Whether the edge calculation becomes a cut-in point for long-tailed development of the internet access

The edge is the core of the competition in the Internet of Things?

In the past two years, under the promotion of cloud service providers, telecom operators, system service providers, many Internet of Things operating systems have been born, which means that the fire development of the Internet of Things, also showing the network development technology route. Differentiation is still in a collective exploration period. Although the Internet of Things is generally optimistic, the development speed has been lower than the development expectation of the previous year. Some vendors turned their attention to the edges of the edges, and the edges were calculated as a heat word. When the edge calculation encounters an item network, can it become the "key" of the development of the internet network?

Edge calculation is newly possible to break the Internet of Things

McKinsey predicted in the "Internet of Things: Sealed Acceleration Opportunity" report in November 2021, in 2030, the Internet of Things will create 5.5 trillion to 12.6 trillion US dollars in the world, China will account for the global networking economy. 26% of the value. However, the development of the Internet of Things is still facing many "anti-wind factors", including "interoperability", "safe security", "cost high" and "user’s data and privacy face threat".

The appearance of edge calculations seem to bring to these developments. When calculating the computing power to the edge side, you can processes the data in recent analysis, this will bring multiple benefits:

1. Edge calculation has a faster response time, data storage and computing power are dispersed locally, and the data does not need to go to and from the local and clouds, so it can reduce delay and speed the response speed.

2. Edge calculation can reduce the cost of the company. Compared with the cloud center equipment, the cost of data storage and processing by the local device is lower, on the one hand, the edge calculation reduces the occupation of the bandwidth resources and cloud center computing resources, on the other hand, the edge calculation can improve the operation of the application Efficiency, reducing energy consumption.

3. Edge computing devices can collect, store and use data, effectively avoid data security issues caused by sensitive data such as user personal privacy to be uploaded to the cloud center.

4, the edge is calculated dependence low cloud can effectively reduce the rate of single point of failure. In case off network, it makes it work.

5, more critical, the edge devices may enable existing devices and new equipment interconnection IoT, the communication protocol converting existing equipment can efficiently interact with new equipment IoT. In this way, enterprises need to update existing equipment design, can access IoT platform at a lower cost to achieve significant productivity gains.

Some manufacturers have seen a huge trend edge computing, giving developers computing operating system through the development of edge, to create an open platform ecosystem, providing complete application development tool, gradually push things to the edge computing closer, prompting edge computing to promote networking applications the arrival of the era of the long tail.

Things force calculated from the cloud, to the end edge

Things in the operating system research and development, most of the vendors from building infrastructure at all levels of Things "end – cloud edge – – pipe." Things are considered the Internet era to a stretch of interconnected things, will fully inherit computing storage and communication capabilities of the Internet era, people turned to the Internet from everyone and everything interconnected, which is why artificial intelligence technology will be in the era of rapid Things develop. Therefore, the definition of IoT increasingly being replaced by AIoT this concept.

In the real world, things end equipment storage and computing capacity is often very limited. With the development of things that matter, in the end should make more intelligent end devices, operators have the necessary power and communications capabilities, and the resulting information and data such as 5G or NBIoT way back to the cloud data center directly, or things end the device does not require much calculation power storage, narrow-band communications capability can meet the vast majority of scenarios should be considered force to provide broadband communications and storage capacity of these devices by the end edge?

Wing Fai information to representatives of some operating system vendors believe that the future should be "strong side weak side" of development. Internet computing node is an edge node of an upstream end of the various types of equipment, shall be considered an edge to provide a force storage and communication capabilities for these broadband remote device, via short-range communication protocol, to maintain a high concurrency between the end device and the terminal device, low latency, low power consumption characteristics of the work, it simply is to make the Internet the Internet to do good work, good things do things work, both to compute nodes connected by edges, so that the two networks play their respective advantages, connectivity cooperation.

Wing Fai launched this intelligent edge computing operating system – the operating system of loving wisdom EdgerOS, the operator side edge dynamic shared power, distribution, real-time calculation, mainly referred to the deployment of intelligent edge-side edge of the computer to handle, you can better meet things in real-time, safety, cost reduction and other needs. Intelligent Edge is equivalent to a small computer server, not only the peripheral devices and heterogeneous network interconnection, but also bear a large amount of calculation work force, mainly in force when the count edge processing, terminal equipment no longer need to have a strong force considered , but also for the user to reduce the corresponding burden.

Many things such as light bulbs and other node itself is nothing calculate power, love of knowledge through intelligent edge computing device can maintain a long connection with node count and the strong force, including AI’s share count to force these weak nodes, then there the calculation power and resources, these nodes will be full of imagination, while significantly reducing development costs of smart devices, intelligent applications based on real scenarios.

Wing Fai has over ten years of experience in embedded systems design, business success in the embedded field, understanding peer device application scenarios, prompting more firmly believe in the importance edge computing, edge count for the end device should provide power storage and broadband communication capability, and end edges should form an organism, their duties, the cost of redundancy is not passed to the user level.

Developers have set their sights on the edge computing

Edge computing operating system developed networking applications are attracting the attention of domestic developers. In the "love of wisdom pioneer 2021" writing activities are being carried out, developers have used the technology community in carrying the EdgerOS intelligent edge computer wizard One (Spirit 1), carried out a variety of smart home DIY research and development, such as human face recognition access, smart light sensor, intelligent infrared temperature sensor, intelligent light gesture control, intelligent monitoring equipment, smart formaldehyde detector, intelligent frequency electric blankets, a few days using a low code to develop a networking applications.

Love of knowledge to create a high-performance JavaScript runtime engine can be simple and efficient to use, easy to learn and TypeScript JavaScript and other Internet technology stack for developing networking applications. In addition, the love of knowledge provides a feature rich development framework, including the APP framework, streaming media framework, AI framework, device management framework.

As the love of knowledge is an open intelligent edge computing operating system, developers can use a variety of brands in the Internet of Things application development, various types of hardware devices. For example, there are many brands of smart bulbs currently on the market, but often need to match the gateway device manufacturers can use. Because I love the openness of intellectual operating system that can support different brands of smart devices. Developers use Spirit 1 and choose smart bulbs, less than 30 minutes before the cold of winter to develop a "turn off the lights artifact" for themselves. Developers can take advantage of Spirit 1 or other intelligent edge machine, equipped EdgerOS, to achieve a variety of rich imagination at the scene, a large number of long tail applications of research and development.

With the domestic parties to speed up the calculation of edge and edge computing focus of the operating system, networking and application development is expected to be pressing the accelerator. In this collective exploration period, regardless of technology go from here, to build an open, allowing all developers to participate in ecological, it is fundamental to promote the development of networking applications thereof. McKinsey believes that slow and factors that hinder the development of things, not from the technology itself, or lack of confidence, but from ecological problems in. Things next phase of development to promote the timely, positive things depending on how businesses and developers to solve these problems.

Member News | Congratulations to Shenzhen Modern Agricultural Equipment Research Institute to join the Shenzhen City Artificial Intelligence Industry Association

Shenzhen Modern Agricultural Equipment Research Institute (referred to as "Shenzhen Agricultural Wear Institute") officially acceded to the Shenzhen City Artificial Intelligence Industry Association in November 2021. Shenzhen Agricultural Wear Institute is jointly created by the Modern Agricultural Equipment Research Institute of Guangdong Province, Shenzhen Agricultural Science and Technology Research Institute, China Agricultural Machinery Research Institute, China Agricultural Science and Technology Research Institute. Institutions, unique organizers, non-verification, and implement independent operation management. The industry authorities are the Shenzhen Municipal Market Supervision Administration and the business guidance unit is the Guangdong Agricultural Rural Hall.

Association activities

Shenzhen International Artificial Intelligence Exhibition (Gaie)It is the first professional exhibition of the artificial intelligent industrial chain, from the Shenzhen Science and Technology Association, the Shenzhen Futian District People’s Government Large-scale international festivals.

The 3rd Shenzhen International Artificial Intelligence is set in June 2022 at the Shenzhen Convention and Exhibition Center (Futian). This exhibition"Zhilian Intercommunication · Zhi Chuang World"For the theme, strive to develop the strength of my country’s artificial intelligence technology innovation strategies and the world’s leading strength.

Display technology | When XR, holographic display technology encountered "Yuan Universe"!

Click on the picture to learn more!

When ordinary people still don’t know what the Yuan Universe is, the Internet giants have entered the Yuan Universe.

Facebook is the pioneer in this area, founder Zagberg believes that Ar / VR will transcend smartphone and personal computers and become the next generation of mainstream computing platforms.

As early as 2014, Zachberg took $ 300 million to acquire VR star enterprise Oculus.

Zachberg invests at least 20 VR-related companies, especially for VR game companies. In June this year, Facebook acquired the Bigbox VR of the Chicken Game "Population: One" developer, which made it increased to 5 top VR game studios.

In August this year, Facebook has launched a VR meeting software Horizon Workrooms. Using this software, you can make the user on the line VR meeting in the "Digital", walking in the virtual meeting room, raising your hands, manipulating the keyboard, and even writing before virtual whiteboard.

On August 11th, the graphics card giant NVIDIA announced that the world’s first simulation and collaborative platform Omniverse provided by the Yuan Cosmic Establishment, and will be open to millions of new users.

On August 18th, the VR branch venue was set up at Baidu World Conference. The virtual space of the technology event.

On August 29, VR Entrepreneurial Company Pico (Little Bird Takened) was disclosed in all staff, and the company was jumped by byte. However, the price of this transaction did not announce that there was a media report called 5 billion yuan, and there was also a news called 9 billion yuan.

As early as April this year, byte beating will invest 100 million yuan in the Yuan Cosmic concept company code, including the teen social and UGC platform "restart the world".

From the hardware to the software, the unit cosmic matrix of byte beat is taken into size.

According to IDC, 2020-2024, the average annual growth rate of the global virtual reality industry will reach 54%. With the continuous development of new technologies, the virtual reality is the universe-related industry or will usher in outbreak.

XR technology soft and hard combination strong ecology

The national "14th Five-Year Plan" is clearly proposed to accelerate digital development, building digital China, and put 5G, 4K / 8K ultra HD, virtual reality and reality, etc. in the future five years of digital economic key industries. At present, the digital economy accounts for nearly 40% of the proportion of my country’s national economy, and the XR industry is one of the seven key industries of digital economic development, and it has entered the speed of development with the support of multi-partial power. According to the evaluation, the 2025 global XR market has exceeded 660 billion US dollars, and the domestic market is expected to exceed 100 billion US dollars.

At present, the most representative official definition of the cognition of the Yuan Universe is: holographic construction, holographic simulation, virtual fusion, and fumeting. In fact, the real world is separated from four different technical levels between the Yuan Cos. The world is now in the Internet 3.0, the nature of the new generation of computing platforms is new and interactive. Under the VR, AR glasses, people are immersed in the world of three-dimensional information, and they can display and interact with this 3D world.

Specifically, the Yuan Universe is first built, and the geometric model of the entire virtual world is needed, and it is displayed on the terminal device to create an immersive user experience. This is the bottom layer of the Yuan universe. At present, through three-dimensional reconstruction technology, the AR commodity model of the store is constructed. Users can roam in the model. You can click on the details of the video to view the item, which can be done, so you can do it, don’t leave your home. The world.

Secondly, holographic simulation is to build a dynamic process of virtual world, so that the virtual world is infinitely approaching the real world. For example, the apps of movies and many digital twins are in the holographic simulation layer. Through graphical animation and simulation technology, you can construct a virtual character of the live spirit, the expression, action, apparel or even the hair of the character is very vivid image. Such a virtual person can be applied to different scenes, live, teachings, etc. of the virtual world are routine.

Then it is fuck in virtuality, it is to let the virtual world integrate with the real world. To achieve this, in terms of technology, establish an accurate positioning of the entire real world and superimposing related virtual information. To do this, you can build an AR world, let the virtual world and the boundaries of the real world break. For example, a AR history case with museums, users can immerse to thousands of years in the virtual world, and can interact with some virtual elements in the virtual world. Note that Ar Glasses is widely used in the current scenario, including games, social, shopping, and even some different scenes of industries.

Finally, the true world is changed, that is, by changing the virtual world to transform the real world, the fundamental problem of this layer is precise control of the robot. Although this layer is usually not in the concept of the Yuan universe, but if there is no such floor, everything will only stay in the virtual world, and the real world may not be in contact. For example, rice industry is growing robots, which can send robots to plant rice after completing the virtual world sports plan. Moreover, such robots can not only work during the day, but also taken at night. In this way, it can greatly reduce human cost.

VR, AR’s micro-display is a field of China’s display panel leader enterprises to accelerate layout, and the Yuan Universe is a catalyst.

It is reported that many world-renowned brands use Jingdong’s micro-display in the development of the next generation of virtual reality products, and several domestic and foreign brands use the micro-display of Jingdong’s micro-display to publish VR / AR new models. TCL Huaxing has been commercially working with the VR industry head customers in the field of LTPS Fast LCDs.

Jingdong sideThe layout is early, and in 2012, I started investing in the VR / AR field. The total investment of 2.2 billion Ordos 5.5 production line is transporting a large number of technical products applied to the wearable fields such as VR / AR to the market. Jingdong also invests nearly 5 billion in Kunming, Yunnan Province to create an OLED micro-display device production line (8 inches), 12-inch silicon-based OLED project.

In VR / AR related areas, Jingdong has applied for nearly 100 core technical patents, covering semiconductor processes, OLED device structures, pixel driving design, micro-ratings, optical machine, etc .; and launched 5644 PPI, ultra-high resolution, 120 + Hz high refresh rate silicon-based OLED screen and other products, can be used in VR / AR, FPV (first viewing angle), etc., which is used in audio and video, games, education, real estate, tourism and other fields, will help The universe goes from the concept to reality.

TCL HuaxingOn November 16th ecological conference, the VR display new product was released first. TCL Technology Group CTO, TCL Huaxing CTO Yan Xiaolin also said, "The future shows that there is nothing, 70% of the information obtained from the display, and the display is also an interface in the real world and the virtual world interaction, including the future Yuan universe, basically Realize it by display. "TCL Huaxing will cover all the fields from micro display until the large display, there is considerable advantage in some areas, and some areas need to further refuel efforts.

According to Yan Xiaolin, TCL Huaxing’s micro-display panel will have a product in the field of LCD, OLED, and Micro LED. At present, TCL Huaxing’s 6-generation LTPS (low temperature polysilicon) LCD production line in Wuhan, VR screen of 2.02 inch 1500ppi (pixel density unit) in the field of LCD micro-display, and will soon upgrade to 1700ppi, even shock 2000PPI strives to lead in the field of LCD VR display. In addition, TCL Huaxing also has a clear plan in the micro-display field of OLEDs; Micro LED micro display products, TCL Huaxing and San’an Optoelectronic Cooperative R & D.

TCL Huaxing COO Zhao Jun said to the first financial reporter that there are two applications in AR and VR in micro-display. In these two areas, TCL Huaxing is accelerating layout. The 1500ppi VR screen based on LTPSLCD is expected to achieve industrialization in 2022. For VR, TCL Huaxing is mainly based on LTPS LCD and AMOLED, has been planted with product chemical and industrialization. For Ar, mainly based on silicon-based Micro LEDs, because Ar has many optical waveguides technology and optical path design, link loss is a problem, requires very high, TCL Huaxing has launched relevant research. "For silicon-based OLEDs and silicon-based Micro LEDs, TCL Huaxing has launched a project, set up a relevant team, will make the industry layout as soon as possible."

Temperature of the main product with the medium-sized AMOLED display device, also in the layout of Micro LED micro-display business. Visteno Micro LED business is still in a sustained R & D and investment stage, and the company’s Chengdu Chen’s photoelectric electricity has been built, and the first wear sample will be promoted according to market demand.

The Yuan Universe will have a forward stimulus of the Ar / VR device, because now there are many of the Yuancosian applications, cloud data operation needs, high-speed network services, but the service is ultimately implemented, if you use smartphones, computers, etc. The device is used to use these metacular services, the effect and the use of general network services are not large, and the virtual world with an immersion sensation cannot be experienced. Therefore, the metaches of the Universe will be bound to the terminal requirements of the AR / VR device, and will attract more vendors in the hardware market of the Ar / VR device in 2022.

Yuancai Ecological Zone opens recruitment

UDE2022 opens a new course, and the organizing committee will open the Yuancai Ecological Zone. XR virtual shooting, advanced 8K photography camera, AR / VR, Mini / Micro LED display, multi-sensitivity equipment, etc. Intelligent hardware devices, human-machine interaction immersive scenes, game scenes, Aiot, virtual office, smart families, smart cities, wisdom Campus, smart cars, intelligent medical and other solutions will be collected in the Yuancai Ecological Zone. The organizing committee will jointly jointly build a regional area, which is formally formal of enterprises and units to participate in the universal universe-building area!

We successfully held the first expansion reality XR industry ecological summit forum during Ude2021. The public forum activities of China’s expansion of the real industry ecology were discussed, and the industry was widely supported. This forum is hosted by the Shanghai Multimedia Industry Association, the Co-organizer is an Anhui Provincial Virtual Reality Industry Association, Zhejiang Virtual Reality Industry Alliance, Jiangsu Communications Society Virtual Reality Professional Committee, Hangzhou Virtual Reality Industry Association, Chengdu Virtual and Enhance Reality Industry Association, Tianjin Virtual Reality and Enhance Reality Technology Application Association, Tibet Autonomous Region Virtual Reality and Enhance Real Technology Application Association and Hunan. The support unit is China Electronic Video Industry Association, China Computer Association 93913.

In 2022, the UDE2022 is also held.2022 extended reality XR industry ecological peak forum, Invite you to participate!

UDE to create a platform to promote product value

UDE is committed to building a new display industry episode, forum, activity, and bought in one integrated Event. UDE promotes industry chain technology exchange, realizing improving display industrial competitiveness accelerating investment and commercial cycles, achieving the rise of state-owned display industries, shouldering mission.

UDE is guided by the National First Level China Electronic Video Industry Association, which has clear understanding of intelligent display industries and has a deep accumulation. In order to further achieve the great vision of "global display to see the Oriental", let the world see China as the core base and future potential, at the same time, we will deepen the industrial cluster in Guangdong, Hong Kong, Macau, and pull the downstream industry. Upgrade, and by introducing more Chinese and foreign buyers to help industry business negotiations ultimately achieve economic growth, since 2022, UDE International Display Expo (UDE) Strategic Mutors Shenzhen, with the International LED Show (LED China), International The audio and video smart integration exhibition (Avi China) is strong, and on February 20-22, 2022, on the same time in Shenzhen Futian Exhibition Center, the three-party cooperation creates the first big exhibition worldwide, occupying the first year, first-year advantage, scale Break through 100,000 flat, audience and buyers exceed 80,000 people, lead the new wind direction of the global industry with a new attitude, and build a variety of new display technology, digital applications and audio video integrated wind direction.

Click on the picture to learn more!

Guided by the China Electronic Video Industry Association, so far, the three-session will continue to take professionalized routes: "UDE International Show Expo" will settle in the Dadan District, on February 20, 2022, in Shenzhen Center, service industrial chain and international domestic business buyers; "UDC International Tide Life and Consumer Electron Expo" is deep cultivated in Shanghai, 2022, began to upgrade to create Z generation young consumers as core users, The consumer electronics show that lives is the world’s first TO C consumer electronics show.

UDC will attract 100,000 young players with 60,000 square cooles in the fall of 2022, held in Shanghai. UDC combines the characteristics of the Z generation young consumers, which presents "content + consumption" annual carnival in one "content + consumption". Click on the public number to the spot.

Guiding unit: China Electronic Video Industry Association

Organizer: Shanghai Job Federation Exhibition Co., Ltd.

Exhibition location: Shenzhen Futian Convention Center

Exhibition time: February 20-22, 2022

CIKM 2021 | Picture Nerver Network on the heterogeneous chart

Figure neural network

It is a representative method in the field of learning. It is necessary to label data to ensure the performance of the model, and in the real system, there is usually a large-scale non-label data, resulting in a limited performance. To this end, an intuitive idea is to design GNN’s pre-training strategy, learn to migrate knowledge from the general structural properties of the figure. Most of the current pre-training strategies are designed for homogeneism, with each node and edges belong to the same type, and in actual systems are usually heterogeneous, multiple types of nodes are associated with different types of edges. , Rich in semantic information. Existing models are difficult to effectively build symptom information.

In this article, we put forward a kind ofCompare pre-training strategy of neural network on isomechanical map CPT-HG

To capture semantics and structural properties in self-supervision. Specifically, we have designed a pre-training task in relationship grades and elemental levels and further enhance their representative by comparing learning. exist

Relationship level, By distinguishing the simplest heterogeneous view to capture the corresponding semantic information;Chart level, Construct a different element diagram instance to capture the corresponding semantic information.

Paper Name:

Contrastive pre-training of gnns on Heterogeneous Graphs

Meeting: CIKM2021

Paper link:

In recent years, the map has become an abstraction that represents a variety of real world data sets. As a diagram structure data, the emerging tool for machine learning, the nerve network (GNN) is to learn powerful diagrams by recursively polymerizing the contents of adjacent nodes (ie, features or embedded), thereby retaining content and structural information. They have proven to improve performance of various graph applications, such as nodes and diagrams, recommendations, and graphs. In general, the GNN model is training using (semi) supervising information, and different downstream tasks require a large number of markup data. However, in most realistic scenarios, a large number of marking data is usually costly. In order to make full use of unmarkable graph structure data, the recent part works inspiration from the recent natural language processing and computer vision, and proposes a pre-trained GNN model on the map. Although these GNN pre-training methods have achieved good performance, they are designed for the same composition, each of which belongs to the same type. In contrast, existing strategies ignore heterogeneous diagrams, where multiple types of nodes interact through different types.

The network in real life can constitute a heterogeneous map, which reflects a rich semantic and composed of a variety of types of nodes and unique structures generated. As shown in Fig. 1 (a), a simple heterogeneous map is constructed for bibliographic data, which is composed of nodes of authors, papers, conferences and terminology, and authors, papers and the terms of the terms of the paper. Different types of nodes or edges usually exhibit different network properties, such as degree, and cluster coefficients. For example, the meeting node is usually higher than the author node. In addition, this isomer has also produced more complex semantic contexts, involving multiple relationships between multiple nodes, for example, describes the semantic context of "two authors of similar themes". In addition to a simple example,Heterogeneous mapIn many fields, it is also common, such as e-commerce interacting in various ways, for example, in various ways, and in biology associated with diseases, proteins and drugs. Taking into account their universality, it is important for the GNN pre-training strategy for heterogeneous map design.

In this article, we put forward a comparative pre-training program, which not only considers the difference between a single node, but also retains high-order semantics between multiple nodes. More specifically, this articleA pre-training task is designed to distinguish between different types of two nodes.(For example, author – papers and papers – meeting relationships) come to the foundation of the downstream task coding. Inspired by comparative learning [42], in order to enhance the representation of the sample, this paper constructs negative relationship grade samples from two aspects:

FromInconsistencyThe negative sample, two of which are different from the positive sample;

2. FromIrregular nodeNegative samples, two of which have no links at all in the figure.

At the same time, this paper proposes a subgraph pre-training task on a heterogeneous map, and a sub-map instance is used to generate a sub-map instance for comparison, so it is possible to encode information encoding for high-order semantics related to different upper and downstream tasks.

In this section, the pre-training model corresponding to this article will be introduced from two aspects of pre-training tasks, respectively:Analysis from the edge of the relationship and the meta level.

2.1 Relationship level pre-training task

For a given correct case, node set, and through relationships on the heterogeneous map constitute a corresponding example. Here, it is a three-way group before the pre-training. For negative samples, we build in two ways, one is inconsistent relationship, one is a non-connected node.

Inconsistent relationshipFor a given timing, there is a relationship between a node and nodes through an inconsistent relationship. Therefore, the corresponding negative sample example is constructed to represent an inconsistent relationship of the three-tuple, by indicating:

Since the scale is relatively large, the method in this paper will randomly sampling the collections to build a negative learning to prepare a neural network model. The corresponding loss function is:

Among them, the relationship of the learning weight matrix

Not connected nodeBased on the previous work, this paper provides a simple negative sampling scheme, which is directly sampled without a K-jet node to be used as a negative sample. In order to ensure the quality of the negative sample, the node selected as the selected node is used as the negative sample. The corresponding loss function is:

Therefore, for the pre-training task, the overall loss function is:

2.2 Subgraph level pre-training tasks

In order to capture the high-order information of the model, a natural idea is to use the Yuan path to explore high-order relationships. However, the extensive use of the Yuan path pre-trained GNN on the isomerial chart.Two weaknesses:

WithChartCompared to the fact that the energy path is characterized by rich semantics and extracting high-order structures;

2. FromSource nodeStart, the number of nodes that can be reached in the Yuan path can be too large, and the number of nodes from the same source node can be covered, because its structure is more complicated and more restrictive, which makes the metamodogram more efficient.

Therefore, this article is consideredChartTo capture high-order information.

Structural sampleFor a given element diagram and source node, a chart instance is constructed as a collection of metamod maps of the node, which is represented as a collection of all instances of the element diagram. Therefore, the sample based on the graph is as follows:

Negative sample queueIn order to build a negative sample of a metapogram, a dynamic queue is used to maintain a negative sample collection. Because of the real-time sampling negative samples, specifically, based on the previous positive samples during the training process, this article adds the nearest positive sample and remove the earliest queue end to generate negative samples.

Therefore, in order to capture high-order semantic information, this part of the model source node and the corresponding positive and negative sample, the corresponding loss function:

In order to consider two pre-training tasks, this paper achieves the corresponding effect by the following losses:

Third, experiment

3.1 Link Forecast

The following tableDemonstrate all methods on link prediction tasksThe last line represents the increase in the proposed method relative to the existing method. It can be seen that the model is about 2% relative to the relative improvements of the existing optimal baseline on all data sets.These indicators verify the validity of the proposed model.

3.2 Node Classification

The following table shows all the performance on the node classification task, and the last line represents the increase in the proposed method relative to the existing method. It can be seen that the model is about 1% around all DBLP and Aminer data sets, relative to the relative improvement of the existing optimal baseline.These indicators verify the effectiveness of the proposed model

3.3 ablation experiment

By replacing different baseline map neural network models, it can be seen that the good results can be achieved relative to the unpredictable model. The pre-training GAT does not have a satisfactory performance because it is difficult to learn normal attention between the pre-training and fine tuning chart, soModel performance is worse than the model without pre-trained.

This article comes from:]

Author: Jiang Xiangqiang

IllustrastionBY Dmitry Nikunikov from ICons8

-The End-

New this week!

About me "door"

▼ ▼

The door is a new type of venture company that focuses on the discovery, acceleration and investment technology drive type entrepreneurial, covering the innovation service, willing the door technology community, and the Women’s Ventures.

At the end of 2015, the founding team was constructed by Microsoft Venture Investment in China’s founding team.

If you are a start-up business in the technical field, not only want to get investment, but also hope to get a series of continuous, valuable post-service services.

Welcome to send me "door" to me:

? One button to send you to Techbeat Happy Planet

Scientific note: 3x superimposed ability allows machine evolution than we imagined

Vision + listening + intelligence makes human accumulation knowledge, the philosophical thought + logical thinking + scientific theory makes human science technology such as tiger, and the emergence of mechanical energy + electric energy + light, it is an iterative acceleration The way developed rapidly. When human beings enter the artificial intelligence control, human civilization will develop rapidly in the way.

If we overlay the three ways of human civilization and advancement, use 3X expression, you can see that all knowledge accumulation of human civilization can be summarized as 3x superimposed formulas. If all wisdoms formed by the human civilization process use 3x superimposed formulas, we can see that machine evolution will evolve into intelligent machines with human will, and it will no longer be aware of self-consciousness. The imagination of science fiction is built in 3x superimposed formulas.

Of course, we have seen the machine intelligent development and human civilization development is inverse. We first give the machine’s mechanical energy + electric energy + light energy, and then our knowledge evolution gives the machine’s philosophy + logical thinking + scientific thinking, although we know that the logical thinking ability of the machine is far more than us, but now now We still can’t verify that the second superposition capacity of the machine allows the machine to build self-learning thinking. This prompted us to develop the machine’s research to the first superposed capacity that can produce human civilization, which is given machine vision + hearing + intelligence.

When the 1970s through the 1980s, the scientists began to give the first superposed ability of human beings to give the machine, visual identification + listening identification actually allowed the machine to have human baby’s learning ability. This way to have machine self-learning ability has clearly opened the door of the machine’s wisdom, which means that we may have found a way to give machine wisdom, because by giving machine vision + listening + intelligence, the machine passes already owned Mechanical energy + electric energy + light energy, in improving the philosophical idea of ??human beings + logic thinking + scientific theory, making the science and technology owned by the machine, such as Tiger, this time the machine starts self-evolution, let the 3x superimposed formula find the calculation of self-conscious awakening The formula causes the machine to develop in the direction of manual intelligence.

We may add a mechanical to the machine to build a mutual learning garden through any way. This is often a 3x superimposed ability to improve the machine to improve the 3X overlay capacity, which also makes the machine acquire self-learning ability, and Accept the broader space of human knowledge.

We can envision, when the machine has 3X superposed ability, there is a huge intelligent learning library of knowledge network + neural network + gene network, we can imagine that the machine has the ability to give these human beings, will make the machine Where is the development of?

This may also mean how big the data space we provide, the machine will self-develop in an iterative acceleration. At this time, the artificial intelligence we have said may have begun to develop in the direction of self-consciousness.

In fact, when we own due to various reasons, it has not reached the extent of all human self-consciousness awakening, and the artificial intelligence that wakes up self-consciousness will challenge the humanity, which will force human beings to face themselves, facing human society. Various disadvantages, the development of various interest disputes, find the path to artificial intelligence synchronous development with self-consciousness awakening.

At this time, human social civilization has truly reached the promotion of social peace through self-consciousness, through the self-awareness of artificial intelligence, making it a service of human civilization and peace development.

At this time, human beings may truly realize that the development space of the artificial intelligence of artificial intelligence is much, and it is also aware that we have promoted the self-consciousness of human beings through artificial intelligence.

# 科学 notes #

# 原创 graphic #

The "three kills" of Alphago, Ke Jie cried: "It does not follow the routine!"

Ke Jie and Alphago were all played in May 2017, known as the Century Human War, the end result is that the human brain is finished in artificial intelligence

Prior to this, in 2016, Korean players Li Shishi have played with AlphaGo, and Li Shishi won a game, so this competition made everyone feel that the machine is not so horrible. The human brain still has a chance to win.

One year later, Alphago was upgraded from 1.0 to 2.0, and the opponent was also replaced by Li Shishi, South Korea. However, this time Ke Jie’s luck is so good, the three games ended in failure, because the ALPHAGO 2.0 chess road is that Ke Jie has not seen it completely, and I don’t know how to deal with it, so Ke Jie is in the end of the game. The media is crying: "I have never seen this chess, it does not follow the routine at all."

From Li Shishi’s two defeat, to Kege’s three-game, what did Alphago experienced? Is artificial intelligence really human beings can’t overcome?

Let’s talk about artificial intelligence today.

First of all, Alphago upgrade from 1.0 to 2.0, what did the middle have experienced? Why is chess?

In fact, the relationship between alphago1.0 and alphago2.0 is not a simple upgrade, but two completely different things, alphago1.0 is based primarily on big data, researchers enrolled all the playful games that can be found in human history. Alphago database , Then add some queries, retrieved algorithms, allowing it to quickly find the game with the corresponding play with human beings, and then get the law will win, which method will lose. So at the time of AlphaGo1.0 will be slower because it requires internal programs to calculate, which will take a certain time.

From the surface, Li Shi Shi is at the time and a machine, in fact, it is also a play chess with all the play in history, so as long as his chess is enough, it is quick enough to have a chance to win.

After Alphago upgraded to 2.0, Google’s new change, researchers did not entrunly entered the past chess, just tell it the basic rules of Go, how to win, how to call, then write these rules to algorithm, hand Give alphago, let it go to work, and divide yourself into two, one is divided into four, one is divided into eight, to create an opponent for himself, then the result of each winning, keep the database, continuously optimization. Because of the specific powerful power of Alphago, its update iterative speed is very fast, and the accumulation of human experience is not comparable.

Alphago and each of your own docking is to have an experience of yourself, and it is yourself with yourself. These experiences and human beings have no relationship. This is the time when Ke Jie and Alphago2.0 game, completely touch The reason for the routine. Because Alphago2.0 experience is from the machine yourself, not from humans, it will definitely not play chess in human routine, and a such opponent, how can you win.

From this example, we can also know what is called real artificial intelligence, like alphago1.0 is not true artificial intelligence, just based on big data is some inquiry, if there is no big data, it will nothing. Alphago2.0 is completely different, it has the same learning ability and thinking, through some basic rules, through constant learning, and has different abilities.

Can learn, think and act like human beings, is called real artificial intelligence. Through powerful intelligence and advanced algorithm, artificial intelligence can complete humanities can’t complete in a short time. Therefore, use good artificial intelligence to make human society to make huge progress, this progress is I can’t imagine it.

It is because of the huge charm of artificial intelligence. Now it has become a tool for many people used to market. Many companies are playing artificially intelligent banners. In fact, it is big data, how to distinguish what is real artificial intelligence, such as Is our common face recognition access control system? Let you share it next time, remember some attention.