Beyond Softmax: A New Perspective on Gradient Bandits

Emerson Melo, David Müller

We establish a link between a class of discrete choice models and the theory of online learning and multi-armed bandits. Our contributions are: (i) sublinear regret bounds for a br…