Embedded Files
Thursday 26 June
Thursday 26 June
11:00-12:15 Registration, coffee/tea and networking
12:15-13:45 Lunch (DIY)
13:45-14:00 Registration
14:00-15:30 Presentations
14:00 Julian Hall
HiGHS: Introductions, review and outlook
14:30 Filippo Zanetti
Introducing the new HiGHS interior point solver
15:00 Oscar Dowson
15:30-16:00 Coffee/tea
16:00-17:10 Presentations
16:00 Luke Marshall
Accelerating Column Generation via Template Pricing
16:25 Mark Turner
Interesting aspects of MIP in HiGHS
16:50 Md Shahrukh Anjum
A cardinality-based extended formulation for the unsplittable multicommodity network design problem
18:30-21:00 Workshop dinner at Scotts Kitchen: Aperitif (18:30); Dinner (19:15).
Numbers at Scotts are limited to 30, so dinner for students will be at a venue to be arranged by local PhD students attending the workshop.
Due to late registrations and the number of true guest bookings being uncertain, we may not be able to offer places at the dinner to anyone registering after 20 June.
An after-dinner venue for all will be arranged.
Friday 27 June
Friday 27 June
09:30-11:05 Presentations
09:30 Tom Lauwers
highsv, A graphical front end to the HiGHS library
09:55 Sebastian Van Thienen
10:20 Renzo Wijngaarden
Optimising Airport Operations with HiGHS: Current Use and Future Potential
10:45 Ivet Galabova
11:00-11:30 Coffee/tea
11:30-12:15 Presentations
11:30 Qi Huangfu
Early history and recent observations of HiGHS
11:55 Andrew McGee
Optimizing Pet Food production with HiGHS
12:15-13:15 Lunch (DIY)
13:15-15:30 Presentations
13:15 Eligius Hendrix
On Augmented Lagrangian for LP Feasibility
13:40 Dimitris Kousoulidis
GB Battery Revenue Optimisation for Good, Fun, and Profit!
14:05 Harley Mackenzie
Back to Basics, Forward to HiGHS: juLinear.jl in the Age of Open-Source and AI
14:30 Kristin Braun
14:55 Christian Valente
15:20 Julian Hall
16:00 The Pear Tree
Abstracts
Abstracts
Kristin Braun
Accelerating Linear Programming Performance: A Hardware-Software Co-Design Approach for the Simplex Method
Linear programming plays a key role in mathematical optimization, forming the basis for solving a wide range of practical problems. Many real-world applications can be directly modeled as linear programs, while more complex problems, such as Mixed-Integer Programs (MIP) and Mixed-Integer Nonlinear Programs (MINLP), often rely on repeatedly solving linear relaxations. However, CPU-based methods face limitations when targeting extremely low computation times or substantial energy reductions, particularly in edge applications such as real-time robot control.
To address these challenges, we propose a dedicated hardware accelerator for the Simplex algorithm, developed using a hardware-software co-design approach. Specifically, we introduce the Simplex Processing Unit (SXPU), an accelerator for the computationally expensive pricing step in the Simplex method. Our design seamlessly integrates with the open-source HiGHS solver, accelerating the solution of large-scale LP relaxations in MIP and MINLP contexts. The SXPU is primarily designed for ASIC implementation to maximize performance and energy efficiency but also supports FPGA-based prototyping. It achieves significant speedups over software-based approaches, demonstrating its potential for high-performance, energy-efficient optimization in mathematical programming solvers.
Oscar Dowson
JuMP + HiGHS
JuMP is an algebraic modeling language for mathematical optimization written in the Julia programming language. We have been an enthusiastic adopter of HiGHS since early 2021. In this talk we will discuss how we adopted HiGHS and who is using it in the Julia ecosystem.
Julian Hall
HiGHS: Introductions, review and outlook
The HiGHS team has grown considerably over the past 12 months, so this presentation will introduce the new members of the team. Developments in HiGHS over the past year and future plans will be summarised.
Julian Hall
PDLP: Navigating the hype
GPU accelerated solution of LPs using Chambolle and Pock's primal-dual hybrid gradient technique (PFLP) is undoubtedly exciting, but it's important not to consider it as a silver bullet. It is less robust and can be very much slower and less accurate than traditional techniques. That said, some HiGHS users have seen big performance gains! This presentation will explore the pros and cons of the technique.
Dimitris Kousoulidis
GB Battery Revenue Optimisation for Good, Fun, and Profit!
Utility-scale battery energy storage systems (BESS) offer a compelling solution to the problem of intraday renewable intermittency and are already being rapidly deployed and helping displace the most polluting sources of electricity (good!).
However, their development and construction are very capital intensive processes. This makes maximising the revenue produced by batteries crucial to their financial viability and the rate of expansion of the industry, while also offering a plethora of interesting problems to work on (fun and profit!)!
In this talk, I will focus on the problem of short-term (one day ahead until real time) revenue optimisation for BESS in Great Britain (GB). I will introduce the key markets - wholesale electricity, frequency response services, and balancing mechanism and reserve - and provide an overview of how we think about and tackle this problem now.
I will also discuss where we would like to get to in the future and highlight how open source mixed integer solvers, and HiGHS in particular, have been enabling us throughout this journey.
Tom Lauwers
highsv, A graphical front end to the HiGHS library
As part of my course on linear optimizations I got introduced to Lingo, but I found it frustrating due to its proprietary limitations. In response I began working on a free and open-source graphical tool oriented towards educational settings that makes solving linear problems more accessible.
Harley Mackenzie
Back to Basics, Forward to HiGHS: juLinear.jl in the Age of Open-Source and AI
This presentation introduces juLinear.jl, an open-source linear programming solver developed in Julia with the support of HARD software. The motivation for juLinear.jl arises from the need for a solver that not only functions, but also approachable and transparent, making it ideal for students, educators, and researchers who want to understand the inner workings of linear programming solvers from the ground up.
juLinear.jl’s clear and mathematically expressive Julia codebase allows new learners to engage directly with core solver concepts, experiment with algorithmic techniques, and build confidence before contributing to major projects like HiGHS. By demystifying solver internals, juLinear.jl helps foster the next generation of open-source solver contributors and provides a platform for rapid experimentation and teaching.
The talk will also explore how AI tools can assist with solver development, highlighting both their potential benefits, such as code generation and bug detection, and their limitations, including the need for expert oversight and mathematical rigor. By combining a student-friendly open-source codebase with new development tools, juLinear.jl aims to accelerate learning, innovation, and collaboration in the global optimization community.
Luke Marshall
Accelerating Column Generation via Template Pricing
Column Generation can be an effective technique for dealing with large-scale integer programming problems – however, it is known to suffer from convergence issues. There is much research on stabilization to avoid this issue, with various levels of success and implementation difficulty.
Using a simple example, I’ll introduce a new approach “Template Pricing” that converges orders of magnitude faster than the “standard”. It is easy to implement and, surprisingly, yields high-quality integer solutions as a side-effect. In the talk, I’ll give insights on why it works so well – and how it might be generalized to other problem
Andrew McGee
Optimizing Pet Food production with HiGHS
The manufacture of pet food has many steps and processes leading to a nonlinear model. This paper outlines the how Datacor has solved these programs using the HiGHS solver.
Md Shahrukh Anjum
A cardinality-based extended formulation for the unsplittable multicommodity network design problem
Given an underlying directed graph, the multicommodity capacitated fixed charge network
design problem (MCND) is concerned with the selection of optimal paths and flows for all
commodities in the network at a minimal cost, while respecting flow balance constraints at each
node and capacity constraints along each arc. In this research, we consider the unsplittable
multicommodity capacitated fixed charge network design problem (UMCND), a variant of the
MCND, wherein the flow of a commodity from its origin to its destination is restricted along a
single path in the network. Practical instances of the UMCND can be found in
telecommunication networks, single sourcing production-distribution problems, express
package delivery, etc. In this research, we develop a new cardinality-based extended
formulation for the UMCND, wherein we associate new variables based on the cardinality of
each arc, where arc-cardinality is defined as the maximum number of possible commodities
that can flow along each arc in the network. This extended formulation, which consists of
𝑂(𝑛𝑘!) variables and constraints where 𝑛 and 𝑘 denote the number of nodes and commodities,
respectively, is proven to be stronger than its original formulation, and furthermore, the
extended formulation can also be gainfully exploited to develop specialized valid inequalities,
which generalize the well-known Cover, (1, 𝑝)-Configuration inequalities, Flow Forcing
inequality and the Missed Commodity Cutset inequality. Computational results on benchmark
instances demonstrate substantial savings in the number of subproblems taken in a branch-and-bound
process taken to determine the optimal solution.
Sebastian Van Thienen
Airline Crew Scheduling
Crew scheduling in aviation involves assigning pilots and cabin crew to sequences of flights while respecting a wide range of legal, contractual, and operational constraints. Solving these large-scale optimization problems efficiently is essential for airline operations and cost control. In this talk, we explain how we model the movement and activities of pilots on a time-space graph, and how this gives rise to a mixed-integer linear programming problem. To solve the relaxation of this problem, we can leverage the Dantzig-Wolfe decomposition structure and apply the column generation algorithm. In this algorithm, we solve a sequence of linear programs, that are growing in size. We show how we can apply HiGHS to solve these master problems.
Renzo Wijngaarden
Optimising Airport Operations with HiGHS: Current Use and Future Potential
At Edinburgh Airport, HiGHS is currently used to support our security operation by forecasting the number of security lanes required at 5-minute intervals throughout the day. This model helps minimise passenger queue times while optimising for operational efficiency. In addition to sharing our experience with this application, I will also discuss a future opportunity: applying HiGHS to the stand allocation problem. This complex task involves minimising coaching operations and thus optimising pier service level, while accounting for constraints such as destination-based stand restrictions, aircraft size compatibility, and jet bridge preferences amongst many other things.
Filippo Zanetti
Introducing the new HiGHS interior point solver
In this talk, we introduce the new HiGHS interior point solver. Its main features and the choices taken during its development are discussed, highlighting the differences with the existing solver. Some current limitations and plans to improve the code are mentioned, as well as issues related to parallelisation, accuracy and extension to quadratic programming. We present some results coming from energy modelling problems which highlight the large performance gain that the new solver can achieve compared to the current one.
Page updated
Google Sites
Report abuse