iiinova — where your digital self keeps evolving

Clustering is a powerful tool for decision-making under uncertainty! Work w/ my students Irina Wang (lead) and Cole Becker, in collab. w/ Bart Van Parys 🧵 (7/7)

010

We have several examples in the paper. Here is a sparse portfolio optimization one. Clustering barely affects the solution objective. Speedups are more than 3 orders of magnitude. 🧵 (6/7)

111

By varying the number of clusters K, our method bridges Robust and Distributionally Robust optimization! We also derive theoretical bounds on 1) how to adjust the Wasserstein ball radius to compensate for clustering, and 2) how to exactly quantify the effect of clustering 🧵 (5/7)

110

In Mean Robust Optimization, we define an uncertainty set around the cluster centroids with weights defined by the amount of samples in each cluster. 🧵 (4/7)

100

Our procedure: we first cluster N data points into K clusters. Then, we solve the Mean Robust Optimization problem. 🧵 (3/7)

100

Robust optimization is tractable but, often, very conservative. Wasserstein Distributionally Robust Optimization is less conservative but, often, computationally expensive. How can we bridge the two? 🧵 (2/7)

100

Bartolomeo Stellato @stella.to · 28/11/2024

Our paper "Mean robust optimization" has been accepted to Mathematical Programming: buff.ly/3B3VpIG 📰 Arxiv (longer version): buff.ly/3CT4aWD 👩‍💻 Code: buff.ly/3ATqAXh w/ Irina Wang, Cole Becker, and Bart van Parys A thread 🧵 (1/7)👇

1287

Cool! Thanks for creating this. Could you please add me? :)

000

Reposted by Bartolomeo Stellato

ArXiv math.OC Optimization and Control @optb0t.bsky.social · 26/11/2024

📚 New Arxiv Paper Title: Learning Algorithm Hyperparameters for Fast Parametric Convex Optimization Authors: Rajiv Sambharya, Bartolomeo Stellato Read more: arxiv.org/abs/2411.15717v1

022

Reposted by Bartolomeo Stellato

Thiago Serra @thserra.bsky.social · 21/11/2024

We are very excited to announce that the 2025 INFORMS Computing Society (ICS) Conference will take place March 14-16, 2025, in Toronto: sites.google.com/view/ics-2025 Submissions for contributed talks are due on December 23. We invite talks that showcase the dynamic interface of CS, AI & #ORMS.

sites.google.com

2025 ICS Conference

The 18th INFORMS Computing Society (ICS) Conference welcomes you to Toronto, Canada. We invite researchers, practitioners, and innovators to come together and share insights at the cutting edge where...

0164

Thanks @tmaehara.bsky.social It looks great! I will let you know if I find anything wrong but from a brief look at the first post it looks exactly what one would expect. Thanks again!

110

By the way, do you consider linear optimization a technology? (if use the 1-norm Mosek gives the correct answer)

100

Still, some robotics companies use ADMM-based solvers for MPC. If you need only low accuracy solutions and reoptimize very often, first-order solvers can be very effective (especially with warm-starting).

200

Bartolomeo Stellato @stella.to · 17/11/2024

You are right. For any convex solver, you can construct a nasty problem for which it fails. And if it is a first-order solver, it is quite easy to do it 🙂

100

Sounds great! Thanks a lot

000

Bartolomeo Stellato @stella.to · 17/11/2024

Hey @tmaehara.bsky.social Thanks for the great work on the arxiv bots here! Any chance you could make one for arxiv math.oc (Optimization and Control)? :)

2123

Reposted by Bartolomeo Stellato

Thiago Serra @thserra.bsky.social · 14/11/2024

If your favorite book about algorithms is not Algorithms for Toddlers, then you haven’t read this book yet. Today I used it to talk about greedy algorithms in my Applied Optimization #orms class (some pages below). Here is one of the authors reading the whole book: m.youtube.com/watch?v=nnLO...

2121

Reposted by Bartolomeo Stellato

Emily Tucker @emilyltucker.bsky.social · 08/11/2024

New starter pack if you're looking for the #orms community on here: go.bsky.app/SvBND16 Also if you have recommendations of people to add, feel free to send them my way.

154424

Bartolomeo Stellato @stella.to · 07/09/2024

Very proud of my first PhD student Rajiv Sambharya for defending his thesis! 🎉 Rajiv has done excellent work on learning optimization algorithms for large-scale and embedded optimization, with strong convergence and generalization guarantees. He will soon start a postdoc at UPenn Engineering!

031

Bartolomeo Stellato @stella.to · 08/08/2024

It was great to organize Princeton Workshop on #Optimization, #Learning, and #Control last June! Thanks to everyone who attended and made it a success! 🎉 #OLC24 Missed the live sessions? Catch up on all the talks with the video recordings here: buff.ly/3YwomX6

020

Also, congrats also to the co-winners who developed the PDLP solver! David Applegate, Mateo Díaz, Oliver Hinder, Haihao Lu, Miles Lubin, Brendan O'Donoghue, and Warren Schudy.

000

Interesting fact, Martin Beale generalized Dantzig's simplex method to solve quadratic programs (QPs) buff.ly/3YqZaRU.

100

A huge thanks to the many contributors who helped develop it and maintain it for these years. In particular Ian McInerney, Vineet Bansal, and Amit Solomon; also, to the support from the EU Marie Curie ITN project TEMPO, the Princeton Center for Statistics and Machine Learning, and Princeton OIT.

100

In particular, none of this would have been possible without Goran Banjac with whom and I shared countless hours developing OSQP. Here is a picture of us in 2016 having Korean BBQ in the Bay Area (where it all began!)

100

I am thrilled to accept this award on behalf of my co-authors Goran Banjac, Paul Goulart, Alberto Bemporad, and Stephen Boyd. Although they couldn't attend, this prize is for our great team work.

100