🏢 University of Vienna

Adaptive gradient descent methods are improved by leveraging local curvature information for entirely adaptive algorithms without added computational cost, proving convergence with only local Lipschit…