Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SelectiveFinetuning: Enhancing Transfer Learning in Sleep Staging through Selective Domain Alignment

Siyuan Zhao, Chenyu Liu, Yi Ding, Xinliang Zhou

TL;DR

This work tackles the problem of domain shift in sleep staging caused by EEG variability across subjects and environments. It introduces SelectiveFinetuning, which uses a pretrained Multi-Resolution CNN to extract EEG features and a Domain Aligner based on Earth Mover Distance to selectively finetune on source data that closely matches the target distribution, thereby mitigating negative transfer in a fully transductive setting. The method demonstrates state-of-the-art performance on cross-dataset transfers (SleepEDF and SHHS) and shows improved robustness when labeled target data are scarce. Practically, this approach enables more reliable sleep stage classification in diverse, real-world recording conditions with limited annotation efforts.

Abstract

In practical sleep stage classification, a key challenge is the variability of EEG data across different subjects and environments. Differences in physiology, age, health status, and recording conditions can lead to domain shifts between data. These domain shifts often result in decreased model accuracy and reliability, particularly when the model is applied to new data with characteristics different from those it was originally trained on, which is a typical manifestation of negative transfer. To address this, we propose SelectiveFinetuning in this paper. Our method utilizes a pretrained Multi Resolution Convolutional Neural Network (MRCNN) to extract EEG features, capturing the distinctive characteristics of different sleep stages. To mitigate the effect of domain shifts, we introduce a domain aligning mechanism that employs Earth Mover Distance (EMD) to evaluate and select source domain data closely matching the target domain. By finetuning the model with selective source data, our SelectiveFinetuning enhances the model's performance on target domain that exhibits domain shifts compared to the data used for training. Experimental results show that our method outperforms existing baselines, offering greater robustness and adaptability in practical scenarios where data distributions are often unpredictable.

SelectiveFinetuning: Enhancing Transfer Learning in Sleep Staging through Selective Domain Alignment

TL;DR

This work tackles the problem of domain shift in sleep staging caused by EEG variability across subjects and environments. It introduces SelectiveFinetuning, which uses a pretrained Multi-Resolution CNN to extract EEG features and a Domain Aligner based on Earth Mover Distance to selectively finetune on source data that closely matches the target distribution, thereby mitigating negative transfer in a fully transductive setting. The method demonstrates state-of-the-art performance on cross-dataset transfers (SleepEDF and SHHS) and shows improved robustness when labeled target data are scarce. Practically, this approach enables more reliable sleep stage classification in diverse, real-world recording conditions with limited annotation efforts.

Abstract

In practical sleep stage classification, a key challenge is the variability of EEG data across different subjects and environments. Differences in physiology, age, health status, and recording conditions can lead to domain shifts between data. These domain shifts often result in decreased model accuracy and reliability, particularly when the model is applied to new data with characteristics different from those it was originally trained on, which is a typical manifestation of negative transfer. To address this, we propose SelectiveFinetuning in this paper. Our method utilizes a pretrained Multi Resolution Convolutional Neural Network (MRCNN) to extract EEG features, capturing the distinctive characteristics of different sleep stages. To mitigate the effect of domain shifts, we introduce a domain aligning mechanism that employs Earth Mover Distance (EMD) to evaluate and select source domain data closely matching the target domain. By finetuning the model with selective source data, our SelectiveFinetuning enhances the model's performance on target domain that exhibits domain shifts compared to the data used for training. Experimental results show that our method outperforms existing baselines, offering greater robustness and adaptability in practical scenarios where data distributions are often unpredictable.
Paper Structure (13 sections, 2 equations, 2 figures, 2 tables)

This paper contains 13 sections, 2 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Sleep Staging
  4. Transfer Learning
  5. Method
  6. Pre-trained Model
  7. Feature Extraction
  8. Domain Aligner
  9. Experiments
  10. Dataset and Preprocessing
  11. Baseline Methods
  12. Experiment Results
  13. Conclusion

Figures (2)

  • Figure 1: An overview of our SelectiveFinetuning, which composes of Pre-trained Model, Feature Extraction and Domain Aligner. It extracts features from both source and target domain data using a pre-trained model and calculates their similarity. Then, a policy function selects the source domain data that is most similar to the target domain. These selected data are then used to finetune the model's parameters, reducing the impact of negative transfer and ultimately improving the model's performance on the target domain.
  • Figure 2: This figure shows the effect of aligning data between the source and target domains in the presence of domain shifts. The left plot displays mixed data distributions, while the right plot shows a more aligned distribution after selecting source data similar to the target domain.