1, strengthen the learning field
Reinforcement learning is a method of learning through experiment and error, which is inspired by the process of human learning new skills. In a typical case of reinforcement learning, we ask the experimenter to take action by observing the current state to maximize the feedback results. Every time an action is performed, the experimenter will receive feedback from the environment, so it can judge whether the effect brought by this action is positive or negative.
2. Generate model domain
Through the collection of a large number of samples, the models generated by artificial intelligence have strong similarity. That is to say, if the training data is an image of a face, then the model obtained after training is also a composite image similar to a face.
Ian Goodfellow, a top expert in artificial intelligence, put forward two new ideas for us: one is a generator, which is responsible for synthesizing the input data into new content; The other is a discriminator, which is responsible for judging whether the content generated by the generator is true or false. In this way, the generator must repeatedly learn the synthesized content until the discriminator cannot distinguish the authenticity of the generator content.
3. Storage network field
If the artificial intelligence system wants to adapt to various environments like human beings, it must constantly master new skills and learn to apply them. It is difficult for traditional neural networks to meet these requirements. For example, if a neural network is trained to solve task B after completing the training of task A, the network model is no longer applicable to task A. ..
At present, there are some network structures that can make the model have different degrees of memory ability. Long-term and short-term memory networks can process and predict time series; Progressive neural network, which learns the lateral relationship between independent models and extracts the same features to complete new tasks.
4. Data learning field
For a long time, the mode of deep learning is that we need a lot of training data to achieve the best results. Without large-scale training data, the deep learning model will not achieve the best results. For example, when we use artificial intelligence system to solve the task of lack of data, all kinds of problems will appear at this time. There is a method called transfer learning, which is to transfer the trained model to a new task, so that the problem can be easily solved.
5. The field of simulation environment
If artificial intelligence system is to be applied to real life, then artificial intelligence must have the characteristics of applicability. Therefore, developing a digital environment to simulate the real physical world and behavior will provide us with an opportunity to test artificial intelligence. Training in these simulated environments can help us understand the learning principle of artificial intelligence system and how to improve the system, and also provide us with a model that can be applied to real environment.
6, the field of medical technology
At present, the vertical image algorithm and natural language processing technology can basically meet the needs of the medical industry. There are also many technical service providers in the market, such as Suntech Cloud, which provides intelligent medical imaging technology, Zhiweixin Branch, which develops artificial intelligence cell recognition medical diagnosis system, and Ruoshui Medical, Yitongtianxia, which provides intelligent auxiliary diagnosis service platform. Although intelligent medical care plays an important role in auxiliary diagnosis and treatment, disease prediction, medical image-assisted diagnosis and drug research and development. Due to the non-circulation of medical image data and electronic medical records between hospitals and the opaque cooperation between enterprises and hospitals, there is a contradiction between technology development and data supply.
7. In the field of education
Enterprises such as Iflytek and Chuxue Education have already begun to explore the application of artificial intelligence in the field of education. Through image recognition, you can correct test papers, identify topics and answer questions by machine; Pronunciation can be corrected and improved through speech recognition; Man-machine interaction can answer questions online. The combination of AI and education can improve the unbalanced distribution and high cost of teachers in the education industry to a certain extent, and provide more efficient learning methods for teachers and students from the tool level, but it cannot have a more substantial impact on the educational content.
8, the field of logistics management
By using intelligent search, reasoning planning, computer vision, intelligent robots and other technologies, the logistics industry has automatically transformed the processes of transportation, warehousing, distribution, loading and unloading, and basically realized unmanned operation. For example, using big data to plan the intelligent distribution of goods, optimize logistics supply, demand matching and the allocation of logistics resources. At present, most of the manpower in the logistics industry is distributed in the "last mile" distribution link, and JD.COM, Suning and Cainiao are scrambling to develop unmanned vehicles and drones to seize market opportunities.