Transfer learning is a principled methodology to exploit knowledge of an external agent in order to improve learning of a primary agent. This framework has become immensely useful in a multitude of theoretically and practically oriented scientific fields, including reinforcement learning, deep learning, autonomous driving, natural language processing, etc. However, standard transfer learning strategies are typically based on a complete stochastic dependence structure being specified between the participating learning procedures. The talk will present algorithms for knowledge transfer between Bayesian filters without the need to adopt any such restrictive assumptions. The solution relies on the fully probabilistic design of an unknown probability distribution which conditions on knowledge in the form of an external probability distribution. The talk will demonstrate that the proposed algorithms outperform alternative knowledge transfer strategies in the Kalman filtering context.