CN-122003682-A - Customized device network-on-a-network based on shared infrastructure network

Abstract

A method includes hosting, by a device, an underlying network shared by a plurality of applications. The method also includes receiving, by the device, a notification from a particular application of the plurality of applications. The notification indicates that the particular application is to use the underlying network. The method also includes retrieving, by the device, a subnet associated with the particular application in response to receiving the notification from the particular application. The method also includes generating, by the apparatus, a customized network for the particular application at least in part by inserting the subnet into the underlying network.

Inventors

• WANG JINGTAO
• WANG XUSONG
• Poetry Rain Hu
• Olivier Henry Gilbert Lacombe
• Michael. Christian Butler
• WANG MIAO
• Eugenio Joerg Marchiori
• YAN CHENGJI

Assignees

• 谷歌有限责任公司

Dates

Publication Date

20260508

Application Date

20240813

Priority Date

20230816

Claims (20)

1. 1. A method, comprising: Hosting, by a device, an underlying network shared by a plurality of applications; receiving, by the device, a notification from a particular application of the plurality of applications, wherein the notification indicates that the particular application is to use the underlying network; retrieving, by the device, a subnet associated with the particular application in response to receiving the notification from the particular application, and A customized network is generated by the apparatus for the particular application at least in part by inserting the subnet into the underlying network.
2. 2. The method of claim 1, further comprising, prior to inserting the subnetwork into the underlying network: determining by the device whether a digital signature associated with the subnet indicates that the subnet is compatible with the underlying network, Wherein the subnet is inserted into the base network in response to a determination that the digital signature indicates that the subnet is compatible with the base network.
3. 3. The method of claim 1, wherein the subnet is retrieved from a subnet store that includes subnets for different applications.
4. 4. The method of claim 1, wherein the sub-network corresponds to a neural network topology having associated weights tailored to the particular application.
5. 5. The method of claim 1, wherein inserting the subnet into the base network comprises: Adding parallel branches to a computational graph of the underlying network, and The subnetwork is inserted into the parallel branches.
6. 6. The method of claim 1, wherein inserting the subnet into the base network comprises: adding sequential branches to a computational graph of the underlying network, and The subnetwork is inserted into the sequential branch.
7. 7. The method of claim 1, wherein inserting the subnet into the base network comprises: Resizing an embedded lookup matrix associated with the underlying network, and The embedded lookup matrix is updated based on the subnet after the embedded lookup matrix is resized.
8. 8. The method of claim 1, wherein the network weights in the underlying network are applicable to each of the plurality of applications.
9. 9. The method of claim 1, wherein the network weights in the sub-network are applicable to the particular application, and wherein the network weights in the sub-network are not applicable to different applications of the plurality of applications.
10. 10. The method of claim 1, wherein the underlying network is isolated from servers and other devices.
11. 11. The method of claim 1, wherein receiving the notification from the particular application comprises receiving, by the apparatus, a call from the particular application to the underlying network.
12. 12. An apparatus, comprising: Memory, and A processor coupled to the memory, the processor configured to: hosting an underlying network shared by a plurality of applications; Receiving a notification from a particular application of the plurality of applications, wherein the notification indicates that the particular application is to use the underlying network; In response to receiving the notification from the particular application, retrieving a subnet associated with the particular application, and A customized network is generated for the particular application at least in part by inserting the subnetwork into the underlying network.
13. 13. The apparatus of claim 12, wherein to insert the subnet into the base network, the processor is configured to: Adding parallel branches to a computational graph of the underlying network, and The subnetwork is inserted into the parallel branches.
14. 14. The apparatus of claim 12, wherein to insert the subnet into the base network, the processor is configured to: adding sequential branches to a computational graph of the underlying network, and The subnetwork is inserted into the sequential branch.
15. 15. The apparatus of claim 12, wherein to insert the subnet into the base network, the processor is configured to: Resizing an embedded lookup matrix associated with the underlying network, and The embedded lookup matrix is updated based on the subnet after the embedded lookup matrix is resized.
16. 16. The apparatus of claim 12, wherein prior to inserting the subnet into the base network, the processor is configured to: Determining whether a digital signature associated with the subnet indicates that the subnet is compatible with the underlying network, Wherein the subnet is inserted into the base network in response to a determination that the digital signature indicates that the subnet is compatible with the base network.
17. 17. The apparatus of claim 12, wherein the base network is isolated from servers and other devices.
18. 18. The apparatus of claim 12, wherein the network weights in the sub-network are applicable to the particular application, and wherein the network weights in the sub-network are not applicable to different applications of the plurality of applications.
19. 19. A non-transitory computer-readable medium comprising instructions that, when executed by a processor within a device, cause the processor to perform operations comprising: hosting an underlying network shared by a plurality of applications; Receiving a notification from a particular application of the plurality of applications, wherein the notification indicates that the particular application is to use the underlying network; In response to receiving the notification from the particular application, retrieving a subnet associated with the particular application, and A customized network is generated for the particular application at least in part by inserting the subnetwork into the underlying network.
20. 20. The non-transitory computer-readable medium of claim 19, wherein prior to inserting the subnet into the base network, the operations comprise: determining by the device whether a digital signature associated with the subnet indicates that the subnet is compatible with the underlying network, Wherein the subnet is inserted into the base network in response to a determination that the digital signature indicates that the subnet is compatible with the base network.

Description

Customized device network-on-a-network based on shared infrastructure network Cross Reference to Related Applications The present application claims the benefit of U.S. provisional patent application No. 63/519,985 filed on 8/16 of 2023. The contents of this U.S. provisional patent application are incorporated herein by reference in their entirety. Background A large generated base model may be used by devices such as mobile phones to perform a number of different tasks and functions. As non-limiting examples, the large-scale generative base model may be used to perform open-domain question-answering tasks, open-domain text-to-image generation tasks, and the like. Compared to classical neural network models, large generative base models generally provide improved performance and possess unique capabilities, such as improved generalization, contextual learning capabilities, the ability to interact via natural interfaces (text, image, voice, video), and the like. Running a large generative base model on a device includes several benefits. For example, by running a large generative base model on a device, rather than using an online large generative base model, the large generative base model may function when the device is offline, model behavior may be personalized for the user of the device, and client data remains offline, which results in increased privacy. However, running a large generative base model on a device typically requires the device to have increased storage, memory, and computing resources. For example, if multiple applications on a device require separate large generative base models, storing multiple large generative base models on the device may occupy a large amount of storage space (e.g., occupy a large amount of memory). Additionally, a large amount of computing resources may be used to run a different large generative base model for each application, which in turn may utilize a large amount of power at the device. Disclosure of Invention According to the techniques described herein, a base model may be shared among multiple applications, and may have interfaces to enable each application to customize the base model for its own use. For example, the base model may be divided into a shared base neural network (hereinafter network) and a set of interfaces, where sub-neural networks (hereinafter sub-networks) may be inserted for updating, augmentation, and customization. To illustrate, the shared underlying network may include model weights shared by multiple applications. In order to customize the shared base network for a particular application of the plurality of applications, a particular subnet associated with the particular application may be inserted or plugged into the shared base network. The particular subnetwork may include application-specific network weights such that when inserted into the shared underlying network, the particular subnetwork changes the shared underlying network to a customized network for the particular application. In a first example embodiment, a method includes hosting, by an apparatus, an underlying network shared by a plurality of applications. The method also includes receiving, by the device, a notification from a particular application of the plurality of applications. The notification indicates that the particular application is to use the underlying network. The method also includes retrieving, by the device, a subnet associated with the particular application in response to receiving the notification from the particular application. The method also includes generating, by the apparatus, a customized network for the particular application at least in part by inserting the subnet into the underlying network. It should be appreciated that this approach ensures compatibility, security, authenticity and timeliness of both the underlying network and the subnetwork. In a second example, an apparatus includes a memory and a processor coupled to the memory. The processor is configured to host an underlying network shared by a plurality of applications. The processor is also configured to receive a notification from a particular application of the plurality of applications. The notification indicates that the particular application is to use the underlying network. The processor is further configured to retrieve a subnet associated with the particular application in response to receiving the notification from the particular application. The processor is further configured to generate a customized network for the particular application at least in part by inserting the subnet into the underlying network. In a third example, a non-transitory computer-readable medium includes instructions that, when executed by a processor within an apparatus, cause the processor to perform operations. The operations include hosting an underlying network shared by a plurality of applications. The operations also include receiving a notification from a particular application of