📈 #73 Optimizing with intelligence: integrating Deep Learning into Operations Research

Leveraging complementary strengths to drive innovation

2020.

Covid wasn’t here (yet), and on Saturday mornings, I enrolled in a free Reinforcement Learning course.

We were supposed to learn the basics for 6 weeks, then spend 6 weeks building something with that knowledge. But since we were locked in our houses, it took months to build it.

The project name? SAMURAI.

Did it involve Japan? Not at all.

The name came from SAMUR (Servicio de Asistencia Municipal de Urgencia y Rescate, or Service of Municipal Assistance of Urgencies and Rescues) and AI.

We created a simulation engine that considered traffic and incident predictions, so an agent learned the right policy to manage all the ambulances using Deep Reinforcement Learning.

Putting ‘Deep’ before Reinforcement Learning forces you to use neural networks, and it was the first time I realized that Machine Learning could do more than classify cats and dogs: it could make decisions in complex, dynamic environments.

And that decision-making power, when combined with optimization, could unlock possibilities far beyond what either field could achieve alone.

SAMURAI was my first glimpse into how neural networks could tackle complex decision-making problems traditionally solved with optimization algorithms. It wasn't perfect (what first project is?), but it showed me the potential of combining these approaches.

So today in Feasible, I’ll discuss the combination of Deep Learning and Operations Research:

• 🧠 Why combine DL and OR?

• 🔗 3 ways to combine Deep Learning and OR

• 🔍 Challenges, limitations, and future directions

I couldn’t send you this post yesterday. Before I could finalize and schedule it, we had a massive blackout in Spain and Portugal. The bad thing is the chaos. The good thing is the home conversations. I’ll talk about them on Thursday :-)

Today is the last free long-form article in Feasible. Details are here.

Are you ready? Let’s dive in… 🪂

🧠 Why combine DL and OR?

The integration of Deep Learning and Operations Research creates a powerful symbiosis where each field addresses the other’s limitations.

This combination offers several advantages.

First, let’s compare their problem-solving strengths: