# Side-aware Meta-Learning for Cross-Dataset Listener Diagnosis with Subjective Tinnitus

**Authors:** Yun Li, Zhe Liu, Lina Yao, Molly Lucas, Jessica J. M. Monaghan, Yu Zhang  
**arXiv:** [2205.03231](https://arxiv.org/abs/2205.03231)  
**Date:** 2022-05-03  
**Summary page:** https://www.summarizepaper.com/en/paper/2205.03231v1/

## Key points

- The paper addresses the issue of dataset specificity in machine learning models used for tinnitus diagnosis
- Current models are limited to specific datasets that may vary in symptoms, disease severity, and demographic attributes
- Differences in dataset formats can impact model performance
- The authors propose a side-aware meta-learning approach for cross-dataset tinnitus diagnosis
- The proposed approach effectively classifies tinnitus in individuals of different ages and genders from diverse data collection processes
- The method leverages the superiority of meta-learning instead of relying on large-scale datasets
- A subject-specific training process is introduced to help the model fit the data patterns of different patients or healthy individuals
- Incorporating side information about ears enhances model performance and improves learned features' quality
- The proposed method achieves a high accuracy rate of 73.8% in cross-dataset classification for tinnitus diagnosis
- This paper contributes to advancing tinnitus diagnosis by addressing dataset specificity issues through side-aware meta-learning
- Meta-learning can improve model performance without relying on large datasets commonly used in deep learning approaches

## Summary

The paper titled "Side-aware Meta-Learning for Cross-Dataset Listener Diagnosis with Subjective Tinnitus" addresses the issue of dataset specificity in machine learning models used for tinnitus diagnosis. While machine learning has been successfully applied in EEG-based tinnitus analysis, current models are limited to specific datasets that may vary in terms of symptoms, disease severity, and demographic attributes. Additionally, differences in dataset formats can impact model performance. To overcome these limitations, the authors propose a side-aware meta-learning approach for cross-dataset tinnitus diagnosis. This method effectively classifies tinnitus in individuals of different ages and genders from diverse data collection processes. Unlike conventional deep learning models that rely on large-scale datasets, the proposed approach leverages the superiority of meta-learning. The authors also introduce a subject-specific training process to help the model fit the data patterns of different patients or healthy individuals. Through extensive analysis, they demonstrate that incorporating side information about ears enhances model performance and improves the quality of learned features. The results indicate that the proposed method achieves a high accuracy rate of 73.8% in cross-dataset classification for tinnitus diagnosis, suggesting its potential as a valuable tool for diagnosing subjective tinnitus across diverse populations. Overall, this paper contributes to advancing tinnitus diagnosis by addressing dataset specificity issues through side-aware meta-learning. The findings highlight the effectiveness of incorporating side information and demonstrate how meta-learning can improve model performance without relying on large datasets commonly used in deep learning approaches.