Figure what are the uses and applications of neural networks?

in recent years , Figure neural network can be said to be a new technology in the tuyere of , I believe that we all know more or less about graph neural network . Today, this article mainly talks about the powerful ability of graph neural network , For you to analyze its use and application . Generally speaking , Fig. the use of neural network can be summarized as three points , That is, end-to-end learning 、 Reasoning method and semantic visualization ability of intelligent computing . The application of graph neural network mainly focuses on computer vision 、 natural language processing 、 Biomedical 、 Industrial recommendation and industrial risk control .

Figure what are the uses of neural networks ?

1、 End to end learning

In recent years , Deep learning brings face recognition 、 The successful application of voice assistant and machinetranslation . The three scenarios represent three types of data ： Images 、 Voice and text . The key to the breakthrough of deep learning in these three scenarios is the end-to-end learning mechanism behind it . End to end represents high efficiency , It can effectively reduce the information asymmetry in the intermediate links , Once a problem is found at the terminal , Each link of the whole system can be linked and adjusted .

Since end-to-end learning in images 、 Learning on voice and text data is so effective , Then extend the learning mechanism to graph data with a broader business scenario , It's just a natural idea . If AI To achieve the same human ability , Composition generalization must be a top priority , Structured representation and computation are the key to achieving this goal . Just as in biology, innate factors and acquired factors work together , Combine the advantages of both , Benefit from their complementary advantages .

2、 Reasoning method of intelligent computing

The industry believes that large-scale graph neural network is a powerful reasoning method for cognitive intelligent computing . Graph neural network extends deep neural network from processing traditional unstructured data to higher-level structured data . Large scale graph data can express human common sense and expert rules with rich and logical relations , Graph nodes define understandable symbolic knowledge , The topological structure of irregular graph expresses the dependence of graph nodes 、 Subordinate 、 Logical rules and other reasoning relationships .

Take insurance and financial risk assessment , A complete AI The system doesn't just need to be based on a personal resume 、 Behavior habits 、 Analyze and deal with the health level, etc , And through their relatives and friends 、 colleagues 、 Further credit evaluation and inference shall be conducted for the data and mutual evaluation among students . The learning system based on graph structure can make use of 、 Interaction between users and products , Make very accurate causal and relational reasoning .

3、 Semantic visualization capability

Graph has strong semantic visualization ability , This advantage is shared by all GNN Shared by the model . For example, in the scenario of abnormal transaction account identification ,GNN After judging an account as an abnormal account , The partial subgraph of the account can be visualized . We can find some abnormal patterns , For example, there are multiple accounts on the same device , Or the same account acts on multiple devices . It can also be seen from the dimension of characteristics , For example, this account is very similar to other related account behavior patterns , So as to explain the judgment of the model .

Figure applications of neural networks ?

1、 natural language processing

GNNs There are also many applications in naturallanguageprocessing , Including multi hop reading 、 Entity recognition 、 Relation extraction and text classification . Multi hop reading means that there are many corpora for the machine , Open reading comprehension that allows machines to carry out Multi Chain reasoning , Then answer a complicated question . stay 2019 year , Top papers on naturallanguageprocessing use GNN As a reasoning module, it is already standard .

2、 Computer vision

In the application of computer vision, images are generated according to the provided semantics . Input is a semantic graph ,GNN Through to “man behind boy on patio” and “man right of man throwing firsbee” Understanding of two semantics , The output image is generated . Let's talk about visual reasoning , Human processing of visual information is often mixed with reasoning . Human beings can infer from the dimension of space or semantics , And graph can well depict spatial and semantic information , So that computers can learn to behave like humans , Use this information for reasoning . And, of course, motion recognition , Visual Q & A and other applications , Let's not list one by one .

3、 Biomedical

We were all exposed to biochemistry in high school , Know that compounds are made up of atoms and chemical bonds , They are naturally a form of graph data , So graph neural network is widely used in biomedical field . Including the discovery of new drugs 、 Compound screening 、 Protein interaction site detection 、 And disease prediction . At present, the foreign countries include Yale 、 harvard , Many laboratories in China, such as Peking University and Tsinghua University, have studied the application of neural networks in medicine , And I believe this will be one of the most valuable applications of graph neural networks .

In addition to the above directions , There are also things like autonomous driving and VR The domain will use 3D Point cloud ; Combined with the knowledge map which is also very popular in the past two years ; Traffic flow forecast in smart city ; Prediction of circuit characteristics in chip design ; You can even write code using graph neural networks . At present, it is actually applied in industrial scenes , There are two main scenarios that have achieved remarkable results , One is to recommend , Second, risk control .

4、 Industry recommendations

Recommendation is an important application of machine learning in the Internet . In Internet business , The recommended scenarios specifically say , Such as content recommendation 、 E-commerce recommends 、 Advertising recommendation, etc . here , We introduce three methods of graph neural network enabling recommendation .

(1) Interpretable recommendation

Interpretable recommendation , Is not only to predict the recommended products , Also give the reason for the recommendation . There is a concept in the recommendation called meta path . In the scenes recommended by the movie , As shown in the figure below . We use it U Represent user , use M It means film , that UUM Is a Metapath . It means that one user pays attention to another user , Then we can put the movies that users have seen , Recommended to those who care about him .

(2) Social network based recommendations

Take advantage of the concerns between users , We can also implement recommendations . Users' buying behavior will first be influenced by their friends in their online social circle . If the user A My friend is a sports fan , Frequently post about sports events 、 Sports stars and other information , user A It is likely that I will also learn about sports related information . At present, there are many e-commerce platforms , Including Jingdong 、 Mushroom street 、 Xiaohongshu and others are trying to make social based recommendations .

(3) Recommendation based on knowledge map

Items to recommend 、 Content or product , Based on existing attributes or business experience , You can get a lot of related information between them , These related information is what we usually call the knowledge map . Knowledge map can be naturally integrated into existing users - The commodity network constitutes a larger 、 And contain more information . In fact, whether it is social network recommendation , Or the knowledge map , They all add additional information to the graph network . It can aggregate the complex structural information in the relational network , It can also contain rich attribute information , This is where graph neural networks are powerful .

5、 Industrial risk control

Our company has had some time to use the chart for risk control . In our business scenario, there are many network requests every day , A request to come , It is necessary to judge whether this is a real user or a machine traffic in real time . A simple model , The data used includes equipment ID、IP、 Users and their behavior data . From principle 、 Algorithm 、 Realization 、 application 4 The three dimensions explain the graph neural network in detail . I hope it will be helpful for us to learn and use graph neural network technology .

Here we are. , The use and application of graph neural network , Have you got a general understanding ? In fact, there are many uses and applications of graph neural network , Due to the limitation of space, let's stop here first . All in all , Fig. the future of neural network is expected , Let's see !

Share some of my artificial intelligence learning materials for free , Including some AI Common framework actual combat video 、 Image recognition 、OpenCV、NLQ、 machine learning 、pytorch、 Computer vision 、 Videos such as deep learning and neural network 、 Courseware source code 、 Famous essence resources at home and abroad 、AI Hot papers 、 Industry reports, etc .

For better systematic learning AI, I recommend that you collect one .

Here are some screenshots , Free download method is attached at the end of the article .

One 、 Artificial intelligence courses and projects

Two 、 Famous essence resources at home and abroad

3、 ... and 、 Collection of papers on artificial intelligence

Four 、 AI Industry Report

Learn Artificial Intelligence well , Read more , Do more , practice , If you want to improve your level , We must learn to settle down and learn slowly and systematically , Only in the end can we gain something .

Click on the business card below , Scan the code and download the information for free .