Interview with Young Quantum Researchers - Junyu Liu, Postdoctoral Fellow, University of Chicago
Data centers are globally collaborative, device-specific networks that deliver, accelerate, display, compute and store data information across the Internet infrastructure. According to industry bodies, global data traffic processed by data centers in 2020 will be as high as 1.53 ZB (1 ZB ≈ 1 billion TB), accounting for 99.35% of total global traffic. Meanwhile, as classical data grows in size, scientists are beginning to consider storing and processing data in quantum devices with large Hilbert space dimensions.
Recently, in a preprinted paper [1], Professor Liang Jiang's team at the University of Chicago proposed the concept of quantum data centers, which refers to some specific quantum hardware capable of efficiently processing quantum data and providing an effective interface between classical data and quantum processors. Quantum data centers can provide quantum data generation, processing and application services, and have a wide range of applications in quantum computing, quantum communication and quantum sensing.
PhotonBox took this opportunity to interview Junyu Liu, the first author of this paper - a postdoctoral fellow at the University of Chicago. Junyu Liu, 27, graduated from the Junior College of CSU and went to the U.S. in 2016 to study and research physics and quantum information science under several world-class quantum information scientists, including Professor Liang Jiang at the University of Chicago and Professor John Preskill at the California Institute of Technology - the currently popular NISQ and Quantum Hegemony concepts, both proposed by Preskill.
Not only an excellent researcher, Junyu Liu also has extensive experience in the business world, including work at top quantum computing companies such as PsiQuantum and IBM, and is currently an entrepreneur in the quantum space. What is less known is that Junyu Liu is also a second-generation teenager. Today, let's enter the multi-faceted life of Junyu Liu.
Dr. Junyu Liu
01About Research
PhotonBox: Please describe your educational experience and the reason why you chose the quantum field for your research?
Junyu Liu: I attended the Junior College at CSU from 2012-2016 as an undergraduate, then received my PhD at Caltech from 2016-2021, and am currently a postdoc at the University of Chicago and IBM. I have always been interested in physics, especially the connection between physics and computational science, especially algorithms (my father is a programmer. So I am also fascinated by computer programs). At Caltech, my PhD training in physics originated in fundamental physics, including black hole physics and information paradoxes, and solving quantum critical systems using advanced optimization bootstrap algorithms. My connection to quantum information science came from my Ph.D. advisor, John Preskill, who is a leading figure in quantum computing in the U.S. and who also switched to quantum computing from a background in fundamental physics. He suggested at that time that I should study quantum algorithms, especially those with a physics background. Under his influence, I gradually went into the research direction of quantum science. In this industry, I found the cross-disciplinary background very useful for my research in the quantum information industry. The fundamental physics taught me to think about the general laws of quantum computing from the perspective of first principles, and I found this particularly helpful for me to do quantum computing, especially quantum algorithm design.
PhotonBox: Please tell us why you chose the University of Chicago for your postdoc, what you did, and your connection with IBM?
Junyu Liu: My postdoctoral position is a long-term postdoctoral program sponsored by IBM at the University of Chicago, which is very liberal. I am also a member of the Kadanov Center for Theoretical Physics. I was particularly attracted to the University of Chicago during my Ph.D. I visited the city for a summer and was fascinated by this gorgeous city. The University of Chicago is a traditional powerhouse in quantum science, especially in its connection with quantum experiments and industry. In the last few years, as quantum technology has become more practical, quantum science and computer science have grown exponentially, and Chicago has become the center of the quantum industry in the United States. I chose the University of Chicago because of the presence of top scientists such as my postdoctoral advisor, Professor Liang Jiang, and I am grateful for the opportunities I have been given to work and research here. My main focus is to do research at the intersection of physics and information science, including machine learning, quantum computing, quantum networks, data science, and data security. I am also interested in the commercial implementation of related technologies.
PhotonBox: Please introduce the quantum data center mentioned in your latest paper?
Junyu Liu: This paper is co-authored by me, Dr. Connor Hann and Prof. Liang Jiang. We define the quantum data center in this paper. We believe that a quantum data center will contain at least a "quantum random access memory" and a "quantum network". The quantum memory is responsible for processing and storing quantum data, while the quantum network is responsible for transmission. We have found and constructed a number of examples of the important role of the combination of quantum memory and network, including the resource saving of quantum T-gates in error correction algorithms, quantum privacy requests and quantum encryption computation, quantum data compression and quantum sensing. Accordingly, we establish a general theory of quantum data centers and point out that quantum data centers can provide fast, efficient, secure, and private services to users and can serve as a natural extension of current data centers in the quantum era.
PhotonBox: We noticed that a lot of your research is about neural tangent nucleus theory, what does it refer to specifically? How can you apply it to quantum circuits?
Junyu Liu: Neural tangent kernel theory is a hot topic in the field of classical machine learning in recent years, with a deep physics background. It successfully explains to some extent the variation of generalized delay difference in deep learning with the number of parameters, and points out that the gradient descent method for deep neural networks is actually strictly solvable in the case of hyperparametric training. This theory is expected to be the first principle of deep learning and is currently an important theoretical tool for the industry of deep learning. This algorithm was actually invented by physicists who are active in the field of deep learning. In fact, the hyperparameter limit is very much like a "free quantum field theory", while the ratio of the depth to the width of the neural network is similar to a "coupling constant". We can use perturbation theory to do first-principles calculations of the dynamics of deep learning processes, just as we do in quantum theory. As a researcher with a background in fundamental physics, I was fascinated by this theory.
In my collaboration with Professor Jiang Liang and IBM, we pointed out that if the neural tangent kernel theory of classical machine learning is quantum field theory, then the neural tangent kernel theory of quantum machine learning is similar to the "matrix model" proposed by superstring theorists in the last century. In classical machine learning, we do Gaussian averaging of the training weights and biases; in quantum machine learning, instead of Gaussian averaging, we do Haar averaging, or k-design averaging, on the youngest positive group. We use this approach to solve for the dynamics of quantum neural networks in the overparameterized case. Our theory clears up some of the early discussions about the "barren plateau problem" that have plagued the quantum computing community for many years, and provides direct guidance for the design of quantum neural network circuits. This research was published in PRX Quantum on August 17 [2].
In addition to this, I have completed several studies on quantum neural networks (QNN) so far this year [3][4][5], including a preprinted paper just submitted on August 30 [6], where we proposed the effective quantum neural tangent kernel (EQNTK) and combined this concept with overparameterization theory to quantify the convergence of a quantum neural network to a globally optimal solution. We explain why QNNs using hand-symmetric ansatz typically have better trainability than QNNs using asymmetric ansatz. Guided by EQNTK, we further design a symmetric pruning (SP) scheme to automatically prune symmetric ansatz from over-parameterized and asymmetric ansatz when explicit symmetric information of Hamiltonian quantities is not available, thus greatly improving the performance of QNNs. Numerous numerical simulations have verified the analytical results of EQNTK and the validity of SP.
I am very proud of this series of work.
PhotonBox: What is one of your most impressive research experiences?
Junyu Liu: My supervisor during my first year of PhD was Prof. Cliff Cheung of High Energy Fundamental Theoretical Physics. He is a very powerful person and had a deep influence on my research. When I was in my first year, I did a year of work with him but no good progress was made. At this time, he kept encouraging me. Later on, together with Dr. Grant Remmen, we applied our techniques to a problem called the "weak gravity conjecture" and unexpectedly discovered that we could prove this fundamental conjecture about black holes and quantum gravity by using the information entropy method [7]. This work, which gave a general proof of this problem that has been open for decades, caused a great deal of reaction and a large number of scholars used our results in various places, even in purely gravity-independent quantum physics systems. Our work was then reported by Quanta Magazine [8]. I was very proud of it. It made me realize that good scientific work sometimes needs enough time to be polished, and that high requirements and standards are important for scientific discovery.
02About the mentor
PhotonBox: Your current mentor at the University of Chicago, Professor Liang Jiang, is a quantum information scientist that we are very familiar with. What have you learned from him?
Junyu Liu: I would like to take this opportunity to say a few words from the bottom of my heart. I have been working in academia for so many years, and I think Professor Jiang Liang is one of the most outstanding and best scientists I have ever known. I came to work at the University of Chicago in large part because of his personal charisma. Mr. Jiang Liang is a double-quotient, a true genius. He is very knowledgeable and well published, and his work spans many areas of quantum science, including quantum error correction, quantum hardware, quantum networks, quantum sensing, and so on. As a theorist, he is extremely good at studying the connection between quantum information theory and experiment, and realizing quantum hardware as conceived by theorists. At this point, I think he is the world's number one expert, bar none. His work in this field is irreplaceable. Outside of science, he treats people very well, makes them feel at home, and is very encouraging and supportive of young students like me. I have a great deal to learn from him in both academics and life, and I admire and respect him immensely.
03About Quantum Computing
PhotonBox: What do you think are the advantages of quantum computing over classical computing? What changes will quantum computing bring?
Junyu Liu: I think quantum computing is a natural extension of classical computing. A common assertion is that the scale of computing devices has now reached the nanometer scale. Further down the scale, the classical laws of physics no longer apply, which will lead to the failure of Moore's law. Quantum computing has unique laws that can provide huge speedups in some specific problems (e.g. Shor's cracking of RSA, ECC encryption algorithms). Industrially, I see him as a natural extension of the semiconductor industry and a sure way to meet human arithmetic needs.
04About the venture
PhotonBox: When did you start your business, and can you introduce your business project and company?
Junyu Liu: Actually, I have started my business for half a year, and it is an infrastructure company that uses advanced cryptography technologies including "post-quantum cryptography" to improve the security and scalability of Web3 and blockchain. The word "SeQure" replaces the C in "secure" with a Q, which refers to post-quantum cryptography and my background in quantum computing algorithms. We are mainly applying technologies that are already relatively mature in academics and scientific research to solve the looming security problems of blockchain products nowadays. You know, so far this year, $2 billion of cryptocurrency has been stolen by hackers. This is a staggering amount, so my vision and that of my founder Nancy is that we can make these exchanges, Defi, NFT project parties, etc. knife-proof. At a time when NIST is standardizing on post-quantum cryptography, our team was born with a great technical background, a startup team that came out of the University of Chicago. There are 86% PhDs, and many of them are big names in quantum as well as classical cryptography. We are fortunate to have such a startup team, and I feel even more fortunate that we all share the same startup philosophy: to bring breakthrough improvements to the emerging market of blockchain. We also all believe that digital technology can only flourish under the condition of hard technology to ensure the security of data and digital assets.
05About Quadratic
PhotonBox: On your homepage [9], we see that you often use words such as "art" and "drawing" to make the profound quantum science more understandable and attractive.
Junyu Liu: I was already very fond of anime and secondary elements. I think science and technology are very romantic things, so I wanted to show my own scientific work through various art forms, including comics, animation, music, etc. I had friends who helped me draw comics for some of my PhD theses. For some of my work during my postdoc, I also used DALLE2, a model of artificial intelligence for drawing, to make relevant illustrations. I'm even considering setting up a company dedicated to this one someday, but I don't have the energy for that right now.
Reference links：
[1]https://arxiv.org/abs/2207.14336
[2]https://journals.aps.org/prxquantum/abstract/10.1103/PRXQuantum.3.030323
[3]https://arxiv.org/abs/2203.16711
[4]https://arxiv.org/abs/2206.09313
[5]https://arxiv.org/abs/2205.12004
[6]https://arxiv.org/abs/2208.14057
[7]https://link.springer.com/article/10.1007/JHEP10(2018)004
[8]https://www.quantamagazine.org/black-hole-paradoxes-reveal-a-fundamental-link-between-energy-and-order-20200528/
[9]https://sites.google.com/view/junyuliu/main
