Detail of the student project

Topic:Mixup with neighbors
Department:Katedra kybernetiky
Supervisor:Georgios Tolias, Ph.D.
Announce as:Bakalářská práce, Semestrální projekt
Description:Mixup is a data-dependent regularization technique that consists in linearly interpolating input samples and associated outputs. In this project we will investigate how to improve it by mixing examples according to their proximity in representation spaces given by internal network activations.
Bibliography:Baena et al, arxiv 2022, Preventing Manifold Intrusion with Locality: Local Mixup
Zhang et al, ICLR 2018, mixup: Beyond Empirical Risk Minimization

Responsible person: Petr Pošík