Scientists at Nvidia and from a variety of academic, scientific, and quantum computing institutions published a research paper late last year about the ongoing convergence of AI and quantum computing and how each will play a key role in the development of the other.
“There is enough intuition to motivate us to explore AI as a breakthrough tool. [quantum computing]” researchers wrote in a paper published in 2016. nature communications In December. “The inherent nonlinear complexity of quantum mechanical systems makes them well suited for the high-dimensional pattern recognition capabilities and inherent scalability of existing and emerging AI technologies.”
This AI-quantum partnership is a core part of large-scale classical-quantum hybrid data centers that will become the standard operating environment as quantum computing matures. This manifests itself in a myriad of ways, including classical AI, where quantum systems handle cleaning, encoding, and optimizing data while performing computations.
An important area is the use of AI to solve error correction challenges in quantum computing. Qubits are notoriously fragile and can inherently fall apart, or decoheal, due to environmental factors around them, such as noise, temperature, light, or other qubit activity. This can lead to errors, and error-prone quantum systems are neither practical nor useful.
AI for QEC
A lot of research is being done to see how AI (both generative and agentic) can address error correction in quantum systems. For example, scientists at the University of Pennsylvania and Hao Tan from Peking University in China addressed this issue in a research paper early last year. Nvidia discontinued Ising, a family of open source AI models, in April that the vendor said can deliver up to 2.5x faster performance and 3x higher accuracy in the quantum error correction decoding process. “With Ising, AI becomes the control plane, or operating system, of quantum machines, transforming fragile qubits into scalable and reliable quantum GPU systems,” said Jensen Huang, co-founder and CEO of Nvidia.
That same month, researchers from Amazon Web Services, Quantum Elements, the University of Southern California, and Harvard University wrote about using AI digital twin technology and cloud-based HPC systems to create and run simulation models to advance research in quantum error correction.
In a separate paper in June, IBM researchers outlined a large-scale language model (LLM)-based framework that can classify thousands of code variations used to solve error correction problems to find the appropriate code for a particular situation. They write that this is “an example of the increasing two-way interaction between quantum computing and classical AI, each beginning to inform and accelerate the other.”
This month, a group of researchers led by Google Quantum AI researcher Volodymyr Sivak said that one of AI’s capabilities, the ability to learn from experience, could allow quantum systems to run for long periods of time, days, weeks, or even months, without having to stop computations to adjust calibration when errors are detected.
Reinforcement learning (RL) techniques are explained in detail in a study published by researchers. natureAs shown below, this is done through an AI system that can learn from the errors and adjust the quantum system’s control settings and operating parameters based on what is detected, using the encoded information that the quantum computer already collects as it monitors for errors.
Fault-tolerant quantum error correction relies on errors that leave a detectable signature indicating that an error occurred in a specific region of the circuit. The decoder examines the detected events and infers possible error patterns. After the decoder corrects the logic state, the remaining error is quantified using the logic error rate (LER). According to the researchers, this is the main way to measure the quality of quantum error rate processes.
Reinforcement learning has also been successfully used in other areas of complex computing, from robotics to improving the behavior of LLMs, the researchers noted.
Keeping quantum computers running
In this case, the method addresses the challenges faced by quantum computers regarding errors. When an error is detected, the system must be periodically stopped and readjusted to interrupt ongoing calculations. This could hinder the ability to perform the kinds of complex, long-running jobs that will be required as the industry moves into the era of practical commercial and fault-tolerant quantum computers.
Reinforcement learning essentially allows quantum computers to make the necessary adjustments to correct errors on the fly while continuing to perform calculations.
Researchers created the framework for Google’s Willow superconducting quantum chip (below), a hyperscaler to be introduced in December 2024, touting what was then a breakthrough in error correction.
They introduced hardware drift, in which the physical properties of the system, such as the qubit transition frequency and gate resonance parameters, gradually change over time, and used AI learning algorithms to continuously capture error data and make adjustments based on them.
The researchers found that using this technique, logic error rates were 3.5 times more stable and reduced by about 20% compared to traditional recalibration methods. At the same time, the performance of the system remained stable even while reinforcement learning was underway.
“This agent is able to reach high performance even starting from randomized initial control parameters, suggesting potential for augmenting or replacing elements of traditional calibration stacks,” the researchers wrote.
The group tested the framework with multiple surface and color error correction codes using an agent framework that can manage over 1,000 control parameters. In numerical simulations, the researchers found that the framework can manage as many as 40,000 control parameters.
Reinforcement learning frameworks may be enhanced in the future: “In the future, if the learning framework is enhanced enough, it is conceivable that quantum processors could be calibrated for QEC.” [quantum error correction] “RL is fully ab initio, without relying on traditional calibration paradigms or human experts. By allowing quantum computers to learn from errors, we unlock a scalable pathway to optimize performance in real time, replacing destructive calibration routines with non-disruptive computations,” the researchers wrote. Our research suggests that the path to fault tolerance is built not only on better hardware, but also on more intelligent control. ”
