Researchers at the Max Planck Institute for Optics in Germany have proposed a quantum error correction system based on artificial intelligence algorithms. With sufficient training, this method is expected to surpass other error correction strategies.
Quantum computers can solve complex tasks that traditional computers cannot accomplish. However, quantum states are extremely sensitive to continuous interference from the outside world. Researchers hope to use active protection based on quantum error correction to solve this problem.
Recently, Florian Marquardt, director of the Max Planck Institute for Optics in Germany, and his team published a paper Reinforcement Learning with Neural Networks for Quantum Feedback in the physical journal physical review X, proposing a quantum error correction system based on artificial intelligence algorithms.
In 2016, the artificial intelligence program AlphaGo defeated the world's strongest human chess player in the Go game, attracting worldwide attention. Given that there are more combinations of moves in a game of Go than the estimated number of atoms in the universe, it requires more than just processing power. Instead, AlphaGo uses artificial neural networks, which can recognize visual patterns and even learn. Unlike humans, AlphaGo can practice playing hundreds of thousands of Go games in a short period of time, eventually surpassing the best human players.
Researchers at the Max Planck Institute are trying to use this neural network to develop error correction learning systems for quantum computers.
An artificial neural network is a computer program that simulates the behavior of interconnected nerve cells (neurons)-in this study, there are approximately 2,000 artificial neurons connected to each other.
"We take the latest ideas from computer science and apply them to physical systems," Florian Marquardt said. "In this way, we can benefit from the rapidly advancing field of AI."
Learning quantum error correction: visualization of artificial neuron activity when a neural network is performing tasks
Image source: Max Planck Institute of Optics
The main ideas of the research can be summarized as follows:
Artificial neural network may be able to surpass other error correction strategies
In the paper, the team proved that artificial neural networks can learn how to perform a task that is vital to the operation of future quantum computers-quantum error correction. Even with enough training, this method is expected to surpass other error correction strategies.
In order to understand how it works, you first need to understand how a quantum computer works. The basis of quantum information is quantum bit (quantum bit, or qubit). Different from traditional digital bits, qubits can not only adopt the two states of 0 and 1, but also the superposition of the two states.
In the processor of a quantum computer, there are even multiple qubits superimposed together as part of a joint state. This kind of entanglement gives quantum computers the powerful processing power to solve some complex tasks that traditional computers can't do.
However, quantum information is very sensitive to environmental noise. This characteristic of the quantum world means that quantum information needs to be corrected regularly-that is, quantum error correction. However, the operations required for quantum error correction are not only complicated, but also the integrity of the quantum information itself must be maintained.
Quantum error correction is like a game of Go with strange rules
When introducing the working principle of the research, Marquardt put forward an interesting analogy. He said: "You can imagine the elements of a quantum computer as a Go chess board, and qubits are distributed on the entire board like chess pieces." But, unlike the traditional one. Compared with the game of Go, there are some key differences: all the chess pieces have been arranged on the board, and each piece is white on one side and black on the other side. One color corresponds to 0 and the other color corresponds to 1, and the movement in the quantum game of Go is equivalent to turning the chess piece over. According to the rules of the quantum world, chess pieces can also be gray mixed with black and white-representing the superposition and entanglement of quantum states.
When playing this quantum Go game, the player—let us call her Alice—makes an action to preserve a pattern that represents a certain quantum state. This is the quantum error correction operation. At the same time, her opponent went to great lengths to destroy this model. This represents the continuous noise of excessive interference experienced by the actual qubit in its environment. In addition, the game of quantum Go is particularly difficult, because there is a special quantum rule: Alice is not allowed to see the board. A glimpse of any scene that can reveal the state of a qubit will destroy the current sensitive quantum state of the game.
The question is: With so many restrictions, how can she make the right move?
Auxiliary qubits reveal flaws in quantum computers
In quantum computers, this problem is solved by positioning additional qubits between the qubits that store the actual quantum information. Intermittent measurements can be taken to monitor the status of these auxiliary qubits, allowing the controller of the quantum computer to identify the location of the fault and perform error correction operations on the information-carrying qubits in these areas.
In the analogous quantum game of Go, the auxiliary qubits are represented by auxiliary pieces distributed among the actual game pieces. Alice can look at it occasionally, but only at these auxiliary pieces.
In this research, Alice's role is performed by an artificial neural network. The researchers' idea is that through training, the network will become very good at this role, and can even surpass the corrective strategies designed by humans.
However, when the team studied an example containing five analog qubits, they found that using only one artificial neural network was not enough. Since the network can only collect a small amount of information about the state of the qubit, it can never go beyond the strategy of random trial and error. In the end, these attempts destroy the quantum state instead of correcting it.
One neural network uses its prior knowledge to train another neural network
The solution is to add an additional neural network as the teacher of the first network. With its prior knowledge of quantum computers, the teacher network can train other networks—that is, its students—to guide the network to successfully perform quantum error correction. But first, the teacher network itself needs to fully understand quantum computers or quantum computer components that need to be controlled.
In principle, artificial neural networks use a reward system for training. For quantum error correction systems, to successfully restore the original quantum state, actual rewards are necessary.
"However, if rewards are given after achieving this long-term goal, it will need to try many error corrections, and it will take too long to achieve the goal," Marquardt explained.
Therefore, they developed a reward system that motivates the teacher's neural network to adopt effective strategies during the training phase. In the quantum game of Go, this reward system will provide Alice with the overall state of the game in a given period of time without revealing details.
Student networks can surpass teachers through their own actions
"Our first goal is to enable teacher network learning to successfully implement quantum error correction operations without human assistance," Marquardt said. Unlike student networks, teacher networks can do this not only based on measurement results, but also based on the overall quantum state of the computer. The student network trained by the teacher network will be just as good at first, but through your own behavior, you can get better.
In addition to error correction in quantum computers, Florian Marquardt also envisions other applications of artificial intelligence. In his view, physics provides many systems that can benefit from the pattern recognition of artificial neural networks.
All In One Business Computer is the one of the important All In One PC series, can meet all enter and normal business application scenarios. Therefore, more and more clients choose and recommend this Business All In One Computer for business or education projects. The Best All In One Computer For Business is the similar Apple design, competitive and attractive, the hottest parameters is 19 or 21.5 inch intel i3 or i5 2th or 3th or 4th 4 gb ram 128GB ssd . There are many other all In One Computers For Business with lighter weight. Of course, you can see Colorful All In One Gaming PC, All In One Desktop Touch Screen and All In One Business Pc. Believe you can always find the right one at this store. To save time, You can also contact us by email or calling directly to get matched and valuable information fast.
Any other special configuration interest, also feel free to let us know, will try our best to back you up.
Except all in one Custom All In One PC, 14 inch Education Laptop and 15.6 inch business laptop also available here.
All In One Business Computer,Business All In One Computer,Best All In One Computer For Business,all In One Computers For Business,All In One Business Pc
Henan Shuyi Electronics Co., Ltd. , https://www.shuyilaptop.com