Big data is a hot spot a few years ago.
However, with the development of artificial intelligence, its role is increasing day by day. From "data for people" to "data-driven people's behavior", this has caused tremendous changes.
On August 13th, Yin Xiangzhi, chief data scientist of China Academy of Sciences, was invited to attend the CCF-GAIR Summit. Huayuan Data has been engaged in big data related applications since 2002 and has accumulated for many years. The current direction of development is in AI, robotics, and intelligent driving. industry.
In our communication, Yin Xiangzhi throws a very interesting topic: “Have you found out that dogs and animals are so amazing?†He continued to explain why we can all be called dogs when there are so many differences. The wolf and the dog are all the same from the species of the phylum Compendium, but the wolf is a wolf and the dog is a dog. The difference between a dog and a wolf is far less than the difference between a Chihuahua and a Tibetan mastiff, but people will tell which one is a wolf and which one is a dog. This is a question of understanding and understanding. It is also a difficult point of artificial intelligence. Yin Xiangzhi gave his own opinion.
Difficulty of data
In the automotive industry, Yin Xiangzhi said that now automakers or Internet companies, they have enough capital and demand to get this part of the data. The traditional way of thinking is to own the data and put it into cash, but if you do not have the ability to convert data into more value, then they are still at the bottom of the industry chain. Production data and data collection may generate short-term gains, but the liquidity can be greater than the cost of storage.
If China is to become a big country with artificial intelligence, the government may have subsidies on storage, and new storage technologies will also reduce costs. Yin Xiangzhi said that it is more important to make valuable data conversions.
Difficulties with AI
The current data is to allow the machine to identify it within the definition of the person. The more data the more accurate identification can be made. But the problem is that if you go beyond this range, that is beyond the scope of supervised learning, the machine will not be able to identify, this is the most terrible thing. It is impossible for human beings to exhaust all possibilities, so artificial intelligence is now taking the path of non-supervised learning. Exceeding people's limited scope, when the outside world has no experience and training data samples are provided to the machine, it is entirely on the machine itself. This is also the trend of artificial intelligence in the future. It is a problem of technical security.
The difficulty of security
Another data security is also difficult. Using a car to cover all data calculations, including recognition and deep learning, is computationally expensive, and it is also very costly for the car's power consumption. If placed in the cloud, there will be a delay due to a slight signal problem. These basic reserve issues need to be resolved. This time, V2V and V2X are parallel calculations between cars and cars. They can be partially offline and each other's contribution to computing power is a breakthrough point.
Difficulties in driving
Yin Xiangzhi predicts that although companies in the smart car industry chain are currently developing their own underlying technologies, they will begin to cooperate within three to five years. For smart driving, the difficulty is not only identification, the most important thing is to make decisions through data. It is also necessary to judge through the identified pictures to determine the intention of arriving around.
Some things can not be done by one's own efforts, so Yin Xiangzhi expressed an optimistic view of the data open source, he believes that the key technology is the need to apply for a patent to guard, but for data, through sharing can make the data analysis more accurate and Intelligent.
There is a greater need for more realistic data that has been manually annotatedVision sensing in smart driving technology requires more realistic pictures and requires manual annotation. This is an initial level of processing. The cost of labor is very high, but the value of what is done is valuable. Unlabeled data is not needed. But for big data companies
Guarantee its algorithm accumulation
Have high quality data
Flexible business model
It is very important that Huayuan data selection plays a very neutral role, defined as a data platform, and hosted the BOT big data contest in Shanghai. Yoo Sang-chi saw that in fact, artificial intelligence can't be done for people who are 3 years old. Therefore, he hopes to learn from others and hopes that artificial intelligence can become the industry. Yin Xiangzhi originally started work on big data mining in Taiwan in 2002. He has a deep understanding of artificial intelligence. He also admitted that driving assistance is fine, but talking about smart driving is still too early.
ConclusionFinally, the problem of returning to the wolf and the dog is difficult to distinguish from the appearance. For example, the Huskies and Alaskan Malamutes are very similar to the wolf, but the difference lies in the dog's domestication. The wolf is different, and the "personality" is very different. obvious. This is also a very difficult point for machine learning. There is no recognition of emotions, only data and algorithms. How to make big data into real smart data can make decisions for cars and people? There is still a long way to go. Many people engaged in smart car technology research and development have recognized that it is not possible to let autopilot cars get on the road within five years, and it is also dangerous to have only one or several cars on the road. Nowadays it can only be achieved from the point of assist driving. In smart driving, what is more needed is a rational attitude.
