Fudan University and Ant Group join forces to "beat" the human brain with the human brain

In 2003, a graduate student named Wang Xiaofei from the Department of Computer Science at Tsinghua University took a neurobiology course and submitted a course assignment titled "Connecting the Computer CPU and the Human Brain." The biology teacher commented, "There will be someone like you in the Computer Science Department every ten years."

Since the birth of computers, people have tried to study the human brain through mathematical analysis and computer simulation methods. This is how the branch of computer neuroscience came into being. Computer and biological experts are using new tools and interdisciplinary cross-technology applications to promote brain science research.

What Wang Xiaofei didn't expect was that 20 years later, his dream had a new opportunity.

Recently, at Ant A Space in Hangzhou, there was an extremely brain-burning "Brain Simulation Architecture Based on Graph Computing" project launch meeting. Ant Technology Research Institute and Fudan University Brain Science Institute jointly launched a new industry-university-research integration research project, using graph computing technology to conduct brain-like research, giving people a new way to understand the brain.

A new attempt at cross-border

As a cross-disciplinary scholar who is familiar with both computers and neuroscience, Wang Xiaofei is the matchmaker of this project. He is currently an engineer at the Institute of Brain Science at Fudan University. Chen Wenguang, currently the dean of Ant Technology Research Institute and the main person in charge of Ant Graph Computing, was a professor in the Department of Computer Science at Tsinghua University. Wang Xiaofei has known him for many years.

"Xiao Fei asked me to have coffee at the gate of Tsinghua University one day, and then handed me a paper, saying that the brain had already reached such a level, and my graph calculation should be able to be used." Chen Wenguang recalled that his first feeling after reading the paper was: This is a good thing.

On the other hand, Wang Xiaofei communicated his ideas with Wang Yun, senior PI and doctoral supervisor at the Institute of Brain Science of Fudan University. Although it was the first time he heard the word "graph computing", when he saw the natural similarities between graph computing and brain neural networks, especially the dynamic changes in graph relationships, which are very similar to the growth and plasticity of synapses, Wang Yun felt that this would be the most meaningful thing he would do before retirement.

Both sides have a strong expectation: at a time when new technologies are emerging one after another, it is time to do something groundbreaking.

"Of course, cross-border is a very difficult thing. At the beginning, Teacher Chen Wenguang didn't know the terms I mentioned, and I didn't understand the terms Teacher Chen Wenguang mentioned." Wang Yun admitted that fortunately there was Wang Xiaofei among them.

From left: Chen Wenguang, Wang Xiaofei, Wang Yun, Zhu Xiaowei

Next, this project will rely on the Neuropharmacology Laboratory of the Institute of Brain Science at Fudan University and the Graph Computing Laboratory of Ant Technology Research Institute to create a new generation of large-scale, high-precision brain simulation system through the close integration of graph computing, artificial intelligence and experimental neuroscience, providing new research methods for understanding biological intelligence and treating brain diseases, and assisting cutting-edge scientific research.

Simulating the brain with computers

What is brain simulation? What is graph computing? Many people may be confused after reading this: Can computers really simulate the brain? Just like when the science fiction film "Transcendence" was released in 2017, countless people who watched the film couldn't help but ask: Can we really upload the brain of the starring actor Johnny Depp into a computer one day?

"We all know that the human brain is very complex and extremely important. This is why a person's brain only accounts for 2%-3% of the body weight, but its oxygen consumption accounts for 20% of the whole person." Wang Yun explained to everyone that the basic unit of the human brain is the neuron. There are about 100 billion neurons in our brain, and they are connected by about 100 trillion synapses. Neurons are used to transmit and process information through this intricate connection relationship.

We often say that once bitten by a snake, you will be afraid of snakes for ten years. The principle is that the transmission of information from neurons one by one forms a "fear memory" and affects our behavior in the future.

Human beings have never stopped exploring the frontiers of brain science. At present, Europe, the United States, Japan and other countries have successively launched well-known research projects such as the "Blue Brain Project" and the "BRAIN Project". my country also officially launched the "China Brain Project" in 2021. The brain projects of various countries in the world have acquired a large amount of neurobiological data.

At the same time, on the technical level, IBM released a human brain-like chip in 2014. The chip has one million "neurons" and 256 million "synapses" and can temporarily simulate brain activity. In the same year, Japan's "K" supercomputer (K ​​Computer) used 83,000 processors and successfully simulated one percent of human brain activity in one second. Previously, Musk also claimed that he had uploaded his brain to the cloud.

However, the common architecture of current brain simulation systems is still based on the amount of data, computer capabilities, and biological understanding of nerves 20 years ago. Therefore, computational neuroscience needs a completely new brain simulation architecture to accommodate larger amounts of data and provide greater computing power, thereby achieving more accurate simulation and prediction. From this perspective, graph computing technology inherently has similarities with neurons and brain simulation. The high-precision brain simulation architecture built based on graph computing technology is expected to provide a new research method and means to reveal the secrets of the brain.

A high-precision brain simulation architecture

When it comes to graphs, the most common and easiest to understand for ordinary people is a picture. "But the graph we are talking about here is not a graph of pictures, but more like a knowledge graph, an abstract data structure consisting of two basic data types: vertices and edges," explained Zhu Xiaowei, a researcher at the Graph Computing Laboratory of Ant Research Institute. For example, if we regard projects, employees, and companies as three different vertices, there are various relationships between these different types of vertices. Employees may be friends with each other, employees may be employed by a certain company, and may participate in a certain project. These relationships are edges.

As a computing model that builds relationships based on points and edges, the biggest advantage of graph computing is that it can dynamically and real-time analyze point-edge relationships as large as 100 trillion. Ant began exploring graph computing in 2015, and has deployed related technologies such as graph databases, streaming graph computing engines, and graph learning, creating a world-leading graph computing cluster.

Alipay, as an important financial tool, has been using graph computing in the risk management of some black and gray industries such as fake orders and fraud. Taking the Internet scenario as an example, bad merchants often use different bank cards and acquaintances to complete fake orders or cash out "circuits". The data model of pure capital flow cannot reveal clues, but with the help of graph data models, such closed-loop models can be discovered to prevent risks in advance.

For this reason, in the future, we can try to restore the structure of neurons with high precision through graph computing, simulate biological experiments to obtain data, and further verify each other with biological experiments. Wang Yun said with a smile that when people mention biologists, their first reaction is "studying mice all day long". Indeed, if the research on brain science is limited to biological laboratories, it is undoubtedly very boring and inefficient. However, with the assistance of high-precision simulation of the brain, scientific experiments will become more efficient, fast and low-carbon.

When humans thoroughly study their own brains, the subsequent imagination space will be greater, such as the treatment of diseases, providing models for early target screening in drug research, promoting more sophisticated and efficient drug development, and shortening the time it takes for drugs to be put on the market.

Of course, all this is just the beginning. As Chen Wenguang from Ant Technology Research Institute said: "Interdisciplinary fields are very promising, but scientific exploration itself is an uncertain thing. It is still unknown what kind of results will be produced in this field. It's just that we are on the right path and in the right direction."

Liu Yongli, reporter of Metropolis Express and Orange-Persimmon Interactive