These noises are often caused by interference from the environment, imperfect control signals, and unwanted interactions between qubits (which are the building blocks of a quantum computer . A quantum circuit, which is a series operations known as quantum gates, is required to perform computations on a quantum computing device. These quantum gates are mapped to individual qubits and change the quantum states of specific qubits. Then, the quantum gates perform calculations to solve a problem.
However, quantum gates introduce noise that can slow down a quantum machine’s performance.
Researchers from MIT and other universities are trying to solve this problem by creating a method that makes quantum circuits resilient to noise. These are quantum circuits with adjustable quantum gates that have been “parameterized”. The framework was developed by the team to identify the most robust quantum circuits for any particular computing task. It also generates a mapping pattern specific to the qubits in a targeted quantum device.
QuantumNAS (noise adaptive searching) is a framework that can find quantum circuits. It’s computationally less intensive than other search methods. Their method of identifying quantum circuits for real quantum device was more successful than other methods.
The key idea is that without this technique, it would be difficult to test each quantum circuit architecture or mapping scenario in the design space. We would then have to train them and evaluate them. If it doesn’t work, we can throw it away and start again. This method allows us to obtain multiple circuits and mapping strategies simultaneously without the need for much training,” Song Han, associate professor in the Department of Electrical Engineering and Computer Science, and senior author of this paper.
Hanrui Wang, the lead author, and Yujun Lin (both EECS graduate students) are joining Han for the paper. Yongshan Ding is an assistant professor at Yale University. David Z. Pan is the Silicon Laboratories Endowed chair in Electrical Engineering at Texas at Austin. Jiaqi Gu, UT Austin grad student, Jiaqi Gu, Fred Chong, the Seymour Goodman professor in the Department of Computer Science at Chicago, and Zirui LI, an undergraduate at Shanghai Jiao Tong University. The IEEE International Symposium on High-Performance Computer Architecture will present the research.
There are many design options
Parameterizing a quantum circuit requires you to select a few quantum gates. These are the physical operations that the qubits will perform. It is not an easy task as there are so many gates available. There can be many gates in a circuit. The positions and physical qubits of these gates can change.
The design space is huge because there are so many options. It is difficult to design circuit architectures that are good. QuantumNAS is designed to be extremely robust to noise.” Wang says.
Researchers focused on variational quantum systems, which are quantum gates that have trainable parameters and can learn machine learning or quantum chemical tasks. A researcher can either hand-design a variational quantum circuit or use rule-based design methods. Then, he/she will try to determine the optimal set of parameters for each gate by optimizing the process.
The naive search method in which all possible circuits are evaluated separately, requires that the parameters of each candidate quantum circuit be trained. This creates a huge computational overhead. The researcher must also identify the optimal number of parameters and the best circuit architecture.
Construction of a “SuperCircuit”.
They first create a “SuperCircuit,” which includes all possible parameterized quantum gate designs. This SuperCircuit can be used to create smaller quantum circuits that are easily tested.
The SuperCircuit is trained once and all the other circuits in the design area are subsets, so they inherit the same parameters that were previously trained. This helps reduce the computational overhead.
After the SuperCircuit is trained, they can use it to search circuit architectures that meet their targeted objectives, such as high robustness to noise. This involves simultaneously searching for qubit mappings and quantum circuits using an evolutionary search algorithm.
The algorithm generates a number of qubit mapping candidates and then assesses their accuracy using a noise model or a real machine. The algorithm then uses the results to select the most efficient parts and restarts the process until it finds the perfect candidates.
