Thursday, March 06, 2014

Video: Marguerite Frank - Inventor of the Frank-Wolfe Algorithm -

Zaid Harchaoui and Martin Jaggi let me know of this particular video from the Frank-Wolfe and Greedy Algorithms  NIPS 2013 Workshop " which your readers might find interesting". Thanks Zaid and Martin


***
Honorary discussion panel with Marguerite Frank:


Nearly 60 years ago, Marguerite Frank and Philip Wolfe published a very interesting paper, while at Princeton university. The 1956 paper introduced the famous Frank-Wolfe algorithm (also known as conditional gradient), and was the very first method for general constrained convex optimization. The importance of the method can hardly be emphasized enough, as it marks a historical tipping point enabling the departure from linear programming, going to more general convex optimization.

The paper is still of great significance today, as it paved the way for various currently important and broadly used optimization algorithms, in particular sparse methods in signal processing and machine learning (including matching pursuit, Lasso and related techniques).

In the honorary discussion panel at the NIPS 2013 workshop, Marguerite Frank gives a personal account of the history of the paper, as well as her inspiring biography as the first woman in a newly emerging field,
today called mathematical optimization.


Frank–Wolfe algorithm on wikipedia




2 comments:

Martin said...

Yes, very sad that publishers don't make articles publicly available even after so many years.

Along these lines, not even Marguerite Frank herself could access her own article, she recently asked me to email her a copy :/. Hope open access will gain more momentum soon...

Deeparnab said...

I wonder where Philip Wolfe is these days. I have read a transcript of his interview (http://invention.smithsonian.org/downloads/fa_cohc_tr_wolf721128.pdf) from way back -- it's a pretty interesting read.

Printfriendly