed by tomoon team. We present a transformer-based network for time-series classification on various data modalities. Time series data for each modality is segmented at a fixed interval and then each of the segments is fed into a transformer encoder. After that, we fuse the extracted features and learn multimodal features using a transformer encoder. We achieved 0.963 F1-macro and took first place in the competition.