The Dynamic Trichotomy of Quantum Computing, Machine Learning, and Optimization

The Dynamic Trichotomy of Quantum Computing, Machine Learning, and Optimization

One thing I’ve noticed is that many people think of quantum, ML, and optimization as entirely separate entities – something we at Zapata refer to as the ‘false trichotomy’ — when, in fact, they are highly correlated.

Alejandro Perdomo-Ortiz, Associate Director of Quantum AI

One of the reasons my colleagues and I devote our careers to quantum computing is its ability to solve intractable problems — to perform tasks that are impossible for classical computers. Quantum computing holds the potential of revolutionizing the way we do computation and, thereby, impacting the world. 

One of the most revolutionary aspects is that quantum computers can tap into nature to exponentially increase computational power, which only became possible in the early 1900s with our understanding of quantum mechanics. The laws of quantum mechanics govern nature at the core of the atomistic level, and we can use them today to perform computations. It’s no wonder we named our company after an actual revolutionary! 

Optimization is one of quantum computing’s most impactful opportunities 

I want to emphasize that quantum computing is not meant to be used for any problem. For example, you wouldn’t use a quantum computer to multiply numbers – that’s an easy task for any classical computer. 

The opportunity to use quantum methods to solve optimization problems better than classical resources like your laptop, or even the most powerful high-performance computer (HPC), is one of the best opportunities for quantum advantage to manifest and tackle truly intractable problems that are foundationally optimization challenges, such as financial portfolio optimization, protein design, transportation routing, and many others. 

The term “optimization” is broad, so it’s not surprising that there are different adjectives used in the community. If you speak with any of the Fortune 500s, you will likely hear several iterations of the term. From my own experience working at NASA for more than five years, talking to the engineers meant hearing terms such as “discrete optimization,” “constrained optimization,” and “combinatorial optimization,” among others. What they all had in common was being NP-hard problems. 

We don’t have access to optimal solutions for these types of optimization problems yet, even with the best classical computing resources at our disposal. That’s why there is an opportunity for quantum computing to make a significant impact and why it’s so exciting to be in this industry right now. 

I am optimistic that quantum computing will deliver a real, measurable advantage over the best classical algorithms and computers from a practical and scientific standpoint. For example, if I’m talking to someone doing financial portfolio optimization with the most advanced classical tools, I’m confident that we will provide a quantum solution that is better than what they are doing. For more insight on this topic, I get into some of the science and thinking behind this claim in a paper I wrote with my colleague Javier Alcazar: “Enhancing Combinatorial Optimization with Quantum Generative Models.” 

Finding the sweet spot between quantum computing, machine learning, and optimization 

One thing I’ve noticed is that many people think of quantum, machine learning (ML), and optimization as entirely separate entities – something we at Zapata refer to as the “false trichotomy” — when, in fact, they are highly correlated. The challenge for us is to help customers find the sweet spot between them. 

The above challenge actually taps into my journey with quantum computing. The way I would describe my journey with optimization problems and quantum computing is love/hate, and now I’m in the phase of love again. 

When I started working on quantum computation, my first project was related to quantum optimization, specifically, finding the lowest configuration of a protein. It was a problem that in biophysics is known as the “lattice protein folding problem.” The first question I had to answer was: how can we map this problem to a quantum device? 

The quantum annealing approach to quantum computing was the prominent technology then (though today, there are several more alternatives). Unfortunately for me, it was a tremendous battle to get that biophysical problem into language that could be run on one of these annealing machines. And it was a great challenge since several researchers and colleagues were trying for a while with no success, and it turned out to be an exciting and fun first Ph.D. project! 

Later in my career, at NASA, I found by speaking with many engineers and scientists that there was a constant need for solving combinatorial optimization problems. 

Around 2014, ML was growing rapidly as a technology solution. And, at same time, my optimism for finding a quantum advantage in optimization was decaying almost as rapidly. 

Fortunately for me – and my future career with Zapata — I saw an opportunity while leading the Quantum Machine Learning (QML) efforts at NASA; here begins my love relationship with QML. 

Fast-forward to just a few years ago, I noticed a trend — completely new to ML – where some of the leaders in ML were considering it for solving combinatorial optimization. That is when I started paying very close attention. I realized then that I could combine what I knew about quantum optimization and then use some ML to boost it, so to speak. This led to another Zapata saying that is an internal twist on the “kill two birds with one stone” heuristic: “we could be feeding three birds with one scone here.” 

My colleagues and I feel that quantum-enhanced optimization (QEO) is one of the most exciting developments in the quantum industry. We are working with an algorithm that has the potential to deliver quantum advantage by beating the speed and performance records of state-of-the-art classical optimization algorithms. Personally, I see this algorithm that combines ML and quantum computing for tackling combinatorial optimization problems as one of the most exciting and groundbreaking outcomes to come out of Zapata.  

Stay tuned for more updates as our team is working collaboratively with enterprise organizations in multiple industries to harness this unique and powerful trichotomy to overcome complex business problems and deliver measurable value in the near-term.  

For more insights about quantum computing and optimization, please listen to “The Quantum Pod” podcast I did with my colleagues Yudong Cao and Ethan Hansen. 

Alejandro Perdomo-Ortiz
Zapata Author

Alejandro Perdomo-Ortiz , Ph.D.

Research Director, Quantum AI/ML