Researchers at the Indian Institute of Technology Madras have developed a machine learning technique that can quickly identify the sources of noise in quantum computers. By training artificial neural networks on simulated data and testing them on IBM’s quantum processors, the team showed that it is possible to diagnose disturbances more accurately and design targeted ways to suppress them.
“We make use of artificial neural networks trained on well-designed synthetic data for rapid prediction of the noise features with minimal loss of accuracy,” Professor Siddharth Dhomkar, a co-author of the study, said in an email.
The promise and the problem of quantum computing
Quantum computers are often described as the next big leap in technology. Unlike ordinary computers that use bits — tiny switches that are either 0 or 1 — quantum computers use qubits, which can be in a mix of states at the same time. This ability gives them a huge advantage for certain kinds of calculations, from designing new materials to breaking codes that would stump even the fastest supercomputer.
The catch is that qubits are fragile. They depend on delicate quantum effects that can vanish with the slightest nudge from the outside world. As Dhomkar explains, “Anything that can interact with the qubits can potentially destroy quantum coherence (the degree of quantumness), which is essential for the operation of any quantum computer. These mostly uncontrollable interactions of the qubits with their immediate surroundings result in the so-called dephasing noise.”
Researchers have long searched for ways to protect qubits from interference. But the first step is figuring out exactly where the noise comes from — and that turns out to be tricky. The disturbances can change over time, and measuring them directly is slow and complicated.
“Deciphering the exact nature of these intricate interactions requires the implementation of time-consuming and complex quantum protocols,” says Dhomkar. Even when scientists try to measure the disturbances, they often end up with only an average picture that leaves out important details. As a result, many of the strategies for shielding qubits remain hard to put into practice.
A machine learning solution
To break this deadlock, Dhomkar and his colleagues turned to artificial intelligence. Their approach is inspired by the same techniques used in other fields, where computers learn to identify cats or faces by being shown thousands of examples. The researchers created a large set of simulated data showing how qubits are disturbed by their environment. Once the computer had “learned” these patterns, it was able to quickly spot the same signatures in real experimental data.
“The architecture of the neural network is derived from the models that are generally employed for image recognition and processing tasks,” Dhomkar explains. The payoff is speed. Instead of spending weeks running complex tests to figure out what is disturbing the qubits, the machine learning system can deliver an answer in a fraction of the time.
The team tested their method on IBM’s superconducting quantum processors. These devices use tiny electrical circuits cooled to near absolute zero, where they behave like qubits. Because electricity flows without resistance in this state, the circuits can create and maintain fragile quantum states long enough to be useful for computation.
“We use this methodology to characterize various IBM qubits to illustrate the time variation of the underlying noise and to construct customized sequences that, in principle, help in its suppression,” says Dhomkar.
The results were promising. By diagnosing the disturbances more quickly and precisely, the researchers could suggest ways to improve the performance of the qubits. “We have already implemented our protocol on IBM qubits, and the plan is to use this technique to benchmark and compare superconducting qubits being investigated in various labs, all over the world,” says Dhomkar. “This may provide valuable insights to improve fabrication strategies, thereby enhancing the quality of qubits.”
Beyond one type of qubit
Although this study focused on superconducting qubits, the team believes their method can be used for other designs as well. “The method that was first developed here is hardware agnostic, however, the current implementation was geared towards transmon qubits,” says Dhomkar. The key is to model the environmental disturbances for each qubit type. “We have already implemented a similar strategy to an optical spin system, and it can indeed be extended further.”
That flexibility means the approach could help push forward the entire field, which is still experimenting with many competing technologies.
The researchers are not stopping here. “We are now developing ways to tackle more complex noises,” says Dhomkar. In simpler terms, they are working on methods to deal with even more complicated and unpredictable types of disturbances. They are also exploring new forms of artificial intelligence to actively design better ways of controlling quantum computers.
“We are also exploring new AI methods that can design customized ways of running quantum operations more efficiently, even when the hardware is not perfect.”
The road to practical quantum computers will still be long, but this study shows a promising step forward. By teaching machines to recognize and counteract the hidden disturbances that plague qubits, researchers are finding new ways to bring the dream of quantum computing closer to reality.
Reference: Bhavesh Gupta et al, Expedited Noise Spectroscopy of Transmon Qubits, Advanced Quantum Technologies (2025). DOI: 10.1002/qute.202500109
Feature image credit: Gerd Altmann via Pixabay
