Marvin, Computer Engineer and Biostatistician, is the head of the Emmy Noether research group on interpretable machine learning, funded by the German Research Foundation, at the Leibniz Institute for Prevention Research and Epidemiology – BIPS in Bremen, Germany. Since February 2021, he is also Professor of Machine Learning in Statistics at the University of Bremen. He has a research focus on statistical learning and interpretable machine learning and is interested in epidemiological applications to high-dimensional genetic data and longitudinal register data. Marvin is also author of several R packages, including the random forest package ranger. Marvin presented the results of his latest paper, just accepted in Machine Learning journal, explaining the conditional predictive impact (CPI), a model-agnostic interpretable machine learning method which can handle correlated predictor variables and adjust for confounders. The method builds on the knockoff framework of Candès et al. (2018) and works in conjunction with any valid knockoff sampler, supervised learning algorithm, and loss function. Marvin briefly described the method, show selected simulation results and give an example (with R code) of the application. The CPI has been implemented in an R package, cpi, which can be downloaded from this https URL.