Don’t Forget About Simulation

When I was going through the Information and Systems Engineering program (think stats + software + consulting mindset) during my undergrad program, one of my favorite classes was Discrete Event Simulation. The premise of the class was to teach us how to use a combination of event distribution modeling with simulation software to solve real world business problems. It was awesome.

We used the Rockwell Arena simulation software to build these animated models of specific events based on hypothesized or observed poisson distributions. I thought it opened up my eyes to how the business world must use simulation extensively to solve their issues.

The real world never quite works as a junior in college expects it to.

As life would have it, there’s been a huge interest in quantitative methods to solve problems, but those methods have focused heavily on machine learning methods. The world of simulation, at least in the larger industry, has shrunk.

But I’m noticing an interesting phenomenon right now. Simulation is making a slight comeback, at least in my world.

Reinforcement learning seems to rely quite extensively on simulation. The AlphaGo champ that Google built heavily relied on a version of Monte Carlo Tree Search that could simulate what would happen in the games it played based on expected moves it would take. Also, based on my understanding, in the world of robotics, simulation is used extensively to train robots on a variety of tasks.

Finally, in my day job, we rely, off-and-on at least, on a simulation engine that can analyze and backtest different investment strategies (strategies that rely on the output of our DL models). In addition, we’re using a simulation engine to conduct research into our own reinforcement learning agent.

I fully understand that this is all anecdotal evidence, and that simulations have their own sets of issues (i.e. how do you model the real world, do you need data distributions before you start your simulations in order to accurately model the real world, etc). But when it comes to my point of view, I’m still a big fan of simulation as a quantitative tool to answer business problems, and I’m excited to see how the world of simulation progresses in the future.

If you’re interested in learning more about how we use simulations at Apteo, feel free to get in touch: info@apteo.co