Interpretability and adaptation to diverse tasks are desirable properties of machine learning (ML) algorithms, especially when they are part of critical systems such as biomedical signal monitoring. In this context, the recently developed paradigm of automated machine learning (AutoML) is leading to a higher degree of autonomous information management as it allows to adjust ML models with reduced human intervention. Natural signals can often be processed in a way such that relevant information lies in a few spots (e.g., frequencies), as opposed to complex representations that are not human-readable. To this end, sparsity-aware signal processing (SP) algorithms can detect the structure that enhances interpretability. This is important in health-related applications as health issues are generally explained by a restricted set of variables.