The computational bottleneck in applying online learning to massive data sets is usually the projection step. We present efficient online learning algorithms that eschew projections in favor of much more efficient linear optimization steps using the Frank-Wolfe technique. We obtain a range of regret bounds for online convex optimization, with better bounds for specific cases such as stochastic ...

A formula for the transition density of a Markov process defined by an infinite-dimensional stochastic equation is given in terms of the Ornstein–Uhlenbeck bridge and a useful lower estimate on the density is provided. As a consequence, uniform exponential ergodicity and V ergodicity are proved for a large class of equations. We also provide computable bounds on the convergence rates and the sp...

A formula for the transition density of a Markov process defined by an infinitedimensional stochastic equation is given in terms of the Ornstein Uhlenbeck Bridge, and a useful lower estimate on the density is provided. As a consequence, uniform exponential ergodicity and V-ergodicity are proven under suitable conditions for a large class of equations. The method allows us to find computable bou...

We consider the solution of stochastic dynamic programs using sample path estimates. Applying the theory of large deviations, we derive probability error bounds associated with the convergence of the estimated optimal policy to the true optimal policy, for finite horizon problems. These bounds decay at an exponential rate, in contrast with the usual canonical (inverse) square root rate associat...

We present sharp bounds on the supremum norm of DjSh for j ≥ 2, where D is the differential operator and S the Stein operator for the standard normal distribution. The same method is used to give analogous bounds for the exponential, Poisson and geometric distributions, with D replaced by the forward difference operator in the discrete case. We also discuss applications of these bounds to the c...

We consider the stochastic wave equation on the real line driven by space–time white noise and with irregular initial data. We give bounds on higher moments and, for the hyperbolic Anderson model, explicit formulas for second moments. These bounds imply weak intermittency and allow us to obtain sharp bounds on growth indices for certain classes of initial conditions with unbounded support. c ⃝ ...

We present a method for computing lower bounds in the Progressive Hedging Algorithm (PHA) for two-stage and multi-stage stochastic mixedinteger programs. Computing lower bounds in the PHA allows one to assess the quality of the solutions generated by the algorithm contemporaneously. The lower bounds can be computed in any iteration of the algorithm by using dual prices that are calculated durin...

In the stochastic description of biochemical reaction systems, the time evolution of statistical moments for species population counts is described by a linear dynamical system. However, except for some ideal cases (such as zero- and first-order reaction kinetics), the moment dynamics is underdetermined as lower-order moments depend upon higher-order moments. Here, we propose a novel method to ...

This paper studies multiclass loss systems with two layers of servers, where each server at the first layer is dedicated to a certain customer class, while the servers at the second layer can handle all customer classes. The routing of customers follows an overflow scheme, where arriving customers are preferentially directed to the first layer. Stochastic comparison and coupling techniques are ...

Wedevelop computational bounds onperformance for causal state feedback stochastic controlwith linear dynamics, arbitrary noise distribution, and arbitrary input constraint set. This can be very useful as a comparison with the performance of suboptimal control policies, which we can evaluate using Monte Carlo simulation. Our method involves solving a semidefinite program (a linear optimization p...