Quantum Computing

Machine Learning is a Necessary Tool to Scale Quantum Computers


Quantum computing has been hailed as the next big thing in computing, with the potential to revolutionize many fields, including cryptography, drug discovery, and optimization problems.

And over the last decade, we’ve seen major progress in increasing the number of qubits, the basic building blocks of quantum computers.

But Quantum computers, despite their many promises, are not yet a mature technology. The error rate of qubit and inter-qubit operations is too high to allow today’s devices to be useful. In essence, what we have today is actually an engineering prototype of a quantum computer – useful only for physicists and engineers to learn how to build better quantum computers.

Over the last decade, we have made significant progress in increasing the number of qubits in quantum computers. However, despite these advances, quantum computers have experienced an error rate of around 0.25% per operation in the same time period. Even in Google’s “quantum supremacy” results from 2019, 99.8% of the results are noise. Experiments must be run 30,000,000 times to gather enough statistics to prove the results claimed. At such an error rate, even 100 qubits are unusable.

To build fault-tolerant quantum computers that can perform useful computations, the error rate must be reduced by at least an order of magnitude. This is a significant challenge, as errors can arise from a variety of sources, including qubit noise, control electronics and environmental factors.

One approach to reducing error rates is to use error correcting codes, which can detect and correct errors as they arise. However, these codes require a large number of qubits to be effective, which means reducing the error rate is still a critical challenge.

Ultimately, to scale quantum computers, we need to target fault-tolerant quantum computers, which would require hundreds of thousands of qubits to be effective. However, before we can achieve this, we must first reduce the error rate by at least an order of magnitude.

The reason we can’t reduce our error rate is because qubits are like snowflakes – each one is a little different (or their environment and/or control paths are a little different). To obtain a low error rate, we must measure dozens of parameters per qubit per operation and characterize in detail how much noise affects the qubit’s operation. This will allow us to understand in depth what is the root cause of the remaining errors.

Unfortunately, it requires a large number of highly qualified people to execute. That’s too much work – even for very large companies like IBM and Google, let alone smaller startups or academic labs.


The solution to this problem is the ML physicist. ML assistants will permeate most areas of human endeavor over the next decade, and experimental physics is no exception. And ML physicists can help us make rapid progress in reducing error rates in quantum computers.

ML Physicist is a narrow AI specifically designed to work alongside human physicists in development laboratories. By doing the drudgery of measuring parameters and analyzing data, it frees up human physicists to focus on higher-level tasks, such as tackling key problems in the next hardware iteration. This allowed for faster upgrades and eventually led to the development of useful quantum computers.

ML physicists use machine learning algorithms to identify patterns in data that human physicists might not immediately see. By providing new insights and perspectives, it can help guide human physicist efforts towards the most important areas of development. This collaboration between humans and AI can lead to breakthroughs that might not be possible with either alone.

Overall, ML Physicist is a powerful tool for accelerating the development of quantum computers. By automating certain tasks and providing new insights, it can help optimize the efforts of human physicists and ultimately lead to the creation of more useful and practical quantum computing systems.

In conclusion, machine learning is a necessary tool to scale quantum computers. We’ve made great progress in increasing the number of qubits over the last decade, but commensurate improvements in error rates have been lacking, due to the complexity of the devices in question. ML physicists can accelerate progress on this problem. And as ML technology continues to evolve, it is likely to play an increasingly important role in the development of useful quantum computers.

Shai Machnes is the CEO and co-founder of Qruise



Source link

Related Articles

Back to top button