[106th TrustML Young Scientist Seminar] Talk by Wen-Bo Du (Nanjing University) "Non-Parametric Rehearsal Learning via Conditional Mean Embeddings"

An online research talk in RIKEN AIP's TrustML Young Scientist Seminar series. Wen-Bo Du, a Ph.D. student in the LAMDA Group at Nanjing University, presents a non-parametric approach to rehearsal learning, a line of work on avoiding undesired future outcomes by choosing actions based on influence relations. The method formulates the objective in a reproducing kernel Hilbert space using conditional mean embeddings, replaces the discontinuous desirability indicator with a smooth Probit surrogate, and estimates action-dependent outcome distributions through nested kernel ridge regression. The estimator is identifiable from observational data and comes with finite-sample error bounds and consistency guarantees, with experiments on synthetic and semi-synthetic benchmarks.