Characterising the feasibility of private learning in various settings involves identifying parameters or metrics that can accurately predict the success of learning algorithms under privacy constraints. This includes understanding which types of data, learning tasks, and algorithmic approaches are amenable to private learning. Researchers aim to develop theoretical frameworks that can guide the application of differential privacy in diverse contexts, from simple classification tasks to more complex scenarios like regression and multi-class classification. This characterisation not only helps in predicting the performance of private learning algorithms but also aids in designing new algorithms optimised for specific tasks and settings.