01:30 PM - 01:55 PM
Automatic Differentiation - theory, recent developments, and tools
Automatic Differentiation (AD) is a well established technique for computing derivatives and gradients, which can be used in optimization and many other applications. We will review the theory of AD and practical challenges when applying it to real applications. We will talk about the use of AD in gradient-based optimization, and the use of (mostly gradient-free) optimization to accelerate AD. Finally, we will discuss recent developments including the use of AD in machine learning, AD for parallel programs, and emerging AD tools in Julia and other languages.
01:55 PM - 02:20 PM
Online automatic optimization of software for numerical algebra
We introduce a new tool that automatically tunes a solver written in Julia. Given a set of parameters defined by the user, our tool uses NOMAD to solve a black-box optimization problem. The black box measures the solver performance on a user-defined test set as a function of solver parameters exposed by the solver interface. The optimization process is done automatically using continuous integration software and load balancing from inside a pull request. The pull request is updated to suggest a good set of algorithmic parameters.
02:20 PM - 02:45 PM
Singular Perturbated Problems and Julia Package in Optimal Control Problem
We report ongoing work on a Julia package, extending previous developments in numerical methods for optimal control including Hampath  and ct: control toolbox . The focus is on indirect methods and path following combined with automatic differentiation. Applications on singularly perturbated optimal control problems with turnpike properties are presented.
02:45 PM - 03:10 PM
Optimization with partial differential equations constraints in Julia
In this presentation, we showcase a new optimization infrastructure to model and solve PDE-constrained problems in the Julia programming language. We build upon the JuliaSmoothOptimizers infrastructure for modeling and solving continuous optimization problems, and present a collection of packages for PDE-constrained problems. We conclude with examples and simulations.