Network Security Internet Technology Development Database Servers Mobile Phone Android Software Apple Software Computer Software News IT Information

In addition to Weibo, there is also WeChat

Please pay attention

WeChat public account


Mimic human brain nerves: the latest research from Cambridge University appears in Nature, and artificial brain becomes a new direction of AI.

2024-05-30 Update From: SLTechnology News&Howtos shulou NAV: SLTechnology News&Howtos > IT Information >


Shulou( Report--

The latest research from Cambridge University shows that the AI model is similar to the neural structure of the human brain, which may become the key to the design of the AI model in the future.

As the most complex intelligent carrier on the earth, one of the greatest features of the human brain is that it can generate intelligence with high energy efficiency. If the AI system can be created according to the working principle of the human brain as much as possible, the working efficiency of AI will be greatly improved and the energy consumption will be greatly reduced.

Recently, the University of Cambridge did this research to find a way for the AI system to replicate the human brain.

Https:// research / news / ai-system-self-organises-to-develop-features-of-brains-of-complex-organisms literally translates-AI system organizes itself, generating a series of characteristics of the human brain, and even a variety of complex organizations.

AI simulation brain brain is no stranger to everyone, there are a lot of nervous system and tissue inside. All organizations and systems strive for limited energy and resources within the limited physical space.

But in order to live in harmony, the parts have to balance each other's needs.

This is why the brain structure of all species is similar, because after constant optimization and adjustment, people tend to develop into similar tissue solutions. This scheme can bring about a balance between the various parts.

"the brain is not only good at solving complex problems, but also uses very little energy to solve them," said Jascha Achterberg, a Gates scholar in the Cognitive and brain Science Group (MRC CBU) of the Medical Research Council of the University of Cambridge. "

New research from the University of Cambridge helps us understand why the brain looks the way it is. The research approach is to combine the brain's ability to solve problems with very little energy consumption in the process of solving problems, and consider them as a whole.

The co-author, Dr Danyal Akarca, also from MRC CBU, added: this stems from a broad principle that biological systems usually evolve to make the most of their available energy resources. The solutions they find are often very elegant, reflecting the tradeoff between the various forces exerted on them.

The study, published in Nature Machine Intelligence, is an AI system created by Achterberg, Akarca and other members of their team. The core is to simulate a very simplified brain model while imposing physical constraints similar to those of the human brain.

Studies have shown that this designed AI system has indeed developed some key features and developmental strategies similar to those of the human brain.

There are a large number of neurons in the human brain, and this system uses computing nodes, not real neurons. But the function of neuron and node is similar, both receive input, convert input and produce output, and there is no difference that a single node or neuron may connect to multiple other nodes or neurons.

Moreover, all information is calculated after it is entered.

Here is the main part-the physical limitations imposed by the research team.

Each node has a specific location in the virtual space, and the farther away the two nodes are, the more difficult it is for them to communicate. This is similar to the way neurons in the human brain are organized.

First, the researchers assigned the system a simple task-a simplified version of the maze navigation task, usually assigned to animals such as rats and rhesus monkeys when studying the brain. participants must combine a variety of information to determine the shortest route to the destination.

One of the reasons why the research team chose this task is that to accomplish this task, the system needs to remember a series of elements-including the start position, the end point, and the intermediate steps.

Once the system has learned how to perform tasks reliably, it is possible for researchers to observe which nodes are important at different times of the experiment.

For example, a particular node cluster may be responsible for encoding the location of the end of the maze, while other node clusters may focus on encoding available routes.

Therefore, researchers can track which nodes are active at different stages of the task, so as to determine the different functions of each node.

At first, the system didn't know how to get out of the maze and even made mistakes.

However, when the system gets feedback, it will gradually understand how to accomplish this task more efficiently through continuous self-learning.

Specifically, the AI system learns by changing the intensity of connections between nodes, similar to the changes in the strength of connections between brain cells as we humans learn.

After self-learning, the system repeats the task over and over again until it finally learns how to perform all tasks correctly.

However, in the system they designed, there are physical limitations, which means that the farther apart the two nodes are, the more difficult it is to establish a connection between the two nodes based on feedback. The same is true in the human brain-connections across large physical distances are difficult to form, and such connections are even more valuable if they are to be maintained and strengthened.

When the system is asked to perform tasks under these constraints, AI uses some of the same skills as the real human brain to solve the task.

For example, to bypass these limitations, AI spontaneously starts developing hubs, a highly connected node that acts as a conduit for passing information across the network.

What is even more shocking is that the response characteristics of the individual nodes themselves begin to change, in other words, each node is not encoded for a particular attribute of the maze task (such as determining the target location or making the choice of the next step). But will gradually develop a flexible coding scheme.

This means that nodes may encode various attributes of the maze at different times.

For example, the same node can encode multiple locations of the maze without the need for a special node to encode a specific location. This is also a remarkable feature of the brain of complex organisms.

Professor Duncan Astle of the Department of Psychiatry at Cambridge University, one of the co-authors of the paper, says this simple limitation is another feature aimed at the brains of complex organisms.

With a little simple limitation, such as the above mentioned, it is difficult to connect two nodes with longer physical distance, which will force the AI system to produce some rather complex characteristics.

The focus of the study is that these characteristics are common to biological systems such as the human brain, that is to say, through the simulation of AI, we have to study the human brain eventually.

What the team hopes is that their AI system will begin to reveal how these constraints create differences between human brains and how they lead to differences in people with cognitive or mental health difficulties.

Professor John Duncan, one of the co-authors of the paper, from CBU, the UK Medical Research Centre, said: these artificial brains provide us with a way to understand the various data recorded in the real brain related to the activity of real neurons.

Without this step, the data is just data, in other words, abstract.

Achterberg added: "the AI brain allows us to ask problems that cannot be solved in a real biological system. We can train the system to perform tasks, and then experiment with the limitations we impose to see if it starts to become more like the brain of a particular individual."

The impact on the design of future artificial intelligence systems, of course, in addition to helping brain scientists study the human brain, this research can also arouse interest and widespread discussion in the AI community, because they can develop more efficient systems, especially when there may be physical limitations (when physical restrictions are not imposed, but objective).

One of the researchers said that AI researchers have been working on how to develop a more complex nervous system that can code and perform tasks in a flexible and efficient manner.

In order to achieve this goal, developers think that neurobiology will give them a lot of inspiration.

For example, the overall cabling cost of the system they created is much lower than that of a typical AI system.

You know, many modern AI solutions use architectures that are ostensibly similar to the brain. The researchers say the new study shows that the type of problem that artificial intelligence tries to solve affects the question of which architecture is the most powerful.

"if you want to build an artificial intelligence system to solve similar problems to humans, it will eventually be closer to the real brain than a system running on a large computing cluster, which specializes in very different tasks from humans," Achterberg said. "

"the architecture and structure that we see in the artificial brain exists because it helps to deal with specific brain-like challenges. "

This means that AI robots must process large amounts of ever-changing information with limited energy resources, and having a brain structure similar to that of humans allows them to perform many tasks with twice the result with half the effort.

Professor Achterberg added: the brains of AI robots deployed in the real physical world may be more like our brains because they are more likely to face the same tasks as ours.

They need to constantly process the new information from the sensor and control their bodies to move to the target point in space.

Many systems need to run all related calculations with limited power supply. Therefore, in order to balance these energy and resource limitations, and the amount of information that needs to be processed, AI systems are likely to need a structure similar to that of the human brain.

The research, currently funded by the Medical Research Council, Gates Cambridge University, the James S McDonnell Foundation, the Templeton World Charitable Foundation and Google DeepMind, is expected to have a significant impact on both brain science and AI in the future.



Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.

Views: 0

*The comments in the above article only represent the author's personal views and do not represent the views and positions of this website. If you have more insights, please feel free to contribute and share.

Share To

IT Information


© 2024 SLNews company. All rights reserved.