Bayesian Uncertainty-aware Deep Learning with noisy labels: Tackling annotation ambiguity in EEG seizure detection
Deeksha M. Shama, Archana Venkataraman · 2026
Read Bayesian Uncertainty-aware Deep Learning with noisy labels: Tackling annotation ambiguity in EEG seizure detection by Deeksha M. Shama, Archana Venkataraman (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0352191.txt.
Description
Deeksha Shama and Archana Venkataraman tackled a dirty-data problem in automatic seizure detection: the human labels used to train detectors are often wrong about when a seizure starts and ends. Scalp recordings are noisy, so even experts disagree, and standard training treats every label as truth. Their method, BUNDL, instead teaches the network how uncertain each label is, inside a Bayesian frame.
A new training rule built on that idea lets the model lean on confident labels and discount shaky ones, without adding any new parts to the network. Tests on simulated recordings and the follow-on job of locating where seizures begin checked whether cleaner learning follows. For hospitals short on expert annotators, a detector that expects label noise is far more realistic than one that demands perfect teachers.
Catalog record
| CREATOR | Deeksha M. Shama, Archana Venkataraman |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 83123 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |