Adapting Tensor Kernel Machines to Enable Efficient Transfer Learning for Seizure Detection

TL;DR

The paper introduces Adapt-TKM, an efficient transfer-learning framework that personalizes a source tensor-kernel model for seizure detection using behind-the-ear EEG. By regularizing the adapted model toward the source in the primal with a low-rank CPD-based tensor representation, it achieves effective personalization with far fewer parameters than adaptive SVM, enabling faster on-device inference. Applied to SeizeIT data, Adapt-TKM yields improvements for several patients and demonstrates substantial parameter and speed advantages, making it well-suited for resource-constrained wearables while reducing privacy concerns. Overall, the work advances privacy-conscious, parameter-efficient transfer learning for non-linear EEG-based seizure detection with practical edge-device applicability.

Abstract

Transfer learning aims to optimize performance in a target task by learning from a related source problem. In this work, we propose an efficient transfer learning method using a tensor kernel machine. Our method takes inspiration from the adaptive SVM and hence transfers 'knowledge' from the source to the 'adapted' model via regularization. The main advantage of using tensor kernel machines is that they leverage low-rank tensor networks to learn a compact non-linear model in the primal domain. This allows for a more efficient adaptation without adding more parameters to the model. To demonstrate the effectiveness of our approach, we apply the adaptive tensor kernel machine (Adapt-TKM) to seizure detection on behind-the-ear EEG. By personalizing patient-independent models with a small amount of patient-specific data, the patient-adapted model (which utilizes the Adapt-TKM), achieves better performance compared to the patient-independent and fully patient-specific models. Notably, it is able to do so while requiring around 100 times fewer parameters than the adaptive SVM model, leading to a correspondingly faster inference speed. This makes the Adapt-TKM especially useful for resource-constrained wearable devices.

Figures (7)

• Figure 1: Seizure detection pipeline.
• Figure 2: Decision boundaries of different TKRR models on the synthetic target data. The orange dots show the negative samples, the blue triangles the positive samples and the samples used for training are highlighted by the black squares. The shaded blue area shows where the classifier labels samples as positive.
• Figure 3: ROC curves of the TKRR models
• Figure 4: ROC curves of the SVM models
• Figure 5: F1-scores per patient for the TKRR models.