Self‐Certifying Classification by Linearized Deep Assignment

We propose a novel class of deep stochastic predictors for classifying metric data on graphs within the PAC-Bayes risk certification paradigm. Classifiers are realized as linearly parametrized assignment flows with random initial conditions. Building recent literature and data-dependent priors, this approach enables (i) to use bounds training objectives learning posterior distributions hypothesis space (ii) compute tight out-of-sample certificates randomized classifiers more efficiently than related work. Comparison empirical test set errors illustrates performance practicality self-certifying classification method.