US, WASHINGTON (NEWS OBSERVATORY) — Google has released a library of programs designed to work with quantum neural networks. It can be used both on real quantum computers and on their simulations. The product is free and open source.
The new library is called TensorFlow Quantum (TFQ). It complements the well-known TensorFlow tool, designed to work with artificial intelligence on classic computers.
The project was implemented by the Google AI Quantum team in conjunction with students of the University of Waterloo, as well as Alphabet X and Volkswagen.
Recall that a qubit, in contrast to the classical bit, can be not only in the “0” or “1” state, but also in a quantum superposition (“mixture”) of these states. Potentially, this provides quantum computers with tremendous processing power. But at the moment, qubit systems are very vulnerable to interference and errors .
TFQ provides tools for working with the main components of quantum computing: qubits, quantum measurement procedure and so on. The system can be used to program both real quantum computers and their simulations on classic machines. To solve the latter problem, the Google team also released the qsim quantum computing circuit simulator.
As noted experts of the company, the world is entering an era of noisy quantum processors medium scale (Noisy Intermediate-Scale Quantum, or NISQ ). Such devices will have 50-100 qubits (this is the “average scale”). It is assumed that they will already be able to something that is inaccessible to classic computers . At least with the help of the latter, it is already quite difficult to model systems with so many qubits.
On the other hand, qubits are still quite unstable (therefore, processors are called “noisy”). At the same time, the error correction algorithms proposed today will be effective with processing power of millions, not tens of qubits.
Thus, the capabilities of quantum computers of the NISQ era are limited by their vulnerability to errors. For greater efficiency, they should work in the same system with classic processors. And TensorFlow Quantum provides all the possibilities for this, since it is a “descendant” of TensorFlow, which is well adapted for integrating processors with different sets of instructions.
In addition, TFQ is integrated with the Cirq platform . This is also the development of Google AI Quantum, designed to work with quantum computers.
Artificial Intelligence in the service of quantum computing
Recall that NISQ-era systems are vulnerable to interference and computational errors. How to protect yourself from this evil?
Google engineers offer their solution. In fact, they say, we have a widespread task: to filter out random noise from the data and extract useful information. Such tasks often arise, for example, in pattern recognition, and are solved with the help of artificial intelligence.
A quantum neural network reads quantum data thanks to the very measurement procedure. The result of this procedure is already quite ordinary sets of numbers. Their processing can be provided with a classical (non-quantum) neural network with deep learning .
According to the developers, a pair of classical and quantum neural networks will extract the maximum information from “spoiled” quantum data.
Details for specialists are set forth in a preprint of a scientific article published on arXiv.org.
Contact us: [email protected]
Article is written and prepared by our foreign editors from different countries around the world – material edited and published by News Observatory staff in our US newsroom.