First, a deep learning model is trained using a large publicly available dataset of CK + and FER2013 on anger, fear, and surprise in the source domain. The trained model was then finetuned in the target domain using the N-way-k-shot approach on happy, sad, and neutral classes ...