Alipay (Hangzhou) Information Technology Co., Ltd. (Zhejiang, CN)

Methods equipment, systems, and methods to train graph neural networks. One example of this is to get a complete graph and then break it up into several subgraphs. Then, obtain the graph that is used as a training graph for neural net training. This requires obtaining the feature vectors for each node from the diagram of training; obtaining node fusion vectors of each current node in that training graph; determining loss functions on the basis of node labels and node fusions vectors in graphs used for training; and developing the graph neural network to improve the loss function by adjusting the parameters

The increasing amount of processed data has resulted in an increase in the size of graphs. For example social networks could have more than one billion users, and more than 100 billion relationships.If the users and the relationships are abstracted into graphs, the scale of the graph can be more than one billion nodes. The increasing graph scale causes difficulties in graph neural network training.

Graph neural network training methods are, therefore, beneficial.

One of the features of the current specification is that it provides graph neural network methods of training. The methods include the following The complete graph is created and the graph is split into several subgraphs according to apredetermined condition. A choice is made among the many subgraphs to create a training graph. The training graph then contains a node feature vector for each node of the graph. To obtain a nodefusion vector, which is the current node in a training graph that incorporates edge and neighboring features, propagation and aggregation are based on the node feature of each node and edges betweennodes. To build graph neural networks using a loss function, it is calculated using the label for the node and the fusion vectors from the training graph. A graph neural network is iteratively trained using the loss function.

In some embodiments, that the complete graph is divided into a plurality of subgraphs according to a predetermined condition includes the following: The complete graph is divided into the plurality of subgraphs according to the predeterminedcondition by using a community discovery algorithm.

In some embodiments the predetermined condition may include the following: A certain number of neighboring nodes of a subgraph and a quantity of edges included in the subgraph satisfy one of the conditions and the number of nodes that are included in the subgraphis less than or equal to a specified threshold value.

In certain embodiments, the at least one subgraph that will be a part of graph neural network training is selected from the variety of subgraphs. This includes: An updated subgraph for every one of the plurality of subgraphs is determined based on the subgraph and T-th-degree neighbors of the subgraph. at least one subgraph which is to be a participant in training for graph neural networks is selected from a plurality of subgraphs that have been updated.

One embodiment of a training graph created using the most subgraphs to be utilized in graph neural net training is the following: A union set of at least onesubgraph is utilized to generate the training graph.

A graph neural network training system is an additional element of the present specification. The systems include one of the following components: a first determining modulethat is configured to obtain a complete graph, and divide the graph into a plurality ofsubgraphs following a predetermined rule and a second determining module, configured to select at least one subgraph that is to be a part of graph neural network learning from the many subgraphs and create a training graph that is based on at least one subgraph that is to participate in graph neural network training and then obtain from the training graph, a node feature vector of each node in the graph of training; and A fusion module, which is designed to perform the propagation and aggregation process based on thenode feature vector of every node of the training graph , and on edges between nodes, to obtain a node fusion vector that is of each present node in the training graph and contains neighboring node features as well as edge features as well as a training module, configured to obtain a loss function that is based on a label for a node as well as the node fusion feature vector on the training graph and then repeatedly train the graph neural network using the loss function to construct the graph neural network.

Another element of the specifications is the training of graph neural networks equipment. It includes processors. It is built to run the graph neural network training methods mentioned above.

Another element of the specifications is the computer-readable storage media. Computer-related instructions are stored on the storage media. The storage media stores computer instructions. After that, a computer can perform the graph neuralnetwork training methods as explained above.

Click here to view the patent on USPTO website.


Get Patents with PatentPC

What is a patent?

Patents are issued by the government to safeguard the invention. It grants the inventor the exclusive rights to develop, use and market the idea. Society gains when new technologies are brought to market. Benefits can be realized in the direct sense, since it can allow individuals to achieve previously unattainable things. Or indirectly, due to the opportunities for economic growth (business expansion and job creation) which the invention provides.

Many pharmaceutical companies and researchers at universities seek patent protection for their work and research. A patent can cover the physical or abstract nature of a product or process or a method or composition of material new to the area. Patent protection has to be granted to any invention that is valuable unique, innovative, and not previously known to others in the same area.

Patents give inventors a chance to be recognized for commercially successful inventions. They provide a reason for inventors to come up with new ideas. Patents enable entrepreneurs and inventors to know that there is a good chance they will get a profit on their time, effort, and money invested in technology development. They can earn a living by their work.

Companies with the capacity to:

Secure your products and services

Improve the value, the popularity, and appeal of your products on the market;

Stand out and differentiate yourself and your product from the rest.

Get technical and business information.

Beware of the possibility of accidentally using third-party proprietary content or losing valuable data, original outputs, or another creative output.

Patents effectively transform the inventor’s information into a tradeable asset that opens up new possibilities for employment creation and business expansion by licensing or joint ventures.

Small companies that have patent protection are more attractive to investors involved in the commercialization and development of technology.

Patenting can result in new inventions and ideas. The information you create may be eligible for patent protection.

Patents can serve as an obstacle to unscrupulous third parties profiting from the invention’s success.

Patent-protected technology that is commercially profitable could be used for financing technology-related research and development (R&D) which will improve the chances for better technology in the future.

It is possible to use the intellectual property rights of your company to convince investors and lenders that your product has commercial potential. Sometimes, a powerful patent can lead to a variety of financing possibilities. Patents and other IP assets as collateral or security for financing. Investors can also see your patent assets to boost company valuation. Forbes and others have noted that each patent can increase the value of a company by anything from $500,000 to $1 million.

A well-crafted business plan is essential for start-ups. It must be built on IP and demonstrate the way your product or service stands out. In addition, investors will be impressed if you demonstrate that your IP rights are secure or in progress of being secure and that they align with your business strategy.

It is essential to keep an invention secret prior to filing for patent protection. Public disclosure of an invention before it is filed is often detrimental to its novelty and make it ineligible for patent protection. Pre-filing disclosures, such as for investors, test-marketing, or other business partners should be done only following the signing of a confidentiality agreement.

There are a variety of patents and knowing these is vital to safeguard your invention. Utility patents cover the development of new methods and machines. Design patents cover ornamental designs. Patents for utility are the best option and shield the owner from copies and competitors. Patents for utility are usually issued to improve or modify existing inventions. Patents issued under utility can be used to improve or modify existing inventions. For instance, a procedure patent will cover acts or methods for performing one specific thing, while chemical compositions will comprise a mixture of components.

What is the length of time a patent will last? Although utility patents are valid up to 20 years from the initial filing, they can be extended through delays at the patent office.

Are you interested in the patenting of your idea? Since patents are only granted for applicants who are first to file and you must start filing quickly. Call a patent attorney at PatentPC to patent your idea now!

When drafting a patent application, you should do a patent search, as the search will give you an insight into the other applicants’ concepts. This can help you limit the scope of your invention. Also, you can learn about the latest developments in your field of invention. You’ll have a better idea of what your idea should be, and you’ll be better prepared to write the patent application.

How to Search for Patents

Patent searches are the initial step to getting your patent. You can do a google patent search or do a USPTO search. After the patent application has been submitted, the product that is covered by the application can be described as patent-pending. you can locate the patent application on public pair. When the patent office is satisfied with your application, you’ll be able to conduct a patent number look to identify the patent granted. Your product is now patentable. In addition to the USPTO search engine, you can also utilize other search engines, such as espacenet as described below. A patent lawyer or patent attorney can advise you on the procedure. In the US, patents are granted by the US patent and trademark office or by the United States patent and trademark office, which also examines trademark applications.

Interested in finding more similar patents? Here are the steps:

1. Create a list of terms to describe your invention based on its purpose, composition, and application.

Begin by writing down a concise and precise description of your invention. Do not use generic terms like “device”, “process,” or “system”. Think about synonyms for the terms you initially chose. Next, take note of important technical terms as well as key words.

Use the questions below to help you determine key words or concepts.

  • What is the objective of the invention Is it a utilitarian device or an ornamental design?
  • Is invention a way to create something or perform a function? Or is it a thing or process?
  • What is the nature and purpose of the invention? What is the physical structure of the invention?
  • What’s the purpose of the invention
  • What are the technical terms and keywords that describe the nature of an invention? To assist you in finding the right terms, refer to an online dictionary of technical terms.

2. These terms will enable you to look up relevant Cooperative Patent Classifications at Classification Search Tool. To determine the best classification for your invention, go through the resulting classification’s class Schemes (class schedules). If you do not get results from the Classification Text Search, you might consider substituting your words that describe your invention with synonyms.

3. Go through the CPC Classification Definition for confirmation of the CPC classification you’ve found. The link to a CPC classification definition will be available when the classification you have selected is a blue square with a “D” on its left. CPC classification definitions can be used to determine the scope of the classification and therefore you’re sure to select the most appropriate. They may also provide search tips or other suggestions that could be helpful for further investigation.

4. The Patents Full-Text Database and the Image Database allow you to retrieve patent documents that include the CPC classification. By focusing on abstracts and representative drawings, you can narrow down your search to the relevant patent documents.

5. Take advantage of this list of most relevant patent publications to study each in detail for similarities to your idea. Pay close attention to the specification and claims. Refer to the applicant and patent examiner to obtain additional patents.

6. It is possible to find published patent applications that match the CPC classification you selected in Step 3. You can use the same search strategy as Step 4, narrowing your search results to the most relevant patent applications through the abstract as well as the illustrations on every page. Next, carefully examine the published patent applications with particular attention paid to the claims and the additional drawings.

7. Find additional US patent publications by keyword searching in PatFT or AppFT databases, searching for classification of non-U.S. patents as described below, and searching non-patent publications of inventions with web search engines. Here are a few examples:

  • Add keywords to your search. Keyword searches may turn up documents that are not well-categorized or have missed classifications during Step 2. For example, US patent examiners often supplement their classification searches with keyword searches. Think about the use of technical engineering terminology rather than everyday words.
  • Search for foreign patents using the CPC classification. Then, re-run the search using international patent office search engines such as Espacenet, the European Patent Office’s worldwide patent publication database of over 130 million patent publications. Other national databases include:
  • Search non-patent literature. Inventions can be made public in many non-patent publications. It is recommended that you search journals, books, websites, technical catalogs, conference proceedings, and other print and electronic publications.

To review your search, you can hire a registered patent attorney to assist. A preliminary search will help one better prepare to talk about their invention and other related inventions with a professional patent attorney. In addition, the attorney will not spend too much time or money on patenting basics.