[home] [Personal Program] [Help]

---

Session: Brain
Session starts: Friday 27 January, 14:00
Presentation starts: 14:00
Room: Room 558

---

Seline de Rooij (TU Delft)
Kim Batselier (TU Delft)
Borbála Hunyadi (TU Delft)

Abstract:
Due to recent advances in wearable EEG devices, accurate seizure detection algorithms have become even more important. The ever-increasing size of the generated datasets poses a significant challenge to many existing seizure detection methods based on kernel machines such as SVMs. Typically, this problem is mitigated by significantly undersampling the majority class, but in practice, these methods tend to suffer from too many false alarms. Recent works have proposed tensor networks to enable large-scale classification with kernel machines. In our work, we explore the use of one such method, the tensor-network Kalman filter for LS-SVMs (TNKF-LSSVM), in seizure detection, as we hypothesize that using more data will improve the detection performance. This hypothesis was tested using data from the Temple University Hospital Seizure Corpus (TUSZ). Our results show that the TNKF-LSSVM performs comparably to a regular LS-SVM in detecting seizures when both are trained on the same (smaller) dataset (AUCLS-SVM = 0.797 vs. AUC¬TNKF-LSSVM¬ = 0.811). We can also show that it can be trained on truly large-scale data where LS-SVM cannot (N>10^5 ). However, for the presented model configuration detection performance does not seem to improve with more input data.