Learning complementary representations via attention-based ensemble learning for cough-based COVID-19 recognition

Coughs sounds have shown promising as a potential marker for distinguishing COVID individuals from non-COVID ones. In this paper, we propose an attention-based ensemble learning approach to learn complementary representations cough samples. Unlike most traditional schemes such mere maxing or averaging, the proposed fairly considers contribution of representation generated by each single model. The attention mechanism is further investigated at feature level and decision level. Evaluated on Track-1 test set DiCOVA challenge 2021, experimental results demonstrate that feature-level achieves best performance (Area Under Curve, AUC: 77.96%), resulting in 8.05% improvement over baseline.