Method and apparatus for training artificial intelligence (AI) / machine learning (ML) model in single entity

Abstract

Techniques for training artificial intelligence (AI) and machine learning (ML) models for a single entity in wireless communications are described herein. A network trains one or more types of bilateral AI / ML models and provides one or more types of the bilateral AI / ML models to one or more user equipment (UE) devices. A UE device requests a network to deploy a bilateral AI / ML model and receives one or more types of the bilateral AI / ML model from the network based on information contained in the request.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This disclosure claims priority benefit of U.S. Provisional Patent Application No. 63 / 502,424, filed May 16, 2023, the contents of which are incorporated by reference in their entirety. TECHNICAL FIELD

[0003] The present disclosure relates generally to wireless communications, and more specifically to training artificial intelligence (AI) and machine learning (ML) models in a single entity in wireless communications. BACKGROUND

[0004] Unless otherwise stated in this section, the methods described in this section are not prior art to the claims listed below, and nothing mentioned in this section is admitted to be prior art by virtue of its mention in this section.

[0005] In a communication system, such as a wireless communication according to the Third Generation Partnership Project (3GPP) standards, many functions at the user equipment (UE) side tend to have a counterpart at the network side, and vice versa. In the context of artificial intelligence (AI) and machine learning (ML), this can be referred to as bilateral AI / ML models, also known as autoencoders. For example, for a modulation function at the UE / network, there is a demodulation function at the network / UE, for a quantization function at the UE / network, there is a dequantization function at the network / UE, for a forward error correction (FEC) encoder at the UE / network, there is a decoder at the network / UE, for a signal shaping function at the UE / network, there is a de-shaping function at the network / UE, and vice versa. There are also some functions / applications that require complementary modules at both the UE and the network (e.g., channel state information (CSI) compression, denoising (or noise reduction), quantization, coding, error correction code, modulation, peak-to-average power ratio (PAPR) reduction, and image compression). In short, in bilateral AI / ML models, it is ideal to train both sides simultaneously so that the functions at one side are compatible with the corresponding functions at the other side.

[0006] In the context of CSI, CSI compression is a study item (SI) on the air interface in the 5th Generation (5G) New Radio (NR) Release-18 specification. In a bilateral AI / ML model based on autoencoders, the UE of the bilateral AI / ML (encoder / CSI construction) converts the CSI into a compressed representation, and the network (decoder / CSI reconstruction) reconstructs the CSI from its compressed representation.

[0007] With respect to training of AI / ML models in wireless communication systems, there can be several training phases at a single entity (e.g., a UE or a network node of the network). Initially, the required architecture of the encoder and the decoder of the bilateral AI / ML model needs to be designed. Then, both sides need to be trained through forward pass (FP) and backpropagation (BP). In FP, the encoder passes encoded information (e.g., latent vectors) to the decoder, which recovers the information. In BP, the reconstruction error and the gradient of the reconstruction error with respect to the parameters can be propagated through the encoder and the decoder to update the parameters. Finally, the performance of the bilateral AI / ML model as a whole needs to be verified.

[0008] In training type 1, which involves joint training at a single entity (whether at the UE side or the network side), the training entity (UE or network) trains the bilateral AI / ML model through a single training session and respective FP and BP loops at the training phase. Then, at the inference phase, the non-training entity will request the training entity to provide its corresponding part (e.g., the encoder of the UE vendor and the decoder of the network vendor), and the non-training entity will download its corresponding bilateral AI / ML model part to perform a certain task (e.g., CSI compression).

[0009] Training type 1 has both advantages and disadvantages. In terms of advantages, since both the encoder and the decoder are trained in a single entity, the performance can be guaranteed. Moreover, since both the encoder and the decoder are designed by a single entity, matching architectures can be used. Furthermore, differences between input types and output types, between quantizers and dequantizers, and between pre-processing and post-processing can be avoided, thus enabling alignment. Moreover, the efforts of development, training, retraining, fine-tuning, and monitoring of the AI / ML model can be concentrated in a single entity. However, in terms of disadvantages, the AI / ML model is often not optimized for other non-training entities. This can be particularly challenging if the network trains a single encoder and provides the AI / ML model to UEs with compact computation and storage budgets. Therefore, there is a need for a solution to train AI / ML models in wireless communications in a single entity. SUMMARY

[0010] The following summary is provided for purposes of illustration only and is not intended to be limiting in any way. That is, the following summary is intended to introduce novel and non- obvious concepts, highlights, benefits and advantages of the technology described herein. The selected embodiments will be further described in the following detailed description. Thus, the following summary is not intended to identify key or essential features of the claimed subject matter, nor is it used to determine or apportion the scope of the claimed subject matter.

[0011] The subject matter of the present disclosure is to propose solutions or schemes to address the problems described herein. More specifically, various schemes proposed in the present disclosure relate to single entity training of AI / ML models in wireless communications. It is believed that implementation of various proposed schemes can address or alleviate the problems described above.

[0012] In an aspect, a method can involve a processor of a device training one or more types of bilateral AI / ML models. The method can also involve the processor providing the one or more types of bilateral AI / ML models to one or more UE devices.

[0013] In another aspect, a method can involve a processor of a device requesting a network to deploy a bilateral AI / ML model. The method can also involve the processor receiving one or more types of bilateral AI / ML models from the network depending on information included in the request.

[0014] It is worth noting that although the description provided herein can be in the context of certain wireless access technologies, networks, and network topologies of wireless communications, such as Fifth Generation (5G) / New Radio (NR) mobile communications, the proposed concepts, schemes, and any variants / derivatives thereof can be implemented in other types of wireless access technologies, networks, and network topologies, such as, but not limited to, Evolved Packet System (EPS), Long-Term Evolution (LTE), LTE-Advanced, LTE-Advanced Pro, Internet-of-Things (IoT), Narrow Band Internet of Things (NB-IoT), Industrial Internet of Things (IIoT), vehicle-to-everything (V2X), and non-terrestrial network (NTN) communications. Therefore, the scope of the present disclosure is not limited to the examples described herein. BRIEF DESCRIPTION OF DRAWINGS

[0015] The accompanying drawings are included to provide a further understanding of the present disclosure, and are incorporated in and constitute a part of this disclosure. The drawings illustrate embodiments of the present disclosure and, together with the description, serve to explain the principles of the present disclosure. It will be appreciated that the drawings are not necessarily to scale, as some components can be shown exaggerated in scale or with a different aspect ratio, in order to illustrate the concepts of the present disclosure more clearly.

[0016] Figure 1 is an illustration of an example network environment in which various proposed schemes in accordance with the present disclosure can be implemented.

[0017] Figure 2 is an illustration of one example scenario according to the proposed solution of the present disclosure.

[0018] Figure 3 is an illustration of one example scenario according to the proposed solution of the present disclosure.

[0019] Figure 4 is an illustration of one example scenario according to the proposed solution of the present disclosure.

[0020] Figure 5 is a block diagram of one example communication system according to the proposed solution of the present disclosure.

[0021] Figure 6 is a flowchart of one example process according to the proposed solution of the present disclosure.

[0022] Figure 7 is a flowchart of one example process according to the proposed solution of the present disclosure. DETAILED DESCRIPTION

[0023] Detailed embodiments and implementations of the claimed subject matter are disclosed herein. It is understood, however, that the disclosed embodiments and implementations are merely examples of the claimed subject matter and can be carried out in various ways as can be desired in certain implementations. Thus, the present disclosure is not intended to be limited to the examples described herein but rather is to be accorded the widest scope consistent with the claims, to read which with the full scope of equivalents, while the present disclosure is presented in the above summary is provided merely for purposes of summarizing some aspects of the detailed description that follows and is not intended to limit the scope of the disclosure in any way. In fact, it will be apparent to those skilled in the art from examination of the specification that various modifications and variations can be made to the described embodiments and implementations of the present disclosure without departing from the spirit and scope of the disclosure. Additionally, it is contemplated that individuals skilled in the art will be able to devise their own implementation, which, while different from those described in the detailed description, embody the principles of the present disclosure and fall within the spirit and scope of the disclosure. Thus, the present disclosure is not intended to be limited to the implementations described herein, but is to be accorded the widest scope consistent with the claims, to read which with the full scope of equivalents, and customized to an individual implementation. In the following description, numerous specific details are set forth to provide a thorough understanding of the embodiments and implementations. However, it will be apparent to one skilled in the art that embodiments and implementations of the present disclosure can be practiced without some or all of these specific details. In other instances, well known process steps and / or structures have not been described in detail in order to avoid unnecessarily obscuring the description of the embodiments and implementations. It will further be appreciated that the embodiments and implementations of the present disclosure can be used in a variety of different applications.

[0024] SUMMARY

[0025] Implementations of the present disclosure relate to various techniques, methods, solutions, and / or solutions for single entity training of AI / ML models in wireless communications. According to the present disclosure, many possible solutions can be implemented individually or jointly. That is, although these possible solutions can be described separately below, two or more of them can be implemented in one or another combination.

[0026] Figure 1 An example network environment 100 in which various solutions of the present disclosure can be implemented is shown. Figures 2 to 7 Examples of implementing various proposed solutions in the network environment 100 are shown according to the present disclosure. The following description of various proposed solutions is made with reference to the Figures 1 to 7 provided.

[0027] Reference is made to Figure 1Part (A) of FIG. 1 illustrates an example of a network environment 100 in which a UE 110 can communicate wirelessly with a radio access network (RAN) 120 (e.g., a 5G NR mobile network or other type of network, such as a non-terrestrial network (NTN)). The UE 110 can communicate wirelessly with the RAN 120 through a terrestrial network node 125 (e.g., a base station, eNB, gNB, or transmit-and-receive point (TRP)) or a non-terrestrial network node 128 (e.g., a satellite), and the UE 110 can be within coverage of a cell 135 associated with the terrestrial network node 125 and / or the non-terrestrial network node 128. The RAN 120 can be part of a wireless network 130. In the network environment 100, the UE 110 and the wireless network 130 (through the terrestrial network node 125 and / or the non-terrestrial network node 128) can implement various schemes related to training AI / ML models with a single entity in wireless communications, as described below. Figure 1 Part (B) of FIG. 1 illustrates an example of a bilateral AI / ML model implemented at a UE (e.g., the UE 110) and a network (NW) (e.g., the terrestrial network node 125 and / or the non-terrestrial network node 128). Notably, although various proposed schemes, options, and methods can be described separately below, in practical applications, the proposed schemes, options, and methods can be implemented individually or jointly. That is, in some cases, each of one or more proposed schemes, options, and methods can be implemented individually or separately. In other cases, some or all of the proposed schemes, options, and methods can be implemented jointly.

[0028] In a heterogeneous wireless ecosystem, UEs provided by different vendors can have different computing and storage budgets, resulting in heterogeneous UE capabilities. Moreover, even UEs manufactured by the same vendor can vary in capabilities from device to device. On the other hand, network nodes can have different computing and storage budgets, although this can not be a major challenge because the network is expected to have access to a large amount of computing and storage resources. However, network-side training Type 1 can still present challenges. That is, if the network (e.g., the wireless network 130) trains an AI / ML model without considering the specifications of the UE, such a model can not be deployable in the inference phase. For example, a given AI / ML model can require excessive storage that the UE cannot afford. Moreover, the AI / ML model can require excessive computation that can result in significant and intolerable latency.

[0029] In view of the above, according to a first proposed solution of the present disclosure, the network can train and provide dedicated AI / ML models to the heterogeneous UE devices. For example, the network (e.g., wireless network 130) can train specific AI / ML models (alternatively referred to herein as “UE-dedicated AI / ML models”) for one or more types of UE devices. Thus, the feasibility of deployment can be guaranteed, and the inference latency can be guaranteed. Under the first proposed solution, the network can train multiple dedicated AI / ML models.

[0030] Figure 2 An example scenario 200 according to the first proposed solution of the present disclosure is shown. The scenario 200 can involve an example of using dedicated AI / ML models. Referring to Figure 2 , the network can train multiple dedicated AI / ML models. For example, the network (e.g., wireless network 130) can train and provide a dedicated AI / ML model 1 to a first group of UEs ( ), including UE devices of A and B series (or from vendors A and B). In addition, the network can train and provide a dedicated AI / ML model 2 to a second group of UEs ( ), including UE devices of C, D, E, and F series (or from vendors C, D, E, and F). In addition, the network can train and provide a dedicated AI / ML model 3 to a third group of UEs ( ), including UE devices of G series (or from vendor G).

[0031] According to a second proposed solution of the present disclosure, non-dedicated artificial intelligence (AI) / machine learning (ML) models can be trained and provided by the network to the heterogeneous user equipment (UE) devices. Under the proposed solution, the non-dedicated AI / ML models can be designed and trained by the network without targeting any specific type of UE devices. For example, the network (e.g., wireless network 130) can train a general AI / ML model to serve all UEs without considering the vendors, types, and / or capabilities of the UE devices. Thus, the feasibility of deployment and latency can not be guaranteed, and the general AI / ML model can only be suitable for a portion of the UE devices. Under the proposed solution, multiple non-dedicated models can be trained and provided to the heterogeneous UE devices. For example, to accommodate the heterogeneity of UE capabilities, the network can train a series of AI / ML models to cater to UEs with different capabilities (e.g., computation and storage budgets).

[0032] Figure 3 An example scenario 300 according to the second proposed solution of the present disclosure is shown. The scenario 300 can involve an example of training non-dedicated AI / ML models. Referring to Figure 3 , the network (e.g., wireless network 130) can train a series of non-dedicated AI / ML models (e.g., model 1, model 2, … model N). The various models can correspond to respective parameters, floating point operations, and architecture types. As Figure 3As shown, the non-dedicated AI / ML models Model 1, Model 2, … Model N can correspond to parameters P1, P2, … PN, respectively. N , floating point operations F1, F2, …, FN, and architecture types T1, T2, … TN, respectively. N N In addition, as shown in Figure 3 , if a certain model (e.g., Model 2) meets the specifications of a certain UE device (e.g., UE 110), such as storage P (satisfying P2≤P≤P1) and maximum expected floating point operations T (satisfying T2≤T≤T1), then Model 2 can be provided to the UE device.

[0033] According to a third suggested solution of the present disclosure, there can be different types of requests and responses regarding AI / ML model deployment. For example, for a general request and response, a UE device (e.g., UE 110) can not specify its required dedicated AI / ML model or its computation and storage budget. In response, a network (e.g., wireless network 130) can send a general AI / ML model or a low-complexity non-dedicated AI / ML model to the UE device. For another example, for a dedicated request and response, a UE device can specify its dedicated AI / ML model in its request to the network. In response, the network can send a dedicated AI / ML model to the UE device according to the request. For yet another example, for a non-dedicated request and response, a UE device can specify its own storage and computation budget in its request to the network. In response, the network can send a non-dedicated AI / ML model to the UE device that fits the UE device’s computation and storage budget.

[0034] According to a fourth suggested solution of the present disclosure, there can be different types of AI / ML models obtained from training Type 1 of the network. Figure 4 An example scenario 400 is shown according to the fourth suggested solution of the present disclosure. Scenario 400 can involve examples of different types of AI / ML models obtained from training Type 1. Referring to Figure 4 , a network (e.g., wireless network 130) can train and provide different types of AI / ML models, including, for example: (1) a general AI / ML model, (2) multiple dedicated AI / ML models, (3) multiple non-dedicated AI / ML models, and (4) a mix of dedicated and non-dedicated AI / ML models.

[0035] Example Implementation

[0036] Figure 5 ​An example communication system 500 is shown in accordance with one embodiment of the present disclosure, having at least one example device 510 and one example device 520. Each of the device 510 and the device 520 can perform various functions to implement the schemes, techniques, processes, and methods described herein related to CSI compression and decompression, including the various schemes described above with respect to the various proposed designs, concepts, schemes, systems, and methods, including the network environment 100, and the processes described below.

[0037] Each of the device 510 and the device 520 can be part of an electronic device, which can be a network device or a UE device (e.g., the UE 110) such as a portable or mobile device, a wearable device, a vehicle, a wireless communication device, or a computing device. For example, each of the device 510 and the device 520 can be implemented in a smartphone, a smartwatch, a personal digital assistant, an electronic control unit (ECU) in a vehicle, a digital camera, or a computing device such as a tablet, a notebook, or a notebook computer. Each of the device 510 and the device 520 can also be part of a machine type device, which can be an Internet of Things (IoT) device such as an immobile or fixed device, a home device, a roadside unit (RSU), a wired communication device, or a computing device. For example, each of the device 510 and the device 520 can be implemented in a smart thermostat, a smart refrigerator, a smart door lock, a wireless speaker, or a home control center. In the case of implementation in or as a network device, the device 510 and / or the device 520 can be implemented in an eNodeB in an LTE, LTE-Advanced, or LTE-Advanced Pro network, or a gNB or a TRP in a 5G, NR, or IoT network.

[0038] In certain embodiments, each of the device 510 and the device 520 can be implemented in the form of one or more integrated circuit (IC) chips, such as but not limited to one or more single-core processors, one or more multi-core processors, one or more complex-instruction-set-computing (CISC) processors, or one or more reduced-instruction-set-computing (RISC) processors. In the various schemes described above, each of the device 510 and the device 520 can be implemented in or as a network device or a UE. Each of the device 510 and the device 520 can include Figure 5At least some of the components shown in FIG. 5 (e.g., processor 512 and processor 522). Each of devices 510 and 520 can also include one or more other components unrelated to the suggestions of the present disclosure (e.g., an internal power supply, a display device, and / or a user interface device), and thus, for simplicity and brevity, these components of devices 510 and 520 are not shown in FIG. 5 nor described below. Figure 5

[0039] In an aspect, each of processor 512 and processor 522 can be implemented in the form of one or more single-core processors, one or more multi-core processors, or one or more CISC or RISC processors. That is, although the singular term “processor” is used herein to refer to processor 512 and processor 522, each of processor 512 and processor 522 can include multiple processors or, in other embodiments, a single processor, according to certain embodiments of the present disclosure. In another aspect, each of processor 512 and processor 522 can be implemented in the form of hardware (and, optionally, firmware), the electronic components including, for example and without limitation, one or more transistors, one or more diodes, one or more capacitors, one or more resistors, one or more inductors, one or more memristors, and / or one or more varactors, the components being configured and arranged to achieve a particular purpose according to the present disclosure. In other words, in at least certain embodiments, each of processor 512 and processor 522 is a special purpose machine, specifically designed, arranged, and configured to perform certain tasks, including those involving a single entity training an AI / ML model in wireless communications, according to various embodiments of the present disclosure.

[0040] ​In certain embodiments, the device 510 can also include a transceiver 516 coupled to the processor 512. The transceiver 516 can be capable of wirelessly transmitting and receiving data. In certain embodiments, the transceiver 516 can be capable of wirelessly communicating with different types of wireless networks of different radio access technologies (RATs). In certain embodiments, the transceiver 516 can be equipped with multiple antenna ports (not shown) (e.g., four antenna ports). That is, the transceiver 516 can be equipped with multiple transmit antennas and multiple receive antennas for multiple-input multiple-output (MIMO) wireless communications. In certain embodiments, the device 520 can also include a transceiver 526 coupled to the processor 522. The transceiver 526 can include a transceiver capable of wirelessly transmitting and receiving data. In certain embodiments, the transceiver 526 can be capable of wirelessly communicating with different types of UEs / wireless networks of different RATs. In certain embodiments, the transceiver 526 can be equipped with multiple antenna ports (not shown) (e.g., four antenna ports). That is, the transceiver 526 can be equipped with multiple transmit antennas and multiple receive antennas for MIMO wireless communications.

[0041] In certain embodiments, device 510 can also include a memory 514 coupled to processor 512 and accessible to processor 512 for storing data and instructions. In certain embodiments, device 520 can also include a memory 524 coupled to processor 522 and accessible to processor 522 for storing data and instructions. Each of memory 514 and memory 524 can include a type of random access memory (RAM), such as dynamic RAM (DRAM), static RAM (SRAM), thyristor RAM (T-RAM) and / or zero capacitor RAM (Z-RAM). Alternatively, each of memory 514 and memory 524 can include a type of read-only memory (ROM), such as mask ROM, programmable ROM (PROM), erasable programmable ROM (EPROM), and / or electrically erasable programmable ROM (EEPROM). Alternatively, or additionally, each of memory 514 and memory 524 can include a type of non-volatile random access memory (NVRAM), such as flash memory, solid-state memory, ferroelectric RAM (FeRAM), magnetoresistive RAM (MRAM) and / or phase change memory.

[0042] Each of device 510 and device 520 can be a communication entity that communicates using various suggested schemes in accordance with the present disclosure. For illustrative purposes and not by way of limitation, the functionality of device 510 as a UE device (e.g., UE 110) and device 520 as a network node (e.g., network node 125) in a network (e.g., network 130 as a 5G / NR mobile network) is described in the context of example processes 600 and 700.

[0043] Flow Description

[0044] Figure 6An example process 600 is shown in accordance with embodiments of the present disclosure. Process 600 can represent one aspect of various suggested designs, concepts, schemes, systems, and methods for a single entity training AI / ML models in wireless communications, whether partially or entirely, including related content described above. Process 600 can include one or more operations, actions, or functions as shown in one or more blocks. Although shown as discrete blocks, each block can be divided into more blocks, combined into fewer blocks, or removed entirely depending on the implementation of the process. Further, blocks / sub-blocks of each process can be performed in the order shown in each figure, or in a different order. Additionally, one or more blocks / sub-blocks of each process can be iteratively performed. Process 600 can be implemented by or in device 510 and / or device 520, as well as any variants of device 510 and / or device 520. For illustrative purposes only and without limitation of scope, each process is described in the following context as device 510 as a UE device (e.g., UE 110) and device 520 as part of a network (e.g., 5G / NR mobile network) of a communication entity, such as a network node or base station (e.g., terrestrial network node 125). Process 600 can begin with block 610.

[0045] At block 610, process 600 can involve processor 522 of device 520 (e.g., as a terrestrial network node 125 or non-terrestrial network node 128 of wireless network 130) training one or more types of bilateral AI / ML models. Process 600 can proceed from block 610 to block 620.

[0046] At block 620, process 600 can involve processor 522 providing, via transceiver 526, one or more types of bilateral AI / ML models to one or more UE devices (e.g., including device 510).

[0047] In certain embodiments, in providing one or more types of bilateral AI / ML models, process 600 can involve processor 522 providing one or more types of bilateral AI / ML models to different UE devices having different computing and storage capabilities.

[0048] In certain embodiments, in training one or more types of bilateral AI / ML models, process 600 can involve processor 522 training one or more UE-specific AI / ML models, each model targeting one or more types of UE devices.

[0049] In certain embodiments, in training one or more types of bilateral AI / ML models, process 600 can involve processor 522 training one or more non-specific AI / ML models.

[0050] In certain embodiments, in training one or more non-dedicated AI / ML models, the process 600 can involve the processor 522 training a general AI / ML model to serve a plurality of UE devices without regard to the vendors, types, or capabilities of the plurality of UE devices.

[0051] In certain embodiments, in training one or more non-dedicated AI / ML models, the process 600 can involve the processor 522 training a plurality of non-dedicated AI / ML models to serve a plurality of UE devices without regard to any particular types of the UE devices. In certain embodiments, each of the plurality of non-dedicated AI / ML models can correspond to one or more respective parameters, respective floating point operations, and respective architecture types related to one or more of the plurality of UE devices.

[0052] In certain embodiments, in training one or more types of bilateral AI / ML models, the process 600 can involve the processor 522 performing certain operations. For example, the process 600 can involve the processor 522 training one or more UE-dedicated AI / ML models, each model for one or more types of UE devices. Further, the process 600 can involve the processor 522 training one or more non-dedicated AI / ML models. In certain embodiments, in training one or more non-dedicated AI / ML models, the process 600 can involve the processor 522 training a general AI / ML model to serve a plurality of UE devices without regard to the vendors, types, or capabilities of the plurality of UE devices. In certain embodiments, in training one or more non-dedicated AI / ML models, the process 600 can involve the processor 522 training a plurality of non-dedicated AI / ML models to serve a plurality of UE devices without regard to any particular types of the UE devices. In certain embodiments, each of the plurality of non-dedicated AI / ML models can correspond to one or more respective parameters, respective floating point operations, and respective architecture types related to one or more of the plurality of UE devices.

[0053] In certain embodiments, in providing one or more types of bilateral AI / ML models, the process 600 can involve the processor 522 performing certain operations. For example, the process 600 can involve the processor 522 receiving a request from a UE device of the one or more UE devices, the request not indicating a required dedicated AI / ML model or a computational and storage budget of the UE device. Further, the process 600 can involve the processor 522 providing a general AI / ML model or a low-complexity non-dedicated AI / ML model to the UE device.

[0054] Alternatively, in providing one or more types of bilateral AI / ML models, process 600 can involve the processor 522 performing certain operations. For example, process 600 can involve the processor 522 receiving a request from a UE device of the one or more UE devices, the request indicating a desired specialized AI / ML model. Further, process 600 can involve the processor 522 providing the UE device with a specialized AI / ML model.

[0055] Alternatively, in providing one or more types of bilateral AI / ML models, process 600 can involve the processor 522 performing certain operations. For example, process 600 can involve the processor 522 receiving a request from a UE device of the one or more UE devices, the request indicating a desired specialized AI / ML model. Further, process 600 can involve the processor 522 providing the UE device with a specialized AI / ML model.

[0056] Figure 7 According to one example process 700 is illustrated by embodiments of the present disclosure. Process 700 can be representative of an aspect, whether partial or overall, of implementing various designs, concepts, schemes, systems, and methods described above with respect to a single entity training AI / ML models in wireless communications, including related content described above. Process 700 can include one or more operations, actions, or functions as shown by one or more modules. Although shown as discrete modules, each of the processes can be divided into more modules, combined into fewer modules, or deleted in whole or in part depending on the implementation. Further, modules / sub-modules of each process can be executed in the order shown in each figure, or in a different order. Further, one or more of the modules / sub-modules of each process can be iteratively executed. Process 700 can be implemented by or in device 510 and / or device 520, and any variants of device 510 and / or device 520. For illustrative purposes only and without limitation of scope, each process is described in the following context as device 510 as a UE device (e.g., UE 110) and device 520 as a network’s communication entity, such as a network node or base station (e.g., terrestrial network node 125). Process 700 can begin with module 710.

[0057] At 710, process 700 involves the processor 512 (e.g., UE 110) of device 510 requesting, via transceiver 516, a network (e.g., by device 520 as a terrestrial network node 125 or non-terrestrial network node 128) to deploy a bilateral AI / ML model. Process 700 can proceed from 710 to 720.

[0058] At 720, the process 700 can involve the processor 512 receiving, from the network via the transceiver 516, one or more types of bilateral AI / ML models, depending on the information included in the request.

[0059] In certain embodiments, upon receiving one or more types of bilateral AI / ML models, the process 700 can involve the processor 512 receiving a general AI / ML model or a low-complexity non- specific AI / ML model in response to a request that does not indicate a required specific AI / ML model or a computing and storage budget of the UE device.

[0060] Alternatively, upon receiving one or more types of bilateral AI / ML models, the process 700 can involve the processor 512 receiving a specific AI / ML model specific to the UE device in response to a request that indicates a required specific AI / ML model.

[0061] Still alternatively, upon receiving one or more types of bilateral AI / ML models, the process 700 can involve the processor 512 receiving a non-specific AI / ML model relative to the UE device in response to a request that indicates a computing and storage budget of the UE device. In certain embodiments, the non-specific AI / ML model can fit within the computing and storage budget of the UE device.

[0062] Additional Description

[0063] The subject matter described herein is sometimes illustrated using different components contained within, or in connection with, different other components. It should be understood that the depicted architectures are merely examples, and that in fact many other architectures can be implemented to achieve the same functionality. From a conceptual standpoint, any arrangement of components to achieve the same functionality is merely "associated" so that the desired functionality is achieved. Hence, any two components herein combined to achieve a particular functionality can be seen as "associated with" each other such that the desired functionality is achieved, irrespective of architectures or intermediate components. Likewise, any two components so associated can also be viewed as being "operably connected", or "operably coupled", to each other to achieve the desired functionality, and any two components capable of being so associated can also be viewed as being "operably couplable", to each other to achieve the desired functionality. Specific examples of operably couplable include but are not limited to physically connectable and / or physically interacting components and / or wirelessly interactable and / or wirelessly interacting components and / or logically interacting and / or logically interactable components.

[0064] Further, with respect to virtually any plural and / or singular terms herein that are presented with an "a" or "an" preceding the term, those having skill in the art can translate the "a" or "an" into "one", and / or "one or more" where appropriate. For clarity, various singular / plural permutations can be presented herein explicitly.

[0065] Further, those skilled in the art will appreciate that, in general, the terms used herein, especially in the appended claims, such as "comprising," "as comprising," "including," "as including," "having," "as having," "containing," "as containing," "including," "as including," and the like, are to be construed open-ended terms, i.e., to mean "including but not limited to," "comprising but not limited to," "having but not limited to," or "including the foregoing" or the like. Those skilled in the art will further appreciate that if a specific number of an introduced claim statement is intended, the intent will be clearly recited in the claim statement, and if no such recitation is present, no such intent exists. By way of example, to aid in understanding, the following appended claims can include the use of introductory phrases such as "at least one" and "one or more" to introduce claim statements. However, the use of these phrases is not to be construed as implying that the introduction of claim statements by indefinite articles "a" or "an" limits any particular claim containing such introduced claim statements to contain only one such statement, even if the same claim includes the introductory phrase "one or more" or "at least one" and an indefinite article such as "a" or "an," e.g., "a" and / or "an" should be interpreted to mean "at least one" or "one or more"; the same applies to definite articles used to introduce claim statements. Further, even if a specific number of introduced claim statements is explicitly recited, those skilled in the art will recognize that such recitation should be interpreted to mean at least the recited number, e.g., a simple recitation of "two recitations" without further modifiers means at least two recitations, or two or more recitations. Further, in cases where a convention similar to "at least one A, B, and C, etc." is used, in general such structure is understood by those skilled in the art to be a convention for the meaning of the convention, e.g., "a system having at least one A, B, and C" would include, but not be limited to, a system having only A, a system having only B, a system having only C, A and B together, A and C together, B and C together, and / or A, B, and C together, etc. In cases where a convention similar to "at least one A, B, or C, etc." is used, in general such structure is understood by those skilled in the art to be a convention for the meaning of the convention, e.g., "a system having at least one A, B, or C" would include, but not be limited to, a system having only A, a system having only B, a system having only C, A and B together, A and C together, B and C together, and / or A, B, and C together, etc. Those skilled in the art will further appreciate that almost any disjunctive word and / or phrase presenting two or more alternative terms, whether in descriptions, claims, or diagrams, should be understood to contemplate the possibilities of including one of the terms, either of the terms, or both terms. By way of example, the phrase "A or B" would be understood to include the possibilities of "A" or "B" or "A and B." From the foregoing, it will be appreciated that various embodiments of the disclosure have been described herein, and that modifications can be made to these embodiments without departing from the scope and spirit of the disclosure. Accordingly, the various embodiments disclosed herein are by way of example only, and not by way of limitation. The true scope and spirit of the disclosure are indicated by the following claims.

Claims

1. A method implemented at a network side, comprising: training, by a processor of a device, one or more types of a bilateral artificial intelligence / machine learning model; and providing, by the processor, one or more types of the bilateral artificial intelligence / machine learning model to one or more user devices.

2. The method of claim 1, wherein providing one or more types of the bilateral artificial intelligence / machine learning model comprises providing one or more types of the bilateral artificial intelligence / machine learning model to different user devices having different computing and storage capabilities.

3. The method of claim 1, wherein training one or more types of the bilateral artificial intelligence / machine learning model comprises training one or more UE-specific artificial intelligence / machine learning models for one or more types of user devices.

4. The method of claim 1, wherein training one or more types of the bilateral artificial intelligence / machine learning model comprises training one or more non-specific artificial intelligence / machine learning models.

5. The method of claim 4, wherein training the one or more non-specific artificial intelligence / machine learning models comprises training a general artificial intelligence / machine learning model to serve multiple user devices without regard to vendors, types, or capabilities of the multiple user devices.

6. The method of claim 4, wherein training the one or more non-specific artificial intelligence / machine learning models comprises training multiple non-specific artificial intelligence / machine learning models to serve multiple user devices without regard to any particular types of the user devices.

7. The method of claim 6, wherein each of the multiple non-specific artificial intelligence / machine learning models corresponds to one or more respective parameters, respective floating point operations, and respective architecture types associated with one or more of the multiple user devices.

8. The method of claim 1, wherein training one or more types of the bilateral artificial intelligence / machine learning model comprises: training one or more UE-specific artificial intelligence / machine learning models, each model for one or more types of user devices; and training one or more non-specific artificial intelligence / machine learning models.

9. The method of claim 8, wherein training one or more non-specific artificial intelligence / machine learning models comprises training a general artificial intelligence / machine learning model to serve multiple user devices without regard to vendors, types, or capabilities of the multiple user devices.

10. The method of claim 8, wherein training one or more non-specific artificial intelligence / machine learning models comprises training multiple non-specific artificial intelligence / machine learning models to serve multiple user devices without regard to any particular types of the user devices.

11. The method of claim 10, wherein each of the non-specific artificial intelligence / machine learning models corresponds to one or more respective parameters, respective floating point operations, and respective architecture types associated with one or more of the multiple user devices.

12. The method of claim 1, wherein providing one or more types of the bilateral artificial intelligence / machine learning model comprises: ​ receiving a request from a user device of the one or more user devices, the request not indicating a required specialized artificial intelligence / machine learning model or a computing and storage budget of the user device; and providing a general artificial intelligence / machine learning model or a low-complexity non-specialized artificial intelligence / machine learning model to the user device.

13. The method of claim 1, wherein providing one or more types of the bilateral artificial intelligence / machine learning model comprises: receiving a request from a user device of the one or more user devices, the request indicating a required specialized artificial intelligence / machine learning model; and providing a specialized artificial intelligence / machine learning model to the user device.

14. The method of claim 1, wherein providing one or more types of the bilateral artificial intelligence / machine learning model comprises: receiving a request from a user device of the one or more user devices, the request indicating a computing and storage budget of the user device; and providing a non-specialized artificial intelligence / machine learning model to the user device.

15. The method of claim 14, wherein the non-specialized artificial intelligence / machine learning model fits within the computing and storage budget of the user device.

16. A method implemented at a user device, comprising: requesting, by a processor of a device, a network to deploy a bilateral artificial intelligence / machine learning model; and receiving, by the processor from the network, one or more types of the bilateral artificial intelligence / machine learning model depending on information included in a request.

17. The method of claim 16, wherein receiving one or more types of the bilateral artificial intelligence / machine learning model comprises receiving a general artificial intelligence / machine learning model or a low-complexity non-specialized artificial intelligence / machine learning model in response to the request not indicating a required specialized artificial intelligence / machine learning model or a computing and storage budget of the user device.

18. The method of claim 16, wherein receiving one or more types of the bilateral artificial intelligence / machine learning model comprises receiving a specialized artificial intelligence / machine learning model specific to the user device in response to the request indicating a required specialized artificial intelligence / machine learning model.

19. The method of claim 16, wherein receiving one or more types of the bilateral artificial intelligence / machine learning model comprises receiving a non-specialized artificial intelligence / machine learning model relative to the user device in response to the request indicating a computing and storage budget of the user device.

20. The method of claim 19, wherein the non-specialized artificial intelligence / machine learning model fits within the computing and storage budget of the user device.

Images