Processing math: 100%
Skip to yearly menu bar Skip to main content


Poster

Stochastic convex optimization with bandit feedback

Alekh Agarwal · Dean P Foster · Daniel Hsu · Sham M Kakade · Sasha Rakhlin


Abstract: This paper addresses the problem of minimizing a convex, Lipschitz function f over a convex, compact set X under a stochastic bandit feedback model. In this model, the algorithm is allowed to observe noisy realizations of the function value f(x) at any query point xX. We demonstrate a generalization of the ellipsoid algorithm that incurs O(\poly(d)T) regret. Since any algorithm has regret at least Ω(T) on this problem, our algorithm is optimal in terms of the scaling with T.

Live content is unavailable. Log in and register to view live content