1. Reinforcement Learning Domain
Reinforcement learning is a method of learning through experimentation and error that is inspired by the human process of learning new skills. In a typical example of reinforcement learning, we ask participants to take an action and maximize the feedback results by observing the current situation. Each time you perform an action, the experimenter receives feedback from the environment, so it can determine whether the action has a positive or negative effect.
2. Generating Model Fields
Through the collection of a large number of samples, AI generates models with strong similarity. That is, if the training data is an image of a human face, then the model obtained after training is also a synthetic image of a similar face.
Ian Goodfellow, a top AI expert, has proposed two new ideas for us: one is a generator, responsible for synthesizing new content from the input data; the other is a discriminator, responsible for determining 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 is unable to recognize the authenticity of the generator's content.
3. Storing Network Fields
To adapt to various environments as humans do, AI systems must constantly acquire new skills and learn to apply them. Traditional neural networks have difficulty meeting these requirements. For example, after a neural network is trained for task A, if it is trained to solve task B, the network model is no longer suitable for A.
There are a number of network structures that allow models with varying degrees of memory. Long- and short-term memory networks can process and predict time series; asymptotic neural networks learn horizontal relationships between independent models and extract ***same features that can accomplish new tasks.
4. The field of data learning
It has always been the case that deep learning models require large amounts of training data to achieve the best results. Without large-scale training data, deep learning models will not achieve the best results. For example, when we use AI systems to solve tasks that lack data, various problems arise. There is a method called transfer learning, which involves transferring a trained model to a new task so that the problem is easily solved.
5. The field of simulation environments
If AI systems are to be applied to real life, then AI must be characterized by applicability. Therefore, the development of digital environments that simulate the real physical world and behavior will provide us with the opportunity to test AI. Training in these simulated environments can help us to understand well the learning principles of AI systems and how to improve them, and also provide us with a model that can be applied to real environments.
6. Medical technology field
Currently, image algorithms and natural language processing technologies in the vertical field are basically able to meet the needs of the medical industry, and a lot of technology service providers have already appeared in the market, such as Suntech Cloudstar, which provides intelligent medical imaging technology; Zhiwei Xin Branch, which develops an artificial intelligence cell recognition medical diagnostic system; Ruoshui, which provides a platform of intelligent assisted diagnostic services; and Yixing, which provides a platform for statistical processing of medical data. medical treatment, and One World for statistical processing of medical data. Although intelligent medical care plays an important role in assisted diagnosis and treatment, disease prediction, medical image-assisted diagnosis, and drug development. Due to the lack of circulation of medical imaging data and electronic medical records between hospitals, the cooperation between enterprises and hospitals is not transparent, which creates a contradiction between technological development and data supply.
7. Education
Companies such as KDDI and School Education have begun to explore the application of AI in education. Through image recognition, it can be used to correct test papers, recognize questions, and answer questions by machines. Through voice recognition can correct and improve pronunciation; human-computer interaction can answer questions online.AI + education, can to a certain extent improve the distribution of teachers in the education industry as well as the cost of the problem, from the level of tools for teachers and students to provide a more efficient way of learning, but can not have a more substantial impact on the content of education.
8. Logistics management field
The logistics industry uses intelligent search, inference planning, computer vision, intelligent robots and other technologies to automate the process of distribution, loading and unloading, transportation, warehousing and other processes, which basically enables unmanned operations. For example, the use of big data for intelligent distribution planning of goods, optimizing logistics supply, demand matching, and allocation of logistics resources.