Self-Certifying Classification by Linearized Deep Assignment

TL;DR

This paper introduces a new deep stochastic classifier for graph data that leverages PAC-Bayes risk bounds to provide tight, data-dependent risk certificates, improving both training and evaluation efficiency.

Contribution

It presents a novel linearly parametrized deep assignment flow model that enables risk certification and efficient out-of-sample risk estimation within the PAC-Bayes framework.

Findings

Achieves tight out-of-sample risk certificates.

Demonstrates competitive performance on graph classification tasks.

Offers a practical approach for self-certifying classifiers.

Abstract

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 deep assignment flows with random initial conditions. Building on the recent PAC-Bayes literature and data-dependent priors, this approach enables (i) to use risk bounds as training objectives for learning posterior distributions on the hypothesis space and (ii) to compute tight out-of-sample risk certificates of randomized classifiers more efficiently than related work. Comparison with empirical test set errors illustrates the performance and practicality of this self-certifying classification method.