In recent years, the development of intelligence has become faster and faster. Service robots, intelligent cockpits, and autonomous driving devices have gradually penetrated into people's lives. These devices need to interact with people and require good perception capabilities, including vision, hearing, and even touch. Multimodal interaction It will be the development direction of brain technology. It is predicted that by 2035 , brain-like intelligence will account for 18% of traditional artificial intelligence, exceeding the market size of 20 billion US dollars.

In recent years, with the rapid development of intelligence, service robots, intelligent cockpits, and autonomous driving devices have gradually penetrated into people's lives. These devices need to interact with people and require good perception capabilities, including vision, hearing, and even touch, and multi-modal interaction. It will be the development direction of brain technology.

According to a report by market research firm Yole, from 2020 to 2030, brain-like growth will usher in rapid development. It is predicted that by 2035, brain-like intelligence will account for 18% of traditional artificial intelligence, exceeding the market size of 20 billion US dollars. In addition, Qiao Ning, founder and CEO of Times Technology, believes that brain-like intelligence can be used not only for artificial intelligence, but also for logical operations.

At present, the research community and business community have realized the advantages of brain-like intelligence. However, industry insiders generally believe that brain-like intelligence is still half a step away from landing. What are the challenges that brain-like intelligence encounters in landing applications at this stage? If these issues are resolved, which orbit will land first in the future? These issues have been discussed comprehensively by many industry experts recently.

What are the challenges of applying brain-like intelligence to the ground?

First, the current cost is still high. Yu Lei, associate professor and doctoral supervisor of the School of Electronic Information, Wuhan University, said that in recent years, brain-inspired technology has developed rapidly and can indeed solve many problems. However, the problem is too expensive. They need brain-like cameras during their research. It costs 55,000 yuan to buy one, and more than 200,000 yuan to buy four.

Second, there is room for technological improvement. Xu Shu, director of the Advanced Algorithm Laboratory of the South Lake Research Institute of China Electronics Technology, analyzed that there are two main aspects of brain-inspired technology from laboratory to application: First, if the technology is to replace existing technology, it depends on how much it costs. The main problem is solved; secondly, whether the technology can solve the problem that the existing technology cannot completely solve.

What are the different steps in the application of brain-like technology? Brain-like technology shows many performances, such as motion capture of event cameras, privacy protection, etc. , which does show some problems that cannot be solved by existing artificial intelligence, such as very low power consumption, which can recognize and process actions efficiently. However, it is still less than half a step, mainly due to issues such as pixels, imaging quality, stability, etc., and cannot solve the problems encountered in more scenes.

Third, the research on the neural network of the brain is not enough. BP5718A12 Dr. Deng Lei from Tsinghua University said that they often think about a question during their research. What is true brain intelligence? They found that in the study, most of the time, it just completed the mapping relationship from data input to output, and they hope that brain-like intelligence can reflect intelligence at a higher cognitive level, which cannot be solved in the short term.

Deng Lei said that the study of neural networks is to understand more about the working mode of the brain, so as to explain people's behavior. What is being studied now is only the tip of the iceberg of the brain. Research is difficult to apply, and the underlying mechanisms of research are too few. Therefore, it is still difficult to carry out brain-like intelligence on this basis.

In the future, which track brain-like intelligence will be the first to land?

If the current challenges are solved, which track will be the first to apply brain-like intelligence in the future? Xu Shu believes that if the pixels, imaging quality, and chip computing performance can be improved. , it will first land in the smart home, because the brain-like chip focuses on low power consumption, which is very suitable for consumption scenarios. In addition, brain-like has a good protection effect on privacy. After home, various smart robots are a good track.

Brain-like intelligence includes perception level and processing level. From the perspective of perception, there are vision, hearing, touch and so on. Yu Lei said that at present, the visual aspect is relatively mature, such as the event camera, which has a fast response, is basically subtle, and has a good dynamic range. In extreme scenarios, such as sports, strong and weak light, etc., the performance is good, and the scenarios that can be applied in the future are very large.

If the above technical problems are solved, the technical bottleneck of brain-like intelligence may no longer be a big problem, and the difficulty will be ecology. For example, ANN and NLP networks can be applied and are inseparable from ecological construction. Therefore, scientific research institutions and enterprises need to have such consciousness to truly promote the comprehensive application of brain-like intelligence.

Another investor believes that as a new technology, everyone sees the bright spot of brain-like intelligence technology. However, how to promote the application actually has a methodology. The common approach is to find differences in technology, and then find enough segments to match the differences to form bursts at the right landing spots.