Machine learning methods operate on formal representations of the data at hand and the models or patterns induced from the data. They also assume a suitable formalization of the learning task itself (e.g., as a classification problem), including a specification of the objective in terms of a suitable performance metric, and sometimes other criteria the induced model is supposed to meet. Different representations or problem formalizations may be more or less appropriate to address a particular task and to deal with the type of training information available. The goal of LEMUR is to create a novel branch of machine learning we call Learning with Multiple Representations. We aim to develop the theoretical foundations and a first set of algorithms for this new paradigma. Moreover, corresponding applications are to demonstrate the usefulness of the new family of approaches. We regard LEMUR as very timely, as LMR algorithms will allow to flexible representations (e.g., suitable for explainability, fairness) with diverse target functions (e.g., incorporating environmental or even social impact) so as to make the induced models abide by the Green Charter and trustworthy AI criteria by design. We will focus on learning with weak supervision because it addresses one of the major flaws of modern ML approaches, i.e., their data hunger, by means of weaker sources of labelling for training data. The outcome of the DN will be a set of 10 experts trained to implement the third and subsequent waves of AI in Europe. The highly interdisciplinary and intersectoral context in which they will be trained will provide them with research-related and transferable competences relevant to successful careers in central AI areas.