[4B2-GS-1] Fundamental AI, theory (1)

[4B2-GS-1-04] A Study on Estimation of PAC-Bayesian Transportation Distance

〇Kohei Miyaguchi¹ (1. IBM Research - Tokyo)

Keywords:Generalization error bound, PAC-Bayesian theory, PAC-Bayesian transportation distance

Empirically, the PAC-Bayesian analysis is known to produce tight risk bounds for practical machine learning algorithms. However, in its naive form, it cannot incorporate the metric structure of the prediction models. To fill this gap, the PAC-Bayesian transportation (PAC-BT) distance has been proposed recently.
In this paper, we tackle the problem that there is no general method to estimate the PAC-BT bound applicable to a wide class of learning models. As a result, we have found that the distance can be estimated where iterative learning algorithms are involved.