Author Archives: Matthew Saltzman

Nominations are solicited for candidates for the position of director
on the Strategic Leadership Board (SLB). All positions are
for three-year terms (through December 31, 2026). Any interested
individual is eligible to be a director of the Foundation.

COIN-OR is evolving rapidly, and directors can have a significant
influence on the development of the Foundation. Candidates should be
actively interested in COIN-OR and committed to active participation
in Foundation business. For more information on the activities of and
expectations for directors, prospective candidates are encouraged to
contact any current directors or the Foundation Secretary
(secretary at coin-or.org).

Note that membership in COIN-OR is not required. Candidates with
expertise in the software industry or experience as a director for a
non-profit organization will be eagerly welcomed.

The Foundation encourages the nomination of any interested individuals,
particularly users of COIN-OR software, who have not previously
been active participants in COIN-OR projects. Diversity of viewpoints and
skills is both needed and appreciated.

For all director positions, a candidate may be nominated by any
interested individual. Self-nomination is encouraged. All nominations
must be seconded by a full member of the Foundation.

Nominations should be emailed to the Secretary (secretary at coin-
or.org) and will be accepted until January 10, 2024. Candidates must
agree to be nominated and will be contacted to provide a short
biography and statement to be distributed to the full members of the
Foundation prior to the election.

More information on the COIN-OR foundation can be found at
https://www.coin-or.org/about-the-foundation/.

Big congratulations to Sungho Shin, François Pacaud, and Mihai Anitescu for winning the 2023 COIN-OR Cup in a very competitive field with their work ExaModels (https://github.com/sshin23/ExaModels.jl) and MadNLP (https://github.com/MadNLP/MadNLP.jl).

Open-source software packages ExoModels and MadNLP represent a significant breakthrough in efficiently solving large-scale NLPs by leveraging the capabilities of modern GPUs. These packages achieve a remarkable performance enhancement in solving real-world nonlinear optimization problems at scale. A noteworthy example is the recent accomplishment in accelerating alternating current (AC) optimal power flow (OPF) problems by up to a factor of ten compared to state-of-the-art tools. 

We were delighted to celebrate the winning nomination at the INFORMS Annual meeting. Thanks to Jordan Jalving, Carl Laird, Ruth Misener, and Calvin Tsay for serving on the committee.

Congratulations to Francesco Ceccon, Joshua Haddad, Jordan Jalving, Carl D Laird, Ruth Misener, Alexander Thebelt, and Calvin Tsay for winning the 2022 COIN-OR Cup for their submission OMLT (https://github.com/cog-imperial/OMLT). 

We didn’t *only* select OMLT because it makes us hungry for a yummy 🍳 omelet 🍳. OMLT represents neural networks and gradient-boosted trees in an algebraic modeling language (Pyomo) suitable for integrating into optimization formulations. Optimizing over trained surrogate models is important because it allows integration of NNs or GBTs into larger decision-making problems. Computer science applications include maximizing a neural acquisition function or verifying neural networks. In engineering applications, machine learning models may replace complicated constraints or serve as surrogates in larger design and operations problems. OMLT supports GBTs through an ONNX interface and NNs through both ONNX and Keras interfaces. OMLT transforms these pre-trained machine learning models into the algebraic modeling language Pyomo to encode the optimization formulations.

But also, the association with our tummies did help the winners, who we hope will make us breakfast.

The 2020 COIN-OR Cup was awarded at the foundation’s annual meeting, which took place online on November 16, 2020. The winner of the COIN-OR Cup 2020 is:

“Muriqui Optimizer: An open source solver for convex Mixed Integer Nonlinear Programming”

Congratulations to the Muriqui team: Wendel Melo, Marcia Fampa and Fernanda Raup.

Muriqui Optimizer is an open-source solver for convex Mixed-Integer Nonlinear Programming (MINLP) problems. Its development began about 10 years ago with the Master dissertation of Wendel Melo and it is still actively developed today. Muriqui implements a vast number of algorithms from literature (+70,000 lines of code), easing the development of customized algorithms for specific applications. It also provides interfaces to other COIN-OR projects, such as CBC and IPOPT.

In the COIN-OR spirit of reproducible research, new features and computational experiments with this open source-solver were discussed in five papers published in recognized scientific journals.

COIN-OR Cup committee members: Haroldo G. Santos and Samuel S. Brito

The 2019 COIN-OR Cup celebration took place at the INFORMS Annual Meeting in Seattle on Monday, October 21. The selection committee was co-chaired by last year’s winners, Andreas Lundell and Jan Kronqvist. The 2019 winners are Samuel S. Brito and Haroldo G. Santos of Universidade Federal de Ouro Preto, Brazil. The citation appears below.

Congratulations to the winners, and thanks for all you do for COIN-OR.

COIN-OR CUP 2019
The annual COIN-OR Cup Prize recognizes and celebrates effective uses of COIN-OR software or valuable contributions to COIN-OR.

As COIN-OR already consists of over 60 open source projects, it is important to not just recognize new projects, but significant improvements to already existing ones as well.

Mixed-Integer Linear programming (MILP) plays a central role in operations research (OR), with a wide variety of applications. The commercial solvers have impressive performances, but to researchers in OR they mainly remain closed. Furthermore, there are several applications where the licenses of the commercial solvers render them economically infeasible. There is, thus, a strong motivation for an efficient open-source MILP solver, and this year’s winner has made a strong contribution to improve the main open-source alternative.
The project captures the spirit of the COIN-OR initiative, and the importance of their work, and the solver, is also highlighted by the fact that it was used in most of the other submissions this year.

With this motivation, the COIN-OR Cup 2019 has been awarded to the submission

Using Conflict Graphs in COIN-OR Branch-and-Cut Solver

submitted by

Samuel S. Brito and Haroldo G. Santos
Universidade Federal de Ouro Preto, Brazil

The 2017 COIN-OR Cup is awarded to John Chinneck, Mubashsharul
Shafique, and Laurence Smith for their work on the CCGO Global
Optimizer.

In a series of papers, this team developed an improved multi-start
method that can be used to find solutions for nonconvex NLPs. The
methodology is based on an approach called constraint consensus
concentration that attempts to identify disjoint parts of the feasible
region, and then launches a nonlinear solver for each of the disjoint
regions. CCGO relies on a mixture of novel theoretical developments,
heuristics, and algorithm engineering to create a software that
effectively and reliably finds solutions to difficult nonconvex
problems, comparing favorably to existing solvers. The implementation
of CCGO relies on COIN-OR software.

One of the main goals of the COIN-OR Cup is to promote effective use
of COIN-OR software that the community may not know about. The series
of papers nominated by John, Mubashsharul and Laurence is a perfect
example of that, and the prize committee is glad to announce them as
this year’s Cup winners.

Committee members: Andy Conn, Giacomo Nannicini, Thomas Wortmann.

The 2016 COIN-OR Cup is awarded to Giacomo Nannicini (IBM T. J. Watson Research Center) and Thomas Wortmann (Singapore University of Technology and Design) for their contributions to model-based optimization for architectural design. The twelfth annual cup was awarded based on the RBFOpt and Opossum packages developed by the awardees.

RBFOpt has been in COIN-OR since 2015 and is a python library for blackbox global optimization in settings where the objective function is computationally expensive to evaluate. Opossum is a plugin for Grasshopper — a generative design language popular among architectural designers — that provides a GUI to RBFOpt.

This year’s committee commended the awardees for their simultaneous valuable contribution to COIN-OR and effective use of COIN-OR software. By lowering the barrier to entry, RBFOpt and Opossum are bringing state-of-the-art optimization to new communities and exciting classes of problems.

This year’s Cup was awarded at a celebration at the Flying Saucer in Nashville, Tennessee, during the INFORMS Annual Meeting.

2016 COIN-OR Cup Committee:

• Joey Huchette, Massachusetts Institute of Technology
• Jeffrey Linderoth, University of Wisconsin-Madison
• Miles Lubin, Massachusetts Institute of Technology
• Stefan Wild (chair), Argonne National Laboratory