Expected complexity analysis of stochastic direct-search

This work presents the convergence rate analysis of stochastic variants of the broad class of direct-search methods of directional type. It introduces an algorithm designed to optimize differentiable objective functions $f$ whose values can only be computed through a stochastically noisy blackbox. The proposed stochastic directional direct-search (SDDS) algorithm accepts new iterates by imposing a … Read more

Stochastic mesh adaptive direct search for blackbox optimization using probabilistic estimates

We present a stochastic extension of the mesh adaptive direct search (MADS) algorithm originally developed for deterministic blackbox optimization. The algorithm, called StoMADS, considers the unconstrained optimization of an objective function f whose values can be computed only through a blackbox corrupted by some random noise following an unknown distribution. The proposed method is based … Read more

HyperNOMAD: Hyperparameter optimization of deep neural networks using mesh adaptive direct search

The performance of deep neural networks is highly sensitive to the choice of the hyperparameters that define the structure of the network and the learning process. When facing a new application, tuning a deep neural network is a tedious and time consuming process that is often described as a “dark art”. This explains the necessity … Read more

Non-asymptotic Results for Langevin Monte Carlo: Coordinate-wise and Black-box Sampling

Euler-Maruyama and Ozaki discretization of a continuous time diffusion process is a popular technique for sampling, that uses (upto) gradient and Hessian information of the density respectively. The Euler-Maruyama discretization has been used particularly for sampling under the name of Langevin Monte Carlo (LMC) for sampling from strongly log-concave densities. In this work, we make … Read more

The Mesh Adaptive Direct Search Algorithm for Granular and Discrete Variables

The mesh adaptive direct search (Mads) algorithm is designed for blackbox optimization problems for which the functions defining the objective and the constraints are typically the outputs of a simulation seen as a blackbox. It is a derivative-free optimization method designed for continuous variables and is supported by a convergence analysis based on the Clarke … Read more

A Taxonomy of Constraints in Black-Box Simulation-Based Optimization

The types of constraints encountered in black-box simulation-based optimization problems differ significantly from those addressed in nonlinear programming. We introduce a characterization of constraints to address this situation. We provide formal definitions for several constraint classes and present illustrative examples in the context of the resulting taxonomy. This taxonomy, denoted KARQ, is useful for modeling … Read more

Use of a Biobjective Direct Search Algorithm in the Process Design of Material Science Applications

This work describes the application of a direct search method to the optimization of problems of real industrial interest, namely three new material science applications designed with the FactSage software. The search method is BiMADS, the biobjective version of the mesh adaptive direct search (MADS) algorithm, designed for blackbox optimization. We give a general description … Read more

Linear equalities in blackbox optimization

The Mesh Adaptive Direct Search (Mads) algorithm is designed for blackbox optimization problems subject to general inequality constraints. Currently, Mads does not support equalities, neither in theory nor in practice. The present work proposes extensions to treat problems with linear equalities whose expression is known. The main idea consists in reformulating the optimization problem into … Read more

Dynamic scaling in the Mesh Adaptive Direct Search algorithm for blackbox optimization

Blackbox optimization deals with situations in which the objective function and constraints are typically computed by launching a time-consuming computer sim- ulation. The subject of this work is the Mesh Adaptive Direct Search (MADS) class of algorithms for blackbox optimization. We propose a way to dynamically scale the mesh, which is the discrete spatial structure … Read more

Hedge algorithm and Dual Averaging schemes

We show that the Hedge algorithm, a method that is widely used in Machine Learning, can be interpreted as a particular instance of Dual Averaging schemes, which have recently been introduced by Nesterov for regret minimization. Based on this interpretation, we establish three alternative methods of the Hedge algorithm: one in the form of the … Read more