Charlie Vanaret

I will be giving a talk at the SIAM OP21 conference (July 20-23) about UNO, an open-source framework for unifying nonlinear optimization methods that I’m developing with Sven Leyffer.
Link to the program here.

Abstract:
Iterative methods for nonlinear optimization usually share common ingredients, such as strategies to compute a descent direction or mechanisms that promote global convergence. Our new open-source framework for nonlinearly constrained optimization, UNO, offers a selection of off-the-shelf strategies that can be assembled at will. UNO thus unifies a variety of methods (e.g. trust-region filter SQP, line-search penalty Sℓ1QP, line-search filter interior point method, …) and interfaces with specialized solvers (BQPD, MA57) with no programming effort from the user. UNO also provides an interface to the algebraic modeling language AMPL. We present extensive results on a subset of problems from the CUTEst collection, and compare UNO against state-of-the-art solvers filterSQP, CONOPT, IPOPT, KNITRO, LANCELOT, LOQO, MINOS and SNOPT.

I submitted my new article A global method for mixed categorical optimization with catalogs to the European Journal of Operational Research.
A preprint is available here.

Abstract:
In this article, we propose an algorithmic framework for globally solving mixed problems with continuous variables and categorical variables whose properties are available from a catalog. It supports catalogs of arbitrary size and properties of arbitrary dimension, and does not require any modeling effort from the user.
Our tree search approach, similar to spatial branch and bound methods, performs an exhaustive exploration of the range of the properties of the categorical variables ; branching, constraint programming and catalog lookup phases alternate to discard inconsistent values. A novel catalog-based contractor guarantees consistency between the categorical properties and the existing catalog items. This results in an intuitive generic approach that is exact and easy to implement.
We demonstrate the validity of the approach on a numerical example in which a categorical variable is described by a two-dimensional property space.

The slides of my talk An Augmented Lagrangian Filter Method yesterday at the SIAM CSE21 conference can be found here.
Abstract:
We introduce a filter mechanism to enforce convergence for augmented Lagrangian methods for nonlinear programming. In contrast to traditional augmented Lagrangian methods, our approach does not require the use of forcing sequences that drive the first-order error to zero. Instead, we employ a filter to drive the optimality measures to zero. Our algorithm is flexible in the sense that it allows for equality-constrained quadratic programming steps to accelerate local convergence. We also include a feasibility restoration phase that allows fast detection of infeasible problems. We provide a convergence proof that shows that our algorithm converges to first-order stationary points. We provide preliminary numerical results that demonstrate the effectiveness of our proposed method.

Our paper Two-phase approaches to optimal model-based design of experiments: how many experiments and which ones? was published online as open access article.

I will present our work An Augmented Lagrangian Filter Method (joint work with Sven Leyffer) at the SIAM CSE21 conference on March 1 at 10:10 (local time)/16:10 (UTC+1). I will also chair the session on Methods for Constrained Optimization in CSE starting at 9:45 (local time)/15:45 (UTC+1).
More information here.

Our article Two-phase approaches to optimal model-based design of experiments: how many experiments and which ones? has been accepted for publication in the Computers and Chemical Engineering journal.
It was cowritten with Philipp Seufert, Jan Schwientek (Fraunhofer ITWM), Gleb Karpov, Gleb Ryzhakov, Ivan Oseledets (Skoltech), Norbert Asprion (BASF SE) and Michael Bortz (ITWM).

ResearchGate link here.
ScienceDirect link here.

Thorsten Koch and I are giving a lecture at the Technische Universität Berlin called Industrial Data Science. It covers topics such as data visualization, probability and statistics, optimization and machine learning. The 15 sessions take place from Nov 2020 to Feb 2021.

I have the pleasure to start a new contract as a postdoctoral researcher at the Technische Universität Berlin (TUB), under the supervision of Thorsten Koch. I joined the UNSEEN project dedicated to the optimization of energy model systems using a combination of Mixed Integer Programming and Machine Learning techniques. It is a cooperation between the TUB, the Zuse Institut Berlin (ZIB), Jülich Supercomputing Center, GAMS and the DLR. I will be responsible for the project on the TUB/ZIB side.

The paper An Augmented Lagrangian Filter Method that Sven Leyffer and I wrote was accepted in the Mathematical Methods of Operations Research (MMOR) journal. Find it online on ResearchGate.

The slides of my talk Argonot: An Open-Source Software Framework for Nonlinear Optimization at ISMP 2018 are available on ResearchGate:

Iterative methods for nonlinear optimization usually share common ingredients, such as strategies to compute a descent direction or mechanisms that promote global convergence. Our new open-source framework for nonlinearly constrained optimization, Argonot, offers a selection of off-the-shelf strategies that can be assembled at will. Argonot thus implements a variety of methods (e.g. trust-region filter SQP, line-search penalty Sl1QP, …) and interfaces with specialized solvers (BQPD, MA57) with no programming effort from the user. Argonot also provides an interface to the algebraic modeling language AMPL. We present extensive results on a subset of problems from the CUTEst collection, and compare Argonot against state-of-the-art solvers CONOPT, IPOPT, KNITRO, LANCELOT, LOQO, MINOS and SNOPT.