Shulou(Shulou.com)06/02 Report--

With regard to the development level of artificial intelligence in today's scientific and technological circles, academia, industry and media may have different views. A saying I often hear is that artificial intelligence based on big data and deep learning is a completely new form of technology, and its emergence can comprehensively change the social form of mankind in the future, because it can "learn" on its own. as a result, a large number of human labor is replaced. I think there are two misunderstandings:

First, deep learning is not a new technology; second, the "learning" involved in deep learning technology is not the same thing as human learning, because it cannot really "deeply" understand the information it faces.

Deep learning is not a new technology.

From the perspective of the history of technology, the predecessor of deep learning technology is actually the "artificial neural network" technology (also known as "connectionism" technology), which has been popular for a while in the 1980s.

The essence of this technology is to build a simple artificial neural network structure by mathematical modeling, and a typical such structure generally includes three layers: input unit layer, middle unit layer and output unit layer. After the input unit layer obtains information from the outside world, according to the built-in convergence algorithm and excitation function of each unit, it "decides" whether to send further data information to the middle cell layer. the process is just as human neurons can "decide" whether to send electric pulses to other neurons according to the changes of the electric potential in their nuclei after receiving electrical pulses from other neurons.

It should be noted that no matter whether the overall task performed by the whole system is about image recognition or natural language processing, only from the operating state of a single computing unit in the system, the observer has no way to know the nature of the relevant overall task. Rather, the whole system decomposes the recognition task at the macro level into the micro information transmission activities between the components of the system in the way of "breaking up into parts". And through the general trend reflected by these micro information transmission activities, to simulate the information processing process of the human mind at the symbolic level.

The basic method for engineers to adjust the trend of micro-information transmission activities of the system is as follows: first, let the system randomly process the input information, and then compare the processing results with the ideal processing results. If the two do not match well, the system triggers its own "back propagation algorithm" to adjust the connection weight between the various computing units in the system, so that the output given by the system is different from the previous output. The greater the connection weight between the two units, the more likely the phenomenon of "co-excitation" will occur between them, and vice versa. Then, the system compares the actual output with the ideal output again, and if the two still do not match well, the system starts the back propagation algorithm again until the actual output and the ideal output coincide with each other.

The system that completes this training process can not only accurately classify the training samples, but also relatively accurately classify the input information which is close to the training samples. For example, if a system has been trained to identify which photos in the existing photo library are Zhang San's face, then even a new Zhang San photo that has never entered the photo gallery can be quickly recognized as Zhang San's face by the system.

If the reader still doesn't understand the above technical description, you might as well use the following example to further understand the operation mechanism of artificial neural network technology. Suppose a foreigner who does not understand Chinese goes to Shaolin Temple to learn martial arts, how should the teaching activities between teachers and students be carried out? There are two situations: in the first case, there can be language communication between the two (foreigners understand Chinese or Shaolin Temple masters understand foreign languages), so that the master can teach his foreign apprentice directly by "giving the rules". This method of education may barely be compared to the number of rules-based artificial intelligence.

Another situation is that the language of the master and the apprentice do not understand each other at all. in this case, how can the students learn martial arts? The only way to do this is as follows: the apprentice first observes the master's movement, and then follows, and the master tells the apprentice through simple physical communication whether the action is right (for example, if it is right, the master smiles; if not, the master praises the apprentice). Furthermore, if the master affirms a certain action of the apprentice, the apprentice will remember the action and continue to learn; if it is not right, the apprentice will have to guess what is wrong and give a new action based on this guess. and continue to wait for the master's feedback until the master is finally satisfied. Obviously, the learning efficiency of this kind of martial arts is very low, because the apprentice will waste a lot of time in guessing what went wrong. But the word "guessing" is exactly the essence of the operation of the artificial neural network. In a nutshell, such an artificial intelligence system does not really know what the input information it gets means-in other words, the designer of the system cannot communicate with the system at the symbolic level, just as the master is unable to communicate with the apprentice in the previous example. The reason why the "inefficiency" of this inefficient learning can be tolerated by the computer is due to a great advantage of the computer over the natural person: the computer can make a large number of "wild guesses" in a very short physical time, and select a more correct solution. Once we see the mechanism in it, it is not difficult to find that the working principle of artificial neural network is actually very clumsy.

"Deep learning" should be "deep learning".

So why does "neural network technology" now have the successor of "deep learning"? What does this new name mean?

I have to admit that "deep learning" is a confusing term, because it will tempt many laymen to think that artificial intelligence systems can already understand their learning content as deeply as humans do. But the truth is: according to human "understanding" standards, such a system can not achieve the most superficial understanding of the original information.

In order to avoid such misunderstandings, the author is in favor of calling "deep learning" as "deep learning". Because the real meaning of the original English word "deeplearning" technology is to upgrade the traditional artificial neural network, that is, to increase the number of hidden unit layers. The advantage of this is that it can increase the fineness of the information processing mechanism of the whole system, so that more object features can be settled in more middle layers.

For example, in the deep learning system of face recognition, more intermediate levels can deal with the features at different abstract levels, such as primary pixels, color block edges, line combinations, facial features and so on. This kind of fine treatment can certainly improve the recognition ability of the whole system.

However, it should be noted that the mathematical complexity of the whole system and the diversity of data brought about by this kind of "depth" requirements will naturally put forward high requirements for computer hardware and the amount of data used for training. This also explains why deep learning technology is becoming more and more popular after the 21st century. It is the rapid development of hardware in the computer field in the past decade and the huge amount of data brought about by the popularity of the Internet. It provides a basic guarantee for deep learning technology to blossom on the ground.

But there are two bottlenecks that prevent neural networks-deep learning techniques from becoming more "intelligent":

First, once the system is trained to converge, then the learning ability of the system decreases, that is, the system cannot adjust the weight according to the new input. This is not our ultimate ideal. Our ideal is: assuming that due to the limitations of the training sample database, the network converges prematurely, then it can still revise the original input-output mapping relationship independently in the face of new samples. and this revision can take into account both the old history and the new data. But the existing technology cannot support this seemingly grand technological idea. What designers can do now is to return the historical knowledge of the system to zero, put new samples into the sample database, and then train from scratch. There is no doubt that we have once again seen the chilling "Sisyphus cycle".

Second, as the previous example shows us, in the process of neural network-deep learning pattern recognition, a lot of effort is spent on the feature extraction of the original sample. Obviously, the same original sample will have different feature extraction patterns among different designers, which in turn will lead to different neural networks-deep learning modeling directions. For human programmers, this is a good opportunity to show their creativity, but for the system itself, it deprives itself of the opportunity to carry out creative activities. Just imagine: can a neural network-deep learning structure designed in this way be able to observe the original samples, find appropriate feature extraction patterns, and design its own topological structure? It seems difficult because it seems to require a meta-structure behind the structure, which can give a reflective representation of the structure itself. At present, we are still confused about how this meta-structure should be programmed-because it is us who realize the function of this meta-structure. Disappointingly, although deep learning technology has these basic defects, the current mainstream artificial intelligence world has been "brainwashed", believing that deep learning technology is equal to the whole of artificial intelligence. A more flexible and general artificial intelligence technology based on small data obviously requires more effort. From a purely academic point of view, we are still far from this goal.

(the writer works in the School of philosophy of Fudan University.)

[recommended by editors]

China takes the lead in AI technology because it is good at math? American magazine points directly at the gap in mathematics education between China and the United States. Huawei Teng Technology Salon, the next stop is Guangzhou! 2020 China Petroleum and Petrochemical Enterprise Information Technology Exchange Conference informed MIT to issue "arithmetic" warning: deep learning is approaching the computing limit TIOBE July: r and other statistical languages become popular _ Technology Weekly issue 635

Http://blog.sina.com.cn/s/blog_cfa68e330102zvrf.html

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.