Uncertainty-aware Semi-supervised Ensemble for Multilingual Depression Detection

This paper addresses the challenge of multilingual depression detection, particularly in resource-scarce scenarios. The proposed Semi-SMDNet framework leverages semi-supervised learning, ensemble methods, and uncertainty-aware pseudo-labeling to improve performance across multiple languages. The focus on handling noisy data and improving robustness is crucial for real-world applications. The use of ensemble learning and uncertainty-based filtering are key contributions.