Pruning the expression tree with recursive value identification

Today was the release of SCIP.jl v0.11, the first release switching to SCIP 8. The major change in this (massive) release was the rewrite of the nonlinear optimization part, using a so-called expression framework. The rewrite of the wrapper had some fairly tedious parts, debugging C shared libraries is quickly a mess with cryptic error messages. But the nonlinear rewrite gave me the opportunity to tweak the way Julia expressions are passed to SCIP in a minor way.

Mutability, scope, and separation of concerns in library code

It has been about a year since I joined the Zuse Institute to work on optimization methods and computation. One of the key projects of the first half of 2021 has been on building up FrankWolfe.jl, a framework for nonlinear optimization in Julia using Frank-Wolfe methods. You can find a paper introducing the package here. This was an opportunity to experiment with different design choices for efficient, scalable, and flexible optimization tools while keeping the code simple to read and close to the algorithms.

Sets, chains and rules - part II

Differentiating set projections.

Sets, chains and rules - part I

The Pandora box from simple set membership.

Experiments on communicating vessels, constrained optimization and manifolds

Constrained optimization on a fundamental engineering problem

Differentiating the discrete: Automatic Differentiation meets Integer Optimization

What can automated gradient computations bring to mathematical optimizers, what does it take to compute?

Working with binary libraries for optimization in Julia

When going native is the only option, at least do it once and well.

Lessons learned on object constructors

Constructors are a basic building block of object-oriented programming (OOP). They expose ways to build specific types of objects consistently, using arbitrary rules to validate properties. Still, constructors are odd beasts in the OOP world. In Java, this is usually the first case of function overloading that learning programmers meet, often without knowing the term. An overloaded constructor is shown in the following example: class Car { private Motor motor; public Car(Motor m) { this.

Bridges as an extended dispatch system

Compiling mathematical optimization problems in a multiple-dispatch context.

Leveraging special graph shapes in LightGraphs

In a previous post, we pushed the boundaries of the LightGraphs.jl abstraction to see how conforming the algorithms are to the declared interface, noticing some implied assumptions that were not stated. This has led to the development of VertexSafeGraphs.jl and soon to some work on LightGraphs.jl itself. Another way to push the abstraction came out of the JuliaNantes workshop: leveraging some special structure of graphs to optimize some specific operations. A good parallel can be established be with the LinearAlgebra package from Julia Base, which defines special matrices such as Diagonal and Symmetric and Adjoint, implementing the AbstractMatrix interface but without storing all the entries.