Intel song Jiqiang: adhere to the long-term principle of scientific research and promote AI to the 3.0 era

category:Internet
 Intel song Jiqiang: adhere to the long-term principle of scientific research and promote AI to the 3.0 era


From the perspective of the development process of AI, human exploration of AI has lasted for more than 70 years. Looking back at the development of AI, we can clearly capture several key nodes. The first wave of AI is theoretical reasoning through various rules formulated by people. Although it performs well in reasoning, it is limited to a few strictly defined problems and has no learning ability to deal with uncertainty. What really makes AI better is the second wave of AI triggered by deep learning. The massive data generated by the Internet, mobile Internet and so on provide the machine with learning, mining and trial and error objects, so that the system can spontaneously find the rule, make prediction, judgment and decision-making. The growth of data, the improvement of computing power and the evolution of algorithms based on deep learning make some typical deep learning applications reach or even surpass human capabilities. This makes more and more optimists believe that in-depth learning is a valuable and worthy direction for the industry to follow up on a large scale.

However, is deep learning the ultimate answer to AI? With the research of deep learning, we find that there are still some problems to be solved. First, energy consumption is the biggest challenge. According to some research reports, using server level CPU and GPU cluster to train a large-scale AI model, the carbon emissions generated by power consumption is equivalent to the carbon emissions consumed by the entire life cycle of five American cars. Imagine how the human ecological environment would be destroyed if all walks of life followed this AI computing model. Then, data volume is another big challenge. At present, deep learning relies too much on big data. In some scenarios with small amount of data, the use of deep learning will be very limited. AI should learn from small data, just like the human brain. In the training process, how to reduce the energy consumption and the amount of time and data required while ensuring the ability of AI model? This is an important direction for AI to continue to develop. But now it seems that the way of accelerating deep learning training based on large-scale GPU parallel computing can not meet this condition.

A real intelligent system should be the natural intelligence of environment self adaptability. First of all, it can not only deal with the problem of certainty, but also deal with the problem of uncertainty. Second, it must not only be able to do things, but also be interpretable. Third, it is not completely driven by big data, even a small amount of data can achieve higher efficiency of continuous learning. Fourth, it should have high reliability, or be in line with the ethics set by human beings. This is our outlook for the next development stage of AI technology - ai3.0 era.

At present, we are at a turning point from ai2.0 to ai3.0. So, what is expected to become a blade penetrating the future of AI? At present, as a cutting-edge computing mode, neural mimicry computing is most likely to open up a new track from ai2.0 to ai3.0. Neural mimicry computing is an attempt and breakthrough in traditional semiconductor technology and chip architecture. By simulating the structure of human brain neurons and the mechanism of interconnection between neurons, it can continuously self-study under the condition of low power consumption and a small amount of training data, greatly improving the energy efficiency ratio. Obviously, the characteristics of neural mimicry are in line with the development of ai3.0. Therefore, neuromimetic computing is also expected to play an important role in the process of human entering the next generation of AI. Loihi, Intels neuromimetic computing chip, has the ability to smell. Pohoikisprings, a neuromimetic system, has the computing power of 100 million neurons, which is equivalent to the brain of a small mammal. Of course, neuromimetic computing is still in a very early stage. We still have a long way to go before we can really apply this technology to AI. But I believe that the innovation of the underlying technology must adhere to the long-term principle and focus on one direction and track for a long time, so as to fight against all the uncertainties in the development process with this kind of certainty, and finally achieve success. Source: Xue Jingyu, editor in charge of Netease Technology Report_ NBJS10393

At present, we are at a turning point from ai2.0 to ai3.0. So, what is expected to become a blade penetrating the future of AI? At present, as a cutting-edge computing mode, neural mimicry computing is most likely to open up a new track from ai2.0 to ai3.0. Neural mimicry computing is an attempt and breakthrough in traditional semiconductor technology and chip architecture. By simulating the structure of human brain neurons and the mechanism of interconnection between neurons, it can continuously self-study under the condition of low power consumption and a small amount of training data, greatly improving the energy efficiency ratio. Obviously, the characteristics of neural mimicry are in line with the development of ai3.0. Therefore, neuromimetic computing is also expected to play an important role in the process of human entering the next generation of AI.

Loihi, Intels neuromimetic computing chip, has the ability to smell. Pohoikisprings, a neuromimetic system, has the computing power of 100 million neurons, which is equivalent to the brain of a small mammal. Of course, neuromimetic computing is still in a very early stage. We still have a long way to go before we can really apply this technology to AI. But I believe that the innovation of the underlying technology must adhere to the long-term principle and focus on one direction and track for a long time, so as to fight against all the uncertainties in the development process with this kind of certainty, and finally achieve success.