Abstract: Encoding only the task-related information from the raw data, i.e., disentangled representation learning, can greatly contribute to the robustness and generalizability of models. Although ...