Hybridization of Machine Learning and Operational Research is the future of AI

Recently, Graph Neural Network (GNN) gained a lot of attention, but did you know that Graph theory is actually considered by some as a subfield of Operations Research?

Even better, did you know that Machine Learning (ML) was considered as a subfield of Operations Research (OR) a few years ago without (much) controversy?

What is OR you might ask? Not an easy question but think about optimization. Every time you try to optimize something, you probably are in the realm of OR somehow.

ML as a subfield of OR? Basically, in ML you try to ... optimize your predictions! Beyond naming disputes, there are two fields, ML and OR, that are remarkably complementary and when you combine them together you can go (much) further than just using ML or just using OR.

There are at least 4 ways to combine ML and OR. You can use ML and OR as two black boxes without any interaction between the two.

Most of the time, you use ML to get accurate predictions in a first step and then in a second step, you optimize your problem with OR using those predictions as an input.

Then there are two technical ways to use them: use one to improve the other. It works in both directions and it really improves algorithms in both fields. Finally, there is a fourth way to combine both where new algorithms are created.

The first combination is becoming more and more popular and used since about 2019 to solve industrial problems.

The second and third combinations are essentially the subject of academic research by a few teams around the world.

The fourth combination is quite unknown but is probably the most interesting one as - in my view - it could lead to real intelligence.

In this talk, I'll briefly present what OR is and how it is related to ML. Then I'll present real and successful industrial cases for the first three combinations. I will conclude by briefly presenting what I believe can be done with the fourth combination.

Talk+Live Q&A at the Eastern Auditorium in Connected Data World Center

You need an access pass to attend this session: Diversity Access Pass or Full Access Pass apply