An update on developing efficient Bayesian inference approached for validation of RE models is presented. These methods are applied for a JET experiment with an Argon massive gas induced RE beam (#95135). The first proof-of-principle work applied computationally light fluid model with the target to match experimentally observed time evolution of total plasma current in the current quench and RE plateau [1]. After the proof-of-principle, the focus was shifted to DREAM+SOFT simulation of synchrotron emission patterns. This introduces significant complications for the inference as the simulations become computationally significantly more demanding and computing the distance between observed and synthetic images is not trivial. The present focus of the latter is on template matching algorithms. Tests have been started to apply these inference workflows to infer parameters of RE radial transport models, such as Rechester-Rosenbluth, based on observed time evolution of synchrotron patterns. The intention is to present the status of these tests. Another pursued line of research is looking beyond the case based inference. While the Bayesian optimization based approach, discussed above, is already orders of magnitude more data efficient than using the simulator directly, a key disadvantage is that when new observations become available, the inference chain has to be repeated. A more attractive approach is to amortize the upfront training of the surrogate model over many observed data sets from the research domain working on the same model family, such that inference for new observations can be conducted with little computational overhead [2]. The progress in this research will be discussed. [1] A.E. Järvinen, et al. J. Plasma Phys. 88, 905880612 (2022). [2] K. Cranmer, et al. Proc. Natl. Acad. Sci. 117, 30055 (2020).