Icosa Computing: Shaping the Future of Discrete Optimization for Finance Inspired by Quantum Computing
In the world of optimization problems, the quest for the best solution often feels like chasing a mirage. With an astronomical number of possible solutions, the objective shifts instead towards finding an ‘optimal’ solution.
Discrete optimization focuses on optimization problems with many individual variables, such as whether to include a specific stock in your portfolio. Classic solvers have dealt with these for years and offered solutions that worked but still left room for improvement.
The relentless search for better solvers has led optimization enthusiasts down many intriguing avenues. One of these is inspired by quantum physics, a field that inherently deals with quanta—individual, discrete portions of a physical quantity. In 2022, Mert Esencan founded Icosa Computing, leveraging inspiration from quantum physics to develop better solvers for discrete optimization problems.
Learn more about the future of discrete optimization for finance inspired by quantum computing from our interview with the founder and CEO, Mert Esencan:
Why Did You Start Icosa Computing?
When I was young, I read Max Tegmark’s Our Mathematical Universe and learned about David Deutsch’s work and the multiverse, the idea that our universe could be one of many parallel universes. I continued reading about quantum physics to test the multiverse hypothesis and got hooked. This stuff was fascinating!
This prompted me to do a double degree in symbolic AI and physics in my undergraduate studies and management science and engineering in my master’s, during which I was one of the members who initiated the quantum computing association at Stanford. While interning at QC Ware, I fell in love with the applications of quantum physics in computing. I started pursuing a PhD in quantum computing and read The Fabric of Reality by David Deutsch. At the same time, I saw an important new industry growing, and I wanted to be part of it!
While quantum computing itself is still not mature enough for commercial applications, quantum-inspired classical algorithms certainly are. Playing with different quantum-inspired classical solvers, I figured that those worked actually better than established solvers for discrete optimization problems, especially for portfolio optimization in finance, and this led me to found Icosa Computing.
How Does Discrete Optimization Work?
Let’s say you’re trying to pack your bag for a day out at the beach, and you need to decide which items to take with you. For example, you would rather not get sunburned, so you should probably take sunscreen. But you also don’t want your bag to become too heavy, so given this constraint, you need to find the right set of items for the optimal beach experience.
It’s a discrete optimization problem where a decision depends on several discrete variables, and it’s usually impossible to optimize each simultaneously. Similarly, a stock portfolio is a collection of many individual stocks, and optimizing its performance is thus a discrete optimization problem.
Quantum physics is the theory of the fundamental behavior of matter and energy at the smallest scales, where the phenomena are described in terms of discrete, quantized units called quanta. Thus, it’s discrete by nature, and that’s the intuition behind why quantum computing can help solve discrete optimization problems more efficiently. We’re not building hardware but instead turning insight from quantum physics into code, a solver, to help optimize portfolios in finance.
One of the main advantages of our approach is that it scales to larger optimization problems compared to established classical solvers. They can typically handle only portfolios with a few thousand stocks, while we have optimized portfolios with over 25,000 equities beyond what’s possible with current methods. For comparison, circuit-based quantum computing may currently manage portfolios with a few dozen stocks, quantum annealing can handle maybe a few hundred stocks, and hybrid quantum-classical annealing a few thousand, maybe ten thousand stocks.
The great thing about finance and why we started there is that a lot of real data is already available, so we can quickly demonstrate the merits of our solvers and get to first revenue. From there, we can expand into other areas, such as tuning large language models, which rely on a large, discrete number of parameters.
How Did You Evaluate Your Startup Idea?
We participated in the Duality accelerator in Chicago and the Creative Destruction Lab in Toronto, which pushed us hard, especially on the business side, to set and reach milestones on customer development and to get to first revenue. Graduating successfully from the CDL program was conditioned on reaching those milestones, which we luckily managed to do.
Also, we went through a National Science Foundation program, where we were assigned two MBA students to assist with our customer development, such as calling customers directly and gathering insights in weekly presentations. We did this before building the product, and now that we have a prototype working, our main objective is finding product-market fit.
What Advice Would You Give Fellow Deep Tech Founders?
By definition, deep tech hasn’t been done before, so you, as the founder, know the most about what you’re doing. Of course, other people can bring in different expertise, but you need to put that into context as only you have the big picture. Listen to everyone, but take their advice with a grain of salt. Trust yourself more to be the person to make the right decision. And if you think it’s too much responsibility, try to bring in a co-founder. Building a business is hard, but it’s also fun!