Sandia Labs uses machine learning to improve quantum computing

Daniel Hothem, Timothy Proctor, and their Sandia team have developed a machine learning technique that improves the efficiency of quantum computing systems. This approach focuses on modeling the behavior of quantum computers to better understand the system’s capabilities and limitations, as well as diagnosing problems within complex systems without direct access. The team aims to predict potential failures caused by physical errors, the main cause of quantum computer malfunctions, thereby accelerating development across the Department of Energy. This method aims to bridge the gap between theoretical possibilities and current performance of quantum systems by learning from program successes and failures.

Understanding errors in quantum computers

Quantum computers are prone to errors that destroy calculations, creating large gaps between theoretical potential and actual performance. These “physical errors” are a major cause of failure, making system improvements difficult and expensive. Unlike classical computers, the internal workings of quantum computers cannot be easily inspected during calculations, which hinders the ability to identify and correct these errors. To accelerate development, predictive tools that identify possible errors are essential.

To predict these failures, Sandia’s team is building a machine learning model that resembles the evaluation of an old jukebox. By analyzing a digital “snapshot” of a quantum program and its success/failure rates, the model infers the likelihood of physical errors. This approach avoids directly examining the complex quantum computer itself. Similar to jukebox diagnostics, the model learns to identify “flaws” and assess internal components to predict how often a program will succeed.

This modeling approach is designed to scale efficiently as quantum computers become more complex. While traditional error analysis techniques suffer from increased complexity, Sandia’s model does not face the same limitations. This scalability benefits programmers, engineers, and researchers, helping them improve their devices, understand program limitations, and focus on fruitful research directions, ultimately reducing the cost and time to develop next-generation quantum systems.

Modeling quantum systems with machine learning

Sandia researchers are developing machine learning models that improve the efficiency of quantum computing by predicting failures. These models work similar to diagnosing an old jukebox, identifying potential errors before running the program. The team uses neural networks to analyze “digital snapshots” of quantum programs, predicting which physical errors will occur during calculations, and converting them into formulas that estimate success rates. This approach aims to understand and mitigate errors without having direct access to the inner workings of quantum computers.

A key innovation is the ability to scale the model as quantum computer complexity increases. As systems grow, traditional error analysis methods become impractical, but Sandia’s approach avoids this limitation. By focusing on the most important errors and learning from data (the “pictures” of successful and unsuccessful programs), the model helps manage complexity while maintaining accuracy. This scalability is critical to future-proof quantum computing development.

These models benefit the entire quantum computing ecosystem. Programmers can use these to quickly identify and fix errors, and engineers can improve device designs. Researchers in other fields, such as chemistry, can assess whether existing quantum systems can solve particular problems. The ultimate goal is to streamline research and accelerate the application of quantum computers to national security challenges by avoiding “fruitless research directions.”

Jukebox analogy for quantum behavior

Sandia’s team has developed a way to improve the efficiency of quantum computing using techniques similar to troubleshooting old jukeboxes. Similar to testing several recordings to predict whether a machine will work properly without opening it, the team will build a model to predict failures in quantum computers. These models learn from the successes and failures of quantum programs and identify potential physical errors before performing calculations, allowing researchers to avoid mistakes and accelerate development.

The core of this approach utilizes neural networks to process “snapshots” of quantum programs. These models analyze data to predict which physical errors will occur during calculations. By training the model on data from both successful and failed programs, the team can avoid requiring extensive access to the quantum computer itself. The model infers information about potential errors and predicts the program’s success rate, similar to finding “blemishes” in a record.

The “super-efficiency” of this method is key, as quantum computers do not become significantly more complex as they grow. While traditional error analysis techniques struggle to scale, Sandia’s model maintains manageable complexity. This allows programmers and engineers to quickly understand errors, improve devices, and determine which problems can realistically be solved with existing quantum systems, ultimately streamlining research and reducing costs.

Scaling models for future quantum computing

Sandia researchers are developing machine learning models to improve quantum computing by predicting errors. These models aim to understand how physical errors disrupt calculations in real-world quantum computers, which currently frequently fail. The researchers liken the process to diagnosing a problem with an old jukebox. This means you can identify issues like scratches on records or faulty wiring, even if you can’t open up the device and inspect it directly.

The research team uses a neural network that processes data from a quantum program to predict which errors will occur during calculations. By training these models on data from both successful and failed programs, researchers can infer potential errors from digital “snapshots” in the same way they identify flaws in records. This streamlines the research process by making it possible to predict the success rate of a program without running it on a physical quantum computer.

The main advantage of this approach is its scalability. Unlike traditional error analysis techniques, the Sandia model does not significantly increase in complexity as quantum computers grow in size. This will allow programmers and engineers to better understand and improve their devices, and will allow researchers to determine which problems can realistically be solved with today’s quantum systems, ultimately accelerating development and reducing wasted research efforts.