Improving Robustness of Depression Detection Models Using Test-Time Training
Test-time training (TTT) can significantly improve the robustness of depression detection models under various distributional shifts, such as background noise, gender bias, and dataset variations.