The macroscopic world governed by classical physics, as we perceive it, differs vastly from the microscopic world governed by quantum physics. Despite the great strides that have been made in measurement technology for quantum properties, many phenomena observed in the quantum many-body systems, where multiple degrees of freedom influence each other, cannot be addressed directly with current methods due to the complexity of the signals. This limitation is closely related to the fact that we humans are incapable of conceptually comprehending information that is too complex and multi-dimensional. However, even if we cannot understand it, these data carries information of the quantum systems. Thus, the use of rapidly developing AI will facilitate the deciphering of previously incomprehensible quantum properties, and thereby reduce them to something that can be used by humans.

This research aims to develop AI-physical science for reading and using the complex signals produced by quantum systems. Most classical science and technology deals with macroscopic averaging of fluctuations of signals obtained from quantum systems, rarely leveraging the vast microscopic degrees of freedom. However, the emergence of AI has enabled us to actively utilize some of the unused microscopic degrees of freedom and to transcribe it into information human beings can understand. This in turn has increased the possibility of establishing the next-generation processing utilizing various material properties such as quantum nature and nonlinearity, in other words, developing a completely new type of computing that implements functions such as recording and computing information using microscopic degrees of freedom.

In this research, we first aim to decipher and utilize complex quantum output signals called “quantum fingerprints” and “quantum bits (qubits).” AI scientists and researchers in the field of quantum physical property measurement collaborate and boldly integrate their measurement devices, contributing to applications such as quantum computing technology and quantum physical tags. In other words, it would allow for mapping the real world into a complex quantum world beyond the degrees of freedom that humans can comprehend, which in turn would help expand human perception, thinking, and science and technology itself through quantum AI.

Through our research to date, we have accumulated world-leading academic knowledge and expertise in the fields of quantum properties and quantum science and technology. We will accelerate the further progress of both areas by supplementing the respective strengths of our two research fields: advanced measurement technology in quantum physics experiments and information processing in machine learning. Our existing studies have demonstrated that the complexity of physical phenomena arising from microscopic degrees of freedom can be converted into usable information resources using the deep learning methods. We will further pursue our research to trigger a chain reaction of unconventional challenges and innovations.

Discovering a route between the quantum world and our real world through AI would transform science, allowing for the development of new quantum-based technologies. For example, if we can develop a method to ultimately determine the identity of substance, it could be applied to the real world as a “quantum physics tag” to identify products and many other things. In the future, the combination of quantum physics and AI will be likely to make innovation happen in multiple areas: the exploration of the field of quantum AI, in which the complexity and non-linearity of quantum are actively incorporated into AI; and the development of new quantum computing technology, in which the fluctuations of quantum are controlled to a high degree through the use of AI.

The research objective of this project is to extract information hidden in complex quantum properties and quantum fluctuations, which appear in microscopic physical systems governed by quantum mechanics, using artificial intelligence (AI).

Firstly, we aim at the AI-assisted analysis of complex fluctuations in electrical resistance that manifest in microscopic electrical conduction in metals. In the minuscule metals with the nanometer size, the electrons responsible for electrical conduction obey quantum mechanical law, demonstrating properties of both particles and waves, leading to quantum interference phenomena similar to the interference of water waves. The atomic-level impurities and lattice defects in the micro-metal lead to the interference between the wave functions, which are scattered from the impurities (or defects) and reflected at the edges of the metal. As a result, when the magnitude of the magnetic field applied to the metal changes, the electrical resistance exhibits a highly complex pattern of fluctuations in response. This complex pattern is essentially originated from quantum mechanics, and unlike the random thermal fluctuations of the classical world, it retains microscopic information such as the positions of impurities in micro-metals, thus being termed as a 'quantum fingerprint'. However, due to its complexity, the standard procedure has been to analyze these fluctuation patterns by averaging, making it challenging to decode the hidden microscopic information. In this research, we successfully managed to extract information about quantum interference of wave functions in micro-metals from the quantum fingerprints using AI. More specifically, we dealt with micro-metals incorporating various impurity arrangements by numerically solving the Schrödinger equation for electrons, allowing us to determine interference patterns of the wave functions and derive the magnetic field-dependent electrical resistance. We start with the machine learning on image data of these interference patterns based on a Variational Autoencoder (VAE), enabling us to extract their characteristic features in a latent space. Subsequently, we trained a network to estimate the characteristic features extracted by the VAE from the quantum fingerprints in electrical resistance. This approach enabled us to successfully reconstruct microscopic quantum information such as impurity positions and interference patterns from the quantum fingerprint in the micro-metal.

Next, we utilize AI in research on fluctuations that appear in the output of quantum computations performed by quantum computers and lead to errors in computations. Quantum operations performed on quantum bits (qubits), the basic units of a quantum computer, are governed by quantum mechanics, and it is known to be extremely challenging to achieve precise quantum operations due to various external disturbances. The frequency of errors increases with the number of quantum operations required for quantum computation, leading to greater fluctuations in the output results which make meaningful computations impossible. In this project, we performed reinforcement learning in AI for the ideal relationship between inputs and outputs for the quantum circuits (sequences of quantum operations) used in quantum computations. This allowed us to discover new quantum circuits that achieve (almost) the same outputs as well-known quantum circuits but are composed of fewer quantum operations. The discovered quantum circuit, with fewer quantum operations compared to conventional ones, can exhibit smaller errors in the final outputs. Implementing this new quantum circuit using superconducting qubits in an actual quantum computer has successfully led to output results with fewer errors compared to conventional circuits in quantum computations.

We are conducting research using AI not only for known physical systems that exhibit quantum properties, but also exploring new quantum physical systems for the AI-assisted research in this project. One of these is the system of magnons, which are elementary excitations in magnetic materials. Magnons are fluctuations in the quantum spins of electrons responsible for magnetism and have played a crucial role in the field of spintronics, where next-generation, low-power-consumption devices utilizing spins are actively explored. The utilization of magnons as quantum information carriers necessitates the analysis of magnon fluctuations through extensive data acquisition. In this study, we have successfully implemented a technique called state tomography for the first time concerning magnons, allowing the determination of the presence or absence of quantum properties in the states of magnons generated and controlled by magnetic fields and microwaves.

※For more detail, please refer to the URL below.
https://beyondai.jp/contents/wp-content/uploads/2024/02/Research-report_Eiji-Saitoh_202309.pdf