EUVIMEDEUVIMED entdecken
Crossref

Structured Tokenization of BERT for Tabular Classification

Dominic Bazina-Grolinger, Yuqi Liu, Scott Murray, Jacob Schaffer, Yimin Yang

IEEE Transactions on Consumer Electronics · 2026-11 · 72 (4) · 10699-10712

DOI: 10.1109/tce.2026.3700485 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Consumer Electronics, 2026-11. Verfasst von Dominic Bazina-Grolinger und 4 weiteren Autor:innen. Die Arbeit nennt 46 Literaturangaben. Gefördert von: Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant Program, NSERC Alliance Program, Western Research Award.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Recent advances in Transformer-based models such as BERT have significantly improved performance across various natural language processing tasks. However, their application to structured tabular data, particularly in brain–computer interface (BCI) scenarios such as electroencephalography (EEG) classification, remains underexplored due to structural incompatibility and modality mismatch. In this work, we propose a novel framework that adapts text-pretrained BERT models for tabular classification, with a focus on EEG signal interpretation. We design a tokenization strategy that transforms multichannel EEG recordings into sequences compatible with BERT input embeddings, enabling contextualized attention across heterogeneous features. Furthermore, we conduct channel importance analysis to identify critical EEG electrodes, improving model interpretability and reducing computational overhead. To evaluate the effectiveness of our approach, we conduct extensive experiments on 15 public tabular datasets, including complex EEG benchmarks. Results show that our method achieves stable and competitive performance in both full-data and data-limited learning settings. These findings demonstrate the potential of integrating pretrained language models into structured data domains and open new directions for cross-modal representation learning in BERT interface applications.

Förderung

  • Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant Program
  • NSERC Alliance Program
  • Western Research Award

Zitieren

APA

Bazina-Grolinger, D., Liu, Y., Murray, S., Schaffer, J., & Yang, Y. (2026). Structured Tokenization of BERT for Tabular Classification. IEEE Transactions on Consumer Electronics, 72(4), 10699-10712. https://doi.org/10.1109/tce.2026.3700485

Vancouver

Bazina-Grolinger D, Liu Y, Murray S, Schaffer J, Yang Y. Structured Tokenization of BERT for Tabular Classification. IEEE Transactions on Consumer Electronics. 2026;72(4):10699-10712. doi:10.1109/tce.2026.3700485

BibTeX
@article{BazinaGrolinger2026,
  author = {Bazina-Grolinger, Dominic and Liu, Yuqi and Murray, Scott and Schaffer, Jacob and Yang, Yimin},
  title = {Structured Tokenization of BERT for Tabular Classification},
  journal = {IEEE Transactions on Consumer Electronics},
  year = {2026},
  volume = {72},
  number = {4},
  pages = {10699--10712},
  doi = {10.1109/tce.2026.3700485}
}
Kostenlos starten →