10Kv Steel Pole,Metal Pole,Steel Pole Jiangsu Baojuhe Science and Technology Co.,Ltd. , https://www.galvanizedsteelpole.com
Tao Dacheng: The trend of robot commercialization is strong but the foundation is very important Artificial intelligence and robots are getting closer and closer to real life, which can be seen from the flow of talent. In 2016 alone, several well-known academic scholars joined the industry one after another, such as Li Feifei, Ruslan Salakhutdinov, Yoshua Bengio and so on. The close cooperation between the academic community and the industry will greatly help the commercialization of artificial intelligence and robotic products, and the accumulation of application data in the corporate world will also advance the process of academic exploration.
For the two-way cooperation between the academic community and the industry, the world's top artificial intelligence researcher, University of Sydney professor Tao Dacheng has a positive attitude. It is worth noting that the computer vision and machine learning research that Tao Dacheng is engaged in is driving the development of artificial intelligence and robotics. The results of continuous breakthroughs have earned him the prestigious awards such as the Australian Science Highest Honor Eureka Award and his election to IEEEFellow and the European Academy of Sciences. From face recognition to multi-angle learning, from the commercialization of artificial intelligence and robots to the perception of technical singularities, in the story of Tao Dacheng, the development of artificial intelligence and robots has been outlined more clearly.
“In the fastest year, face recognition can be reflected in the robotâ€
In many fields of artificial intelligence, the common people may have a perceptual understanding of the field of face recognition, and major technology companies have set up face recognition groups in recent years, and use this as a starting point to lay out artificial intelligence. Face recognition is precisely the main research direction that Tao Dacheng has studied for many years, won numerous international competition awards, and achieved breakthroughs.
Face recognition is a beginning to help robots recognize humans and thus advance human-computer interaction and collaboration. Tao Dacheng mentioned that on the one hand, humanoid robots can be better interacted with people and integrated into the family through face recognition. On the other hand, they can be used for personnel search by integrating face recognition.
However, face recognition technology still has no small challenge. "The biggest dilemma is the lack of robustness of the face recognition algorithm itself." That is to say, the current rigid algorithm lacks the ability of flexibility and imaginative reasoning, and has insufficient ability to cope with changes in external factors such as illumination and occlusion. The challenges faced by robot face recognition are also more unique, such as the loss of facial lines and the higher real-time requirements.
The Tao Dacheng team's research has achieved breakthroughs by measuring five “outstanding and stable†facial feature points that vary little in different environments for face recognition. The results show that whether you make a sad or happy expression, or different lighting conditions indoors or outdoors, these five points can maintain high stability. This helps to improve the robustness of face recognition.
For the robot, Tao Dacheng team proposed a fuzzy invariant feature learning method based on CNN, which solved the effect of image blur on face recognition. At the same time, they proposed a face recognition model of the backbone branch integrated network, which was improved. While being robust to attitude changes, it also effectively ensures the real-time performance of the recognition algorithm.
Applying these technologies, the Tao Dacheng team won the first place in the 2016PaSC face recognition competition with absolute advantage. According to the evaluation of the game organizers, their algorithms achieved performance consistent with the human visual system.
“The development of machine vision has been going on for many years, face recognition has the longest development timeâ€, and the application of deep learning has increased the recognition rate of face recognition from 20%-50% in the past to over 90%. The continuous improvement of technology and the partial realization of practical level have shown great market prospects. This is an important reason why many domestic and foreign technology companies choose face recognition as the main direction.
“Face recognition is likely to be reflected in the robot in just one year,†Tao Dacheng said.
Face recognition is only the first step. It makes the machine smarter, better recognizes human beings and develops to be able to learn, imitate and create. It also needs a lot of efforts, including multi-visual learning, human pose estimation, behavior analysis and so on. Human pose estimation can help robots perceive human behavior in the family, while behavioral analysis combines scenes to help the machine identify a person's behavior from the video or behaviors in which multiple people participate, such as a person swimming, multiple people Playing football. Through multi-angle learning, robots can understand a thing, an event more accurately.
Tao Dacheng's team has made breakthroughs in these fields of computer vision and machine learning, and published research results in authoritative journals such as IEEETPAMI and TIP. In the 2016 AcTIvity Net competition, the Tao Dacheng team achieved the first performance test task and the second performance recognition task.
The trend of artificial intelligence and robot commercialization is "the foundation is very important"
As technology matures, commercialization will follow. Driven by the influx of academic talents in the industry, the advancement and future layout of artificial intelligence by major technology companies has accelerated the commercialization process.
"The commercialization of artificial intelligence is the general trend." He believes that when a technology has reached maturity, its commercialization prospects will be very impressive. However, the commercialization of artificial intelligence is also facing many challenges. The immature technology, the unemployment rate caused by the development of artificial intelligence, the reverse suppression of artificial intelligence development and the business model are all factors that may be affected.
“A benign commercialization model is at least stable and stable. The product development of the enterprise is based on mature, self-developed core technology.†He suggested that for unresolved application problems, the technology exposed by the academic community cannot be blindly adopted, but Develop your own research team and focus on independently developing this card.
“All in all, the foundation is important.â€
Intelligent robots, driverless, civilian drones, face recognition, smart voice... These commercial artificial intelligence technologies are a good commercial target in his view.
It is impossible to realize the singularity of technology. There is no need to panic. The AI ​​threat theory is in the recent popular American drama "Western World". In the future world, robots with the same appearance and human beings have self-awareness and even manipulate humans in turn, which has also caused many People have a new round of hot discussion about the pros and cons of artificial intelligence and robots. Prior to this, many scholars and celebrities such as Hawking and Bill Gates have expressed their vigilance against the development of artificial intelligence.
“Technical singularity is impossible to achieve,†Tao Dacheng said.
He mentioned that the idea of ​​Hawking et al. stems from the technical singularity, that is, the point in time when machine intelligence begins to leapfrog, and it is also the starting point for a new, intelligent species to begin to rule the earth. However, in Tao Dacheng's view, the application of any thing is good or bad. "Artificial intelligence technology is a double-edged sword, the key is how to use it reasonably."
"In the future, technology and humanities must be interactively developed." Tao Dacheng believes that artificial intelligence will not develop in a situation of out of control, so the public does not have to panic, and as a technology and industry practitioners, "we must follow healthy development. Road".
In fact, artificial intelligence is still in the stage of weak artificial intelligence, and there is still a long way to go before the concept of the public. On the hardware side, artificial intelligence also requires more powerful perceptrons, smoother and more stable mechanical systems, and more efficient computing systems to realize the intelligent analysis capabilities of the machine; in software, not only a general-purpose system design is required. The framework, for each sub-module, also requires efficient algorithm design and substantial theoretical verification.
"Our research and development work in artificial intelligence is to use computers to improve our lives." This is the original intention of many scholars including Tao Dacheng.
Tao Dacheng believes that when a technology has reached the stage of maturity, its commercialization prospects will be very impressive. But the foundation of industrial development is equally important. Intelligent robots, driverless, civilian drones, face recognition, smart voice... These commercial artificial intelligence technologies are a good commercial target in his view.
September 27, 2024