Since mutual independence is the target, pairwise similarity measurements are of no use, and thus traditional clustering algorithms are inapplicable.

We put forth a novel procedure, that starts with a weighted maximal correlation initialization obtainable with a few power iterations, followed by successive refinements based on iteratively reweighted gradient-type iterations . The novel techniques distinguish themselves from prior works by the inclusion of a fresh (re)weighting regularization.

Related work Although first-order OKL methods cannot achieve logarithmic regret, many approximation methods have been proposed to make them scale to large datasets.

In this work, we propose to apply trust region optimization to deep reinforcement learning using a recently proposed Kronecker-factored approximation to the curvature.

We present a probabilistic framework for nonlinearities, based on doubly truncated Gaussian distributions.

Notably, it typically leads to underestimation of the posterior variance.

Neural Information Processing Systems http://nips.cc/

We present a new affine-invariant optimization algorithm called Online Lazy Newton . The regret of Online Lazy Newton is independent of conditioning: the algorithm's

Autonomous agents optimize the reward function we give them.

However, for some applications, this assumption is inappropriate.