Charlie Vanaret

Uno 1.3.0 is out! Now features an interface to the LP solver #HiGHS and a trust-region filter Sequential Linear Programming preset 👉 github.com/cvanaret/Uno Available in #Julia/#JuMP via Uno_jll.jl @SvenLeyffer @UnoSolver #optimization #cpp #opensource #sqp #barrier

Open #PhD position in my #Discrete #Optimization Lab at @utn_nuremberg If you're into #bilevel, #robust, #discrete or #nonlinear #optimization , I'd love to see your application! For details, check the official job posting #disc-opt" target="_blank" rel="nofollow noopener">utn.de/en/career/oppo… Please spread the word! 😉

#Julia #JuMP @JuliaLanguage news! Uno, a next-gen solver for nonconvex optimization, can be installed automatically via Uno_jll.jl and used via AmplNLWriter.jl @UnoSolver unifies SQP and barrier methods + globalization flavors #optimization #numericaloptimization #opensource #cpp

@MLE_Alger @ictmc2024 Damn that's a great idea!!

Inspired to get creative with my poster for @ictmc2024 this week! I shared my work on bayesian generalised linear mixed models to optimise treatment dosage in early phase trials using Patient-reported Outcomes. #ICTMC2024 #clinicaltrials

@EugeneVinitsky To quote Mihai Anitescu during ISMP this year, “if you're not programming in Julia in 2024, you're wasting your time” 🥲

Fairly high score NeurIPS submission rejected because “simulator is written in C++ and people don’t know C++” is a fun new low

Uno 1.1.0 is out 🎉 It now features: - an interface to the linear solver #MUMPS - a new funnel method (performs slightly better than filter methods). Git repo 👉 github.com/cvanaret/Uno Stay tuned to follow the progress on the #Python bindings 🐍 #optimization #opensource #cpp

@leonardsaens @vaiter Thanks for the comparison between forward difference and complex step approximation! I loved the elegance of the latter when I read about it around 2016 while working at @irtSaintEx. Note: what you refer to as round-off error is actually subtractive cancellation.

@vaiter You can try this, and it's more accurate sciencedirect.com/science/articl…

Finite difference approximation is a numerical method to compute derivatives by discretization. The forward difference illustrated here is a first order approximation of the order of the stepsize. ljll.fr/frey/cours/UdC…

Joachim Dahl is giving the #COatWork talk "Linear Programming: Barrier and First Order Methods" at @ZuseInstitute #optimization #linearprogramming #barriermethods #firstordermethods

@SparseRunner giving the #COatWork talk "High Performance Computational Techniques for the Simplex Method" at @ZuseInstitute #optimization #simplex #linearprogramming

@mathOCb @SvenLeyffer

David Kiessling, Sven Leyffer, Charlie Vanaret: A Unified Funnel Restoration SQP Algorithm arxiv.org/abs/2409.09208 arxiv.org/pdf/2409.09208

@deguerre @AlgebraFact Thank you! I'll look into it

@cvanaret @AlgebraFact No idea how this was derived, but the way the "big symbolic maths systems" (e.g. Mathematica) do it is with a rational variant of the Remez Exchange Algorithm. en.wikipedia.org/wiki/Remez_alg…

Surprisingly accurate approximation for cosine from 500 AD.

@deguerre @AlgebraFact Thanks for the insight! Do you think that is how the formula was derived? I'd be interested in a systematic way of computing an accurate rational approximation of a function over an interval

@cvanaret @AlgebraFact Fit a quadratic polynomial p(x) to p(-π/2) = p(π/2) = 0, p(0) = 1, you get p(x) = (π^2 - 4x^2) / π^2. What's interesting is that (π^2 - 4x^2) / (π^2 + a x^2) also matches at those points for all a. Now find the value of a for which this function matches cos(x) best...

New preprint with David Kiessling and @SvenLeyffer We analyze funnel methods for globalizing nonlinearly constrained optimization, prove global convergence, and give extensive results for funnel and filter methods researchgate.net/publication/38… #optimization @UnoSolver #opensource #oss

After #ISMP2024, the @UnoSolver will be at #OR2024 next week on Thursday, Sept 5. Now highly competitive against #filterSQP, #IPOPT and co. Exciting new results of funnel method. Joint work with @SvenLeyffer and David Kiessling conftool.com/or2024/index.p… #optimization #opensource

@sp_monte_carlo Damn, that's cool!

Full house for Mark Turner's talk on PySCIPOpt-ML at #ISMP2024 github.com/Opt-Mucca/PySC… #ml #MachineLearning #scip @zuse

The Uno team will be at #ISMP2024 next week! 😎 @SvenLeyffer will give the keynote presentation “Unifying nonlinearly constrained optimization” on Jul 23, 3pm. I will give a talk about @UnoSolver on Jul 26, 9am #mathematicalprogramming #optimization #opensourcesoftware #oss #cpp

@keenanisalive I'm developing Uno, an open-source C++ solver that unifies nonconvex optimization: github.com/cvanaret/Uno It mimics filterSQP and IPOPT, and is also competitive against SNOPT, MINOS, etc. Header only and L-BFGS approximations are on my to-do list! @UnoSolver #optimization

Eigen has made life easy for doing sparse linear algebra in C++. Is there an equivalent (header only, FOSS, …) for optimization/nonlinear solvers in C++? As in Mathematica or MATLAB, I just want to quickly try out many methods (Newton, L-BFGS, various QP solvers, …whatever).