In the intelligent society with the ternary fusion of human-cyber-physical, the amount of data generated by humans in a year is already more than the sum in previous history. As the core cornerstone of intelligent society, the strategic significance of intelligent computing is becoming increasingly prominent. How can we empower new scientific discoveries through intelligent computing? How will intelligent computing reshape social governance?

Recently, CHEN Hongyang, Deputy Director and Research Fellow of the Research Center for Graph Computing of Zhejiang Lab (ZJ Lab), was selected as one of 2022 China's Intelligent Computing Technology Innovators by DeepTech. On April 14, 2023, thirty young stalwarts with great talent and innovation in the field of intelligent computing gathered at the Intelligent Computing Technology Innovator Summit to discuss the present and future of intelligent computing.

Why is he on the list of top computing talents? What changes will his research bring to computing? We had a conversation with him.

Establishing a Foundation Model to Create Biopharma GPT

CHEN Hongyang's research is on graph computing.

It is about computing graphs rather than images. In the era of big data, there are various simple or complex relationships between data, and graph computing is a way to express and process these relationships. "Most of the current AI technologies can only process structured data, and there are huge amounts of unstructured data that need more matching algorithms to mine and analyze," said CHEN Hongyang. "Graph computing is a very important way to process unstructured data, so the industry also sees graph computing as the cornerstone of the next generation of AI."

Where should we start to build the cornerstone? CHEN's team chose the large-scale efficient graph computing platform, which can be commonly understood as a simple and easy-to-use platform in the field of graph computing. The platform will provide services based on large-scale graph pre-training models and knowledge graph fusion to support scientific computing and other applications.

In recent years, the continuous iterations of foundation models have led the AI all the way to a wild ride, and the emergence of ChatGPT has made general intelligence possible. ChatGPT is expected to solve the problem of AI application fragmentation. The foundation model that CHEN Hongyang wants to make is different from these general-purpose foundation models, and he focuses on scientific computing, and takes the lead in the biopharmaceutical field. CHEN Hongyang named the foundation model "Biopharma GPT".

It is a foundation model for a vertical field. "A biomolecule naturally has a structure that approximates a graph, for example, its atoms can be seen as 'nodes' and its chemical bonds as 'lines'," he said. At the beginning of the project, the first choice of his team is about biopharmaceuticals - the application scenario of the graph computing platform. "Computing is used to inspire scientific research, that is, AI for Science. It is the important goal of ZJ Lab for intelligent computing, also our mission."

The potential applications of the Biopharma GPT include drug design and target discovery, which are crucial in the field of biopharmaceuticals. The traditional process of drug design and target discovery usually requires massive human, material and financial resources, and a long experimental cycle. The Biopharma GPT can generate many diverse molecules in a short period of time, provide a broader library of molecules for screening, and induce generation for specific biochemical properties (e.g., molecular weight, and solubility), thus improving the efficiency and success rate of drug development. Based on a large number of medical materials and biochemical data, the Biopharma GPT can also explore potential drug targets and even predict the interaction between targets and potential drugs, thus reducing the experimental cycle time, saving costs and improving the success rate of drug development.

The development of the Biopharma GPT will be a starting point for the future expansion of related technologies to a wider range of applications such as e-commerce, social networking, chemistry and physics.

Of course, the large-scale efficient graph computing platform is just a small step of his team. "Our goal is to develop from chips and programming frameworks to hardware and software platform integration design, and finally build our own independently developed and controllable graph computers," said CHEN Hongyang.

Switching Research Direction to Intelligent Computing, Doing "Embrace Change" Research

From the University of Tokyo to Fujitsu Limited and then Zhejiang Lab, CHEN Hongyang's research experience and research field have broadened with the development of the AI era. It can be said that he has witnessed and even experienced the wave of the times.

Before joining ZJ Lab, his research focused on network communications. "I followed a team to build some large ICT systems, like the Internet of Things and 5G systems," he said. "I have worked in this field for more than ten years, and also achieved many patents and academic results." Two years before returning to China, he noticed that the explosive growth of big data has played a good catalytic role for the application of AI, and AI began to show strong capabilities in some areas. "When I did my Ph.D., AI was still in an unpopular phase," he said. "With the development of the era of big data, digital transformation began to become the focus in all walks of life, and I also began to consciously expand my research field and get involved in big data and AI."

In his view, network communications represent connectivity, AI and big data represent computing, and connectivity and computing are the base for building an intelligent society. With the advent of the computing era, he feels that it is time to follow the trend and make use of his years of knowledge of network and computing to develop intelligent computing systems that integrate network and computing technologies, including foundation models and efficient graph computing platforms.

In July 2020, CHEN Hongyang returned to China and joined ZJ Lab, officially starting a brand new scientific career. "The Lab targets the strategic direction of intelligent computing to build intelligent computing data reactors," he said. "It happens to coincide with my interests and planning."

Based on his previous technical accumulation, he has been walking solidly on the track of graph computing. Last year, he led his team to self-develop ZJ Zhuque, a large-scale efficient graph computing platform based on distributed memory computing. The platform broke the world record for the edge prediction track and ranked first in the world in the OGB Global Challenge, the top international graph learning benchmark review list. Recently, his team also won the Wu Wenjun Award for AI - Second Prize for Science and Technology Progress Award.

At present, his team has more than 30 members from academia and enterprises, with different scientific research backgrounds such as biology, physics and computer. "AI for Science requires a blend of different domain knowledge, and in addition to the cross-disciplinary team, we also conduct weekly cutting-edge sharing in full English," he said. "Biology, breeding, quantum computing, and other seemingly unrelated professions are what we learn and exchange." High intensity and high frequency of input is the norm for his team. When it comes to computational breeding and computational biology, almost every team member can talk about them eloquently. "To make a pervasive graphing computer, we must be able to see what the industry will look like in five years. Lay out and plan ahead, so we can always stay ahead," he said.

Just like his own experience, "I have always believed that both organizations and individuals need to transform," he said. "For researchers, transformation is to continuously expand their knowledge, and after a certain degree of knowledge accumulation, they should follow the development trend of science and technology to broaden their research horizons. By seeing the future of the industry, you can also see your own future."