When does mixup promote local linearity in learned representations?

Arslan Chaudhry, Aditya Krishna Menon, Andreas Veit, Sadeep Jayasumana, Srikumar Ramalingam, Sanjiv Kumar