User Equipment (UE) Capabilities Report for Machine Learning Applications

Abstract

A method for wireless communication by a user equipment (UE) includes receiving a machine learning model from a base station. The UE reports its machine learning processing capability to the base station. The UE also transmits a gradient update or a weight update to the machine learning model to the base station. The base station transmits the machine learning model to a plurality of UEs. The base station receives a machine learning processing capability report from each of the plurality of UEs. The base station groups the plurality of UEs based on the machine learning processing capability reports to receive gradient updates to the machine learning model.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. patent application No. 17 / 496,650, filed on October 7, 2021, entitled “USER EQUIPMENT (UE) CAPABILITY REPORT FOR MACHINE LEARNING APPLICATIONS,” which claims the benefit of U.S. provisional patent application No. 63 / 090,141, filed on October 9, 2020, entitled “USER EQUIPMENT (UE) CAPABILITY REPORT FOR MACHINE LEARNING APPLICATIONS,” the disclosures of which are expressly incorporated herein by reference in their entirety.

[0003] public domain

[0004] Aspects of the present disclosure relate generally to wireless communications and, more particularly, to techniques and apparatus for user equipment (UE) capability reporting for machine learning applications.

[0005] background

[0006] Wireless communication systems are widely deployed to provide a variety of telecommunication services, such as telephony, video, data, messaging, and broadcasts. Typical wireless communication systems may employ multiple access technologies capable of supporting communication with multiple users by sharing available system resources (e.g., bandwidth, transmit power, etc.). Examples of such multiple access technologies include code division multiple access (CDMA) systems, time division multiple access (TDMA) systems, frequency division multiple access (FDMA) systems, orthogonal frequency division multiple access (OFDMA) systems, single-carrier frequency division multiple access (SC-FDMA) systems, time division synchronous code division multiple access (TD-SCDMA) systems, and long term evolution (LTE). LTE / LTE-Advanced is a set of enhancements to the Universal Mobile Telecommunications System (UMTS) mobile standard promulgated by the Third Generation Partnership Project (3GPP).

[0007] A wireless communication network may include several base stations (BSs) that can support communications for several user equipment (UEs). User equipment (UEs) can communicate with the base stations (BSs) via downlinks and uplinks. The downlink (or forward link) refers to the communication link from the BS to the UE, while the uplink (or reverse link) refers to the communication link from the UE to the BS. As will be described in more detail, a BS may be referred to as a Node B, gNB, access point (AP), radio head, transmit / receive point (TRP), new radio (NR) BS, 5GB node, etc.

[0008] The above multiple access technologies have been adopted in various telecommunication standards to provide a common protocol that enables different user equipment to communicate at a city, country, region, and even global level. New Radio (NR) (which may also be referred to as 5G) is a set of enhancements to the LTE mobile standard promulgated by the Third Generation Partnership Project (3GPP). NR is designed to better support mobile broadband Internet access by improving spectrum efficiency, reducing costs, improving services, utilizing new spectrum, and better integrating with orthogonal frequency division multiplexing (OFDM) with a cyclic prefix (CP) (CP-OFDM) on the downlink (DL), CP-OFDM and / or SC-FDM (e.g., also known as discrete Fourier transform spread OFDM (DFT-s-OFDM) on the uplink (UL), and other open standards that support beamforming, multiple-input multiple-output (MIMO) antenna technology, and carrier aggregation.

[0009] An artificial neural network may include groups of interconnected artificial neurons (e.g., a neuron model). An artificial neural network may be a computing device or represented as a method to be executed by a computing device. A convolutional neural network (such as a deep convolutional neural network) is a feedforward artificial neural network. A convolutional neural network may include layers of neurons that may be configured in a tiled receptive field. It would be desirable to apply neural network processing to wireless communications to achieve higher efficiency.

[0010] Overview

[0011] According to aspects of the present disclosure, a method for wireless communication receives a machine learning model from a base station, reports machine learning processing capabilities to the base station, and transmits gradient updates or weight updates to the machine learning model to the base station.

[0012] In other aspects of the present disclosure, a method of wireless communication transmits a machine learning model to a plurality of user equipment (UEs). The method receives a machine learning processing capability report from each UE. The method also groups the plurality of UEs based on the machine learning processing capability report to receive gradient updates to the machine learning model.

[0013] In other aspects of the present disclosure, an apparatus for wireless communication at a user equipment (UE) includes a processor and a memory coupled to the processor. Instructions stored in the memory, when executed by the processor, are operable to cause the apparatus to receive a machine learning model from a base station. The apparatus may report machine learning processing capabilities to the base station. The apparatus may also transmit gradient updates or weight updates to the machine learning model to the base station.

[0014] In other aspects of the present disclosure, an apparatus for wireless communication at a base station includes a processor and a memory coupled to the processor. Instructions stored in the memory, when executed by the processor, are operable to cause the apparatus to transmit a machine learning model to a plurality of user equipments (UEs). The apparatus may receive a machine learning processing capability report from each UE. The apparatus may also group the plurality of UEs based on the machine learning processing capability report to receive gradient updates to the machine learning model.

[0015] In other aspects of the present disclosure, a user equipment (UE) for wireless communication includes means for receiving a machine learning model from a base station. The UE includes means for reporting machine learning processing capabilities to the base station. The UE also includes means for transmitting gradient updates or weight updates to the machine learning model to the base station.

[0016] In yet other aspects of the present disclosure, a base station for wireless communication includes means for transmitting a machine learning model to a plurality of user equipments (UEs). The base station includes means for receiving a machine learning processing capability report from each of the plurality of UEs. The base station also includes means for grouping the plurality of UEs based on the machine learning processing capability reports to receive gradient updates to the machine learning model.

[0017] In yet another aspect of the present disclosure, a non-transitory computer-readable medium having program code recorded thereon is disclosed. The program code is executed by a user equipment (UE) and includes program code for receiving a machine learning model from a base station. The UE includes program code for reporting machine learning processing capabilities to the base station. The UE also includes program code for transmitting gradient updates or weight updates to the machine learning model to the base station.

[0018] In other aspects of the present disclosure, a non-transitory computer-readable medium having program code recorded thereon is disclosed. The program code is executed by a base station and includes program code for transmitting a machine learning model to a plurality of user equipments (UEs). The base station includes program code for receiving a machine learning processing capability report from each of the plurality of UEs. The base station also includes program code for grouping the plurality of UEs based on the machine learning processing capability report to receive gradient updates to the machine learning model.

[0019] Aspects generally include methods, apparatus, systems, computer program products, non-transitory computer-readable media, user equipment, base stations, wireless communication devices, and processing systems substantially as described with reference to and as illustrated in the accompanying drawings and description.

[0020] The foregoing has broadly outlined the features and technical advantages of examples according to the present disclosure in an effort to facilitate a better understanding of the detailed description that follows. Additional features and advantages will be described. The concepts and specific examples disclosed may be readily used as a basis for modifying or designing other structures for carrying out the same purposes as the present disclosure. Such equivalent constructions do not depart from the scope of the appended claims. The characteristics of the disclosed concepts, both in their organization and method of operation, and the associated advantages will be better understood by considering the following description in conjunction with the accompanying drawings. Each of the figures is provided for illustration and description purposes and is not intended to define limitations on the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to understand the features of the present disclosure in detail, reference may be made to various aspects for a more detailed description, some of which are illustrated in the accompanying drawings. However, it should be noted that the accompanying drawings illustrate only certain aspects of the present disclosure and are not to be considered limiting of its scope, as the description may admit to other equally effective aspects. The same reference numerals in different drawings may identify the same or similar elements.

[0023] Figure 1 is a block diagram conceptually illustrating an example of a wireless communication network in accordance with various aspects of the present disclosure.

[0024] Figure 2 is a block diagram conceptually illustrating an example of a base station in communication with a user equipment (UE) in a wireless communication network according to various aspects of the present disclosure.

[0025] Figure 3 An example implementation of designing a neural network using a system on a chip (SOC) including a general-purpose processor in accordance with certain aspects of the present disclosure is illustrated.

[0026] Figure 4A 、 4B 4C are diagrams illustrating neural networks according to aspects of the present disclosure.

[0027] Figure 4D is a diagram illustrating an exemplary deep convolutional network (DCN) according to aspects of the present disclosure.

[0028] Figure 5 is a block diagram illustrating an exemplary deep convolutional network (DCN) according to aspects of the present disclosure.

[0029] Figure 6 is a block diagram illustrating federated learning in accordance with aspects of the present disclosure.

[0030] Figure 7 is a timing diagram illustrating reporting machine learning capabilities according to aspects of the present disclosure.

[0031] Figure 8is a flow diagram illustrating an example process performed, for example, by a user equipment (UE) in accordance with various aspects of the present disclosure.

[0032] Figure 9 is a flow diagram illustrating an example process, eg, performed by a base station, in accordance with various aspects of the present disclosure.

[0033] Detailed description

[0034] The various aspects of the present disclosure are described more fully below with reference to the accompanying drawings. However, the present disclosure can be implemented in many different forms and should not be interpreted as being limited to any specific structure or function given throughout the present disclosure. On the contrary, these aspects are provided to make the present disclosure thorough and complete, and they will fully convey the scope of the present disclosure to those skilled in the art. Based on this teaching, those skilled in the art will appreciate that the scope of the present disclosure is intended to cover any aspect of the present disclosure, whether it is implemented independently of any other aspect of the present disclosure or implemented in combination. For example, any number of aspects described can be used to implement a device or practice method. In addition, the scope of the present disclosure is intended to cover such devices or methods that are practiced using a supplement to the various aspects of the present disclosure described or other other structures, functionality, or structure and functionality. It should be understood that any aspect of the disclosed disclosure can be implemented by one or more elements of the claims.

[0035] Several aspects of telecommunications systems will now be presented with reference to various devices and techniques. These devices and techniques are described in the following detailed description and illustrated in the accompanying drawings by various blocks, modules, components, circuits, steps, processes, algorithms, etc. (collectively referred to as "elements"). These elements can be implemented using hardware, software, or a combination thereof. Whether such elements are implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system.

[0036] It should be noted that while various aspects may be described using terminology generally associated with 5G and later generation wireless technologies, various aspects of the present disclosure may be applied in communication systems based on other generations, such as and including 3G and / or 4G technologies.

[0037] Standard machine learning approaches centralize the training data on a single machine or in a data center. Federated learning models support collaborative learning of a shared prediction model among user equipment (UE) and a base station (or a centralized server). Federated learning is a process in which a group of UEs receive a machine learning model from a base station and work together to train the model. More specifically, each UE trains the model locally and sends back updated neural network model weights or gradient updates based on, for example, a stochastic gradient descent process performed locally. The base station receives updates from all UEs in the group and aggregates the updates, for example by averaging them, to obtain updated global weights of the neural network. The base station sends the updated model to each UE, and the process repeats round after round until the desired performance level of the global model is achieved.

[0038] In each round of the federated learning process, a group of UEs sends back weight or gradient updates within a given time interval after they receive the model from the base station. If a UE misses the deadline for sending an update, the weights or gradients will become stale, and the base station will not incorporate the updates into the weight or gradient aggregation for that local training round of the federated learning process.

[0039] According to various aspects of the present disclosure, a UE reports its machine learning processing capabilities to a base station. In some aspects, the report may indicate the machine learning hardware capabilities. In other aspects, the report indicates the approximate turnaround time for calculating gradients or weight updates in each round of federated learning. In still other aspects of the present disclosure, the UE reports the approximate turnaround time for calculating gradients or weights, for example, as a function of the UE's battery status.

[0040] The reported machine learning hardware capabilities provide the base station with an approximate training time for the UE to prepare each gradient or weight update. For example, the base station can determine whether the reporting UE is a fast or slow UE based on the reported machine learning capabilities. Therefore, the base station can group UEs according to their machine learning capabilities for different rounds of joint learning. Slow UEs can be grouped with other slower UEs, while fast UEs can be grouped with other faster UEs, thereby improving the efficiency of the joint learning process.

[0041] Figure 11 is a diagram illustrating a network 100 in which various aspects of the present disclosure may be practiced. Network 100 may be a 5G or NR network or some other wireless network, such as an LTE network. Wireless network 100 may include several BSs 110 (shown as BS 110a, BS 110b, BS 110c, and BS 110d) and other network entities. A BS is an entity that communicates with user equipment (UE) and may also be referred to as a base station, NR BS, Node B, gNB, 5G Node B (NB), access point, transmit reception point (TRP), etc. Each BS may provide communication coverage for a specific geographic area. In 3GPP, the term "cell" may refer to the coverage area of ​​a BS and / or the BS subsystem serving that coverage area, depending on the context in which the term is used.

[0042] A BS may provide communication coverage for a macro cell, a pico cell, a femto cell, and / or another type of cell. A macro cell may cover a relatively large geographic area (e.g., several kilometers in radius) and may allow unrestricted access by UEs with service subscriptions. A pico cell may cover a relatively small geographic area and may allow unrestricted access by UEs with service subscriptions. A femto cell may cover a relatively small geographic area (e.g., a residence) and may allow restricted access by UEs associated with the femto cell (e.g., UEs in a closed subscriber group (CSG)). A BS for a macro cell may be referred to as a macro BS. A BS for a pico cell may be referred to as a pico BS. A BS for a femto cell may be referred to as a femto BS or a home BS. In Figure 1 In the example shown in FIG, BS 110a may be a macro BS for macro cell 102a, BS 110b may be a pico BS for pico cell 102b, and BS 110c may be a femto BS for femto cell 102c. A BS may support one or more (e.g., three) cells. The terms "eNB," "base station," "NR BS," "gNB," "TRP," "AP," "Node B," "5G NB," and "cell" may be used interchangeably.

[0043] In some aspects, the cells may not necessarily be stationary, and the geographic area of ​​the cells may move depending on the location of the mobile BS. In some aspects, the BSs may interconnect with each other and / or with one or more other BSs or network nodes (not shown) in the wireless network 100 via various types of backhaul interfaces (such as direct physical connections, virtual networks, etc.) using any suitable transport network.

[0044] The wireless network 100 may also include a relay station. A relay station is an entity that can receive transmissions of data from an upstream station (e.g., a BS or UE) and send transmissions of the data to a downstream station (e.g., a UE or BS). A relay station may also be a UE that can relay transmissions for other UEs. Figure 1 In the example shown in , relay station 110d may communicate with macro BS 110a and UE 120d to facilitate communication between BS 110a and UE 120d. A relay station may also be referred to as a relay BS, relay base station, relay, or the like.

[0045] The wireless network 100 may be a heterogeneous network including different types of BSs (e.g., macro BSs, pico BSs, femto BSs, relay BSs, etc.). These different types of BSs may have different transmit power levels, different coverage areas, and different impacts on interference in the wireless network 100. For example, a macro BS may have a high transmit power level (e.g., 5 to 40 watts), while a pico BS, a femto BS, and a relay BS may have a lower transmit power level (e.g., 0.1 to 2 watts).

[0046] A network controller 130 may be coupled to a set of BSs and may provide coordination and control of these BSs. The network controller 130 may communicate with each BS via a backhaul. The BSs may also communicate with each other directly or indirectly, for example, via a wireless or wired backhaul.

[0047] UEs 120 (e.g., 120a, 120b, 120c) may be dispersed throughout the wireless network 100, and each UE may be stationary or mobile. A UE may also be referred to as an access terminal, terminal, mobile station, subscriber unit, station, etc. A UE may be a cellular phone (e.g., a smartphone), a personal digital assistant (PDA), a wireless modem, a wireless communication device, a handheld device, a laptop computer, a cordless phone, a wireless local loop (WLL) station, a tablet, a camera, a gaming device, a netbook, a smartbook, an ultrabook, a medical device or equipment, a biometric sensor / device, a wearable device (e.g., a smart watch, smart clothing, smart glasses, a smart wristband, smart jewelry (e.g., a smart ring, a smart bracelet)), an entertainment device (e.g., a music or video device, or a satellite radio), a vehicle component or sensor, a smart meter / sensor, industrial manufacturing equipment, a global positioning system device, or any other suitable device configured to communicate via a wireless or wired medium.

[0048] Some UEs may be considered machine type communication (MTC) UEs, or evolved or enhanced machine type communication (eMTC) UEs. MTC and eMTC UEs include, for example, robots, drones, remote devices, sensors, meters, monitors, location tags, and the like, which may communicate with a base station, another device (e.g., a remote device), or some other entity. A wireless node may provide connectivity to or to a network (e.g., a wide area network (such as the Internet) or a cellular network), for example, via a wired or wireless communication link. Some UEs may be considered Internet of Things (IoT) devices, and / or may be implemented as NB-IoT (narrowband Internet of Things) devices. Some UEs may be considered customer premises equipment (CPE). UE 120 may be included inside a housing that houses components of UE 120, such as a processor component, a memory component, and the like.

[0049] In general, any number of wireless networks may be deployed in a given geographic area. Each wireless network may support a specific RAT and may operate on one or more frequencies. A RAT may also be referred to as a radio technology, air interface, etc. A frequency may also be referred to as a carrier, frequency channel, etc. Each frequency may support a single RAT in a given geographic area to avoid interference between wireless networks of different RATs. In some cases, NR or 5G RAT networks may be deployed.

[0050] In some aspects, two or more UEs 120 (e.g., shown as UE 120a and UE 120e) may communicate directly (e.g., without using base station 110 as an intermediary) using one or more sidelink channels. For example, UE 120 may communicate using peer-to-peer (P2P) communication, device-to-device (D2D) communication, vehicle-to-everything (V2X) protocols (e.g., which may include vehicle-to-vehicle (V2V) protocols, vehicle-to-infrastructure (V2I) protocols, etc.), mesh networks, etc. In this scenario, UE 120 may perform scheduling operations, resource selection operations, and / or other operations described elsewhere as being performed by base station 110. For example, base station 110 may configure UE 120 via downlink control information (DCI), radio resource control (RRC) signaling, media access control-control element (MAC-CE), or via system information (e.g., system information blocks (SIBs)).

[0051] UE 120 may include a machine learning (ML) capability reporting module 140. For simplicity, only one UE 120d is shown as including the ML capability reporting module 140. ML capability reporting module 140 may receive a machine learning model from a base station and report the machine learning processing capabilities to the base station. ML capability reporting module 140 may also transmit gradient updates or weight updates to the machine learning model to the base station.

[0052] Base station 110 may include an ML capability grouping module 138. For simplicity, only one base station 110a is shown as including an ML capability reporting module 138. ML capability grouping module 138 may transmit a machine learning model to multiple user equipments (UEs). ML capability grouping module 138 may also receive a machine learning processing capability report from each UE. ML capability grouping module 138 may further group the UEs based on the machine learning processing capability report for receiving gradient updates to the machine learning model.

[0053] As indicated above, Figure 1 These are provided as examples only. Other examples may differ from those regarding Figure 1 Examples described.

[0054] Figure 2 A block diagram shows a design 200 of a base station 110 and a UE 120, which may be Figure 1 One for each base station and one for each UE in . Base station 110 may be equipped with T antennas 234a through 234t, and UE 120 may be equipped with R antennas 252a through 252r, where in general T≧1 and R≧1.

[0055] At base station 110, transmit processor 220 may receive data for one or more UEs from data source 212, select one or more modulation and coding schemes (MCS) for each UE based at least in part on a channel quality indicator (CQI) received from the UE, process (e.g., encode and modulate) the data for each UE based at least in part on the MCS(s) selected for the UE, and provide data symbols for all UEs. Reducing the MCS may reduce throughput but improve transmission reliability. Transmit processor 220 may also process system information (e.g., for semi-static resource partitioning information (SRPI)) and control information (e.g., CQI requests, grants, upper layer signaling, etc.), and provide overhead symbols and control symbols. Transmit processor 220 may also generate reference symbols for reference signals (e.g., a cell-specific reference signal (CRS)) and synchronization signals (e.g., a primary synchronization signal (PSS) and a secondary synchronization signal (SSS)). A transmit (TX) multiple-input, multiple-output (MIMO) processor 230 may perform spatial processing (e.g., precoding) on ​​data symbols, control symbols, overhead symbols, and / or reference symbols, as applicable, and may provide T output symbol streams to T modulators (MODs) 232a through 232t. Each modulator 232 may process a respective output symbol stream (e.g., for OFDM, etc.) to obtain an output sample stream. Each modulator 232 may further process (e.g., convert to analog, amplify, filter, and frequency upconvert) the output sample stream to obtain a downlink signal. The T downlink signals from modulators 232a through 232t may be transmitted via T antennas 234a through 234t, respectively. According to various aspects described in greater detail below, position coding may be utilized to generate synchronization signals to convey additional information.

[0056] At UE 120, antennas 252a through 252r may receive downlink signals from base station 110 and / or other base stations and may provide received signals to demodulators (DEMODs) 254a through 254r, respectively. Each demodulator 254 may condition (e.g., filter, amplify, downconvert, and digitize) the received signal to obtain input samples. Each demodulator 254 may further process the input samples (e.g., for OFDM, etc.) to obtain received symbols. A MIMO detector 256 may obtain received symbols from all R demodulators 254a through 254r, perform MIMO detection on the received symbols where applicable, and provide detected symbols. A receive processor 258 may process (e.g., demodulate and decode) the detected symbols, provide decoded data for UE 120 to a data sink 260, and provide decoded control information and system information to a controller / processor 280. The channel processor may determine reference signal received power (RSRP), received signal strength indicator (RSSI), reference signal received quality (RSRQ), channel quality indicator (CQI), etc. In some aspects, one or more components of UE 120 may be included in a housing.

[0057] On the uplink, at the UE 120, a transmit processor 264 may receive and process data from a data source 262 and control information (e.g., for reports including RSRP, RSSI, RSRQ, CQI, etc.) from the controller / processor 280. The transmit processor 264 may also generate reference symbols for one or more reference signals. The symbols from the transmit processor 264 may be precoded by a TX MIMO processor 266, if applicable, further processed by modulators 254a through 254r (e.g., for DFT-s-OFDM, CP-OFDM, etc.), and transmitted to the base station 110. At the base station 110, uplink signals from the UE 120 and other UEs may be received by the antennas 234, processed by the demodulators 254, detected by the MIMO detector 236, if applicable, and further processed by the receive processor 238 to obtain decoded data and control information sent by the UE 120. Receive processor 238 may provide decoded data to data sink 239 and provide decoded control information to controller / processor 240. Base station 110 may include communication unit 244 and communicate with network controller 130 via communication unit 244. Network controller 130 may include communication unit 294, controller / processor 290, and memory 292.

[0058] The controller / processor 240 of the base station 110, the controller / processor 280 of the UE 120, and / or Figure 2Any other component(s) of the UE 120 may perform one or more techniques associated with machine learning capability reporting, as described in greater detail elsewhere. For example, the controller / processor 240 of the base station 110, the controller / processor 280 of the UE 120, and / or Figure 2 Any other component(s) may perform or direct e.g. Figure 7-Figure 9 Memories 242 and 282 may store data and program codes for base station 110 and UE 120, respectively. Scheduler 246 may schedule UEs for data transmission on the downlink and / or uplink.

[0059] In some aspects, the UE 120 or base station 110 may include means for receiving, means for reporting, means for transmitting, means for grouping, and / or means for scheduling. Such means may include in conjunction with Figure 2 One or more components of the UE 120 or base station 110 are described.

[0060] As indicated above, Figure 2 These are provided as examples only. Other examples may differ from those regarding Figure 2 Examples described.

[0061] In some cases, different types of devices supporting different types of applications and / or services may coexist in a cellular cell. Examples of different types of devices include UE handsets, customer premises equipment (CPE), vehicles, Internet of Things (IoT) devices, and the like. Examples of different types of applications include ultra-reliable low-latency communications (URLLC) applications, massive machine-type communications (mMTC) applications, enhanced mobile broadband (eMBB) applications, and vehicle-to-everything (V2X) applications. Furthermore, in some cases, a single device may simultaneously support different applications or services.

[0062] Figure 3An example implementation of a system on a chip (SOC) 300 according to certain aspects of the present disclosure is illustrated, which may include a central processing unit (CPU) 302 or a multi-core CPU configured to generate gradients for neural network training. The SOC 300 may be included in a base station 110 or a UE 120. Variables (e.g., neural signals and synaptic weights), system parameters associated with a computing device (e.g., a neural network with weights), delays, frequency bin information, and task information may be stored in a memory block associated with a neural processing unit (NPU) 308, a memory block associated with the CPU 302, a memory block associated with a graphics processing unit (GPU) 304, a memory block associated with a digital signal processor (DSP) 306, a memory block 318, or may be distributed across multiple blocks. Instructions executed at the CPU 302 may be loaded from a program memory associated with the CPU 302 or may be loaded from the memory block 318.

[0063] The SOC 300 may also include additional processing blocks tailored for specific functions, such as a GPU 304, a DSP 306, a connectivity block 310 (which may include fifth generation (5G) connectivity, fourth generation long term evolution (4G LTE) connectivity, Wi-Fi connectivity, USB connectivity, Bluetooth connectivity, etc.), and a multimedia processor 312 that may, for example, detect and recognize gestures. In one implementation, the NPU is implemented in the CPU, DSP, and / or GPU. The SOC 300 may also include a sensor processor 314, an image signal processor (ISP) 316, and / or a navigation module 320 (which may include a global positioning system).

[0064] The SOC 300 may be based on the ARM instruction set. In various aspects of the present disclosure, the instructions loaded into the general processor 302 may include code for receiving a machine learning model from a base station. The general processor 302 may also include code for reporting machine learning processing capabilities to the base station. The general processor 302 may further include code for transmitting gradient updates or weight updates to the machine learning model to the base station. In other aspects of the present disclosure, the instructions loaded into the general processor 302 may include: code for transmitting a machine learning model to multiple user equipments (UEs); and code for receiving a machine learning processing capability report from each UE. The instructions loaded into the general processor 302 may also include code for grouping UEs according to the machine learning processing capability report for receiving gradient updates to the machine learning model.

[0065] Deep learning architectures perform object recognition tasks by learning to represent inputs at successively higher levels of abstraction at each layer, thereby constructing useful feature representations of the input data. In this way, deep learning addresses a major bottleneck of traditional machine learning. Before the advent of deep learning, machine learning approaches to object recognition problems might rely heavily on human-engineered features, perhaps combined with shallow classifiers. A shallow classifier might be a two-class linear classifier, for example, where the weighted sum of the feature vector components is compared to a threshold to predict which class the input belongs to. Human-engineered features might be templates or kernels customized for a specific problem domain by engineers with domain expertise. In contrast, deep learning architectures learn to represent features similar to those that a human engineer might design, but they do so through training. Furthermore, deep networks can learn to represent and recognize new types of features that humans might not have considered.

[0066] Deep learning architectures can learn hierarchies of features. For example, if a first layer is presented with visual data, it can learn to recognize relatively simple features (such as edges) in the input stream. In another example, if a first layer is presented with auditory data, it can learn to recognize spectral power in specific frequencies. A second layer, taking the output of the first layer as input, can learn to recognize combinations of features, such as simple shapes for visual data or combinations of sounds for auditory data. For example, higher layers can learn to represent complex shapes in visual data or words in auditory data. Even higher layers can learn to recognize common visual objects or spoken phrases.

[0067] Deep learning architectures can perform particularly well when applied to problems with a natural hierarchical structure. For example, the classification of motor vehicles can benefit from first learning to recognize wheels, windshields, and other features. These features can then be combined in different ways at higher levels to identify cars, trucks, and airplanes.

[0068] Neural networks can be designed with various connectivity patterns. In a feedforward network, information is passed from lower layers to higher layers, with each neuron in a given layer communicating to neurons in a higher layer. As described above, hierarchical representations can be constructed in successive layers of a feedforward network. Neural networks can also have reflow or feedback (also known as top-down) connections. In a reflow connection, the output from a neuron in a given layer can be communicated to another neuron in the same layer. The reflow architecture can help identify patterns that span more than one chunk of input data delivered sequentially to the neural network. The connections from neurons in a given layer to neurons in lower layers are called feedback (or top-down) connections. A network with many feedback connections can be helpful when the recognition of high-level concepts can assist in discerning specific low-level features of the input.

[0069] The connections between layers of a neural network can be fully connected or partially connected. Figure 4A Illustrated is an example of a fully connected neural network 402. In the fully connected neural network 402, a neuron in a first layer may communicate its output to every neuron in a second layer, such that every neuron in the second layer will receive input from every neuron in the first layer. Figure 4B An example of a locally connected neural network 404 is illustrated. In the locally connected neural network 404, neurons in a first layer may be connected to a limited number of neurons in a second layer. More generally, the locally connected layers of the locally connected neural network 404 may be configured such that each neuron in a layer will have the same or similar connectivity pattern, but their connection strengths may have different values ​​(e.g., 410, 412, 414, and 416). The locally connected connectivity pattern may produce spatially distinct receptive fields in higher layers because higher layer neurons in a given area may receive inputs that are tuned through training to properties of a limited portion of the total input to the network.

[0070] An example of a locally connected neural network is a convolutional neural network. Figure 4C An example of a convolutional neural network 406 is illustrated. The convolutional neural network 406 can be configured such that the connection strengths associated with the inputs to each neuron in the second layer are shared (e.g., 408). Convolutional neural networks may be well suited for problems where the spatial location of the inputs is meaningful.

[0071] One type of convolutional neural network is a deep convolutional network (DCN). Figure 4D A detailed example of a DCN 400 designed to recognize visual features from an image 426 input from an image capture device 430 (such as an onboard camera) is illustrated. The DCN 400 of the current example can be trained to identify traffic signs and the numbers provided on traffic signs. Of course, the DCN 400 can be trained for other tasks, such as identifying lane markings or identifying traffic lights.

[0072] DCN 400 can be trained using supervised learning. During training, an image (such as image 426 of a speed limit sign) can be presented to DCN 400, and a "forward pass" can then be calculated to produce output 422. DCN 400 can include a feature extraction section and a classification section. Upon receiving image 426, convolution layer 432 can apply a convolution kernel (not shown) to image 426 to generate a first set of feature maps 418. As an example, the convolution kernel of convolution layer 432 can be a 5x5 kernel that generates a 28x28 feature map. In this example, since four different feature maps are generated in first set of feature maps 418, four different convolution kernels are applied to image 426 at convolution layer 432. Convolution kernels can also be referred to as filters or convolution filters.

[0073] The first set of feature maps 418 may be subsampled by a max pooling layer (not shown) to generate a second set of feature maps 420. The max pooling layer reduces the size of the first set of feature maps 418. That is, the size of the second set of feature maps 420 (such as 14x14) is smaller than the size of the first set of feature maps 418 (such as 28x28). The reduced size provides similar information to subsequent layers while reducing memory consumption. The second set of feature maps 420 may be further convolved via one or more subsequent convolutional layers (not shown) to generate one or more subsequent sets of feature maps (not shown).

[0074] exist Figure 4D In the example of , the second set of feature maps 420 is convolved to generate a first feature vector 424. In addition, the first feature vector 424 is further convolved to generate a second feature vector 428. Each feature of the second feature vector 428 may include a number corresponding to a possible feature of the image 426 (such as "sign," "60," and "100"). A softmax function (not shown) may convert the numbers in the second feature vector 428 into probabilities. Thus, the output 422 of the DCN 400 is the probability that the image 426 includes one or more features.

[0075] In this example, the probabilities for "logo" and "60" in output 422 are higher than the probabilities for other features of output 422 (such as "30," "40," "50," "70," "80," "90," and "100"). Before training, output 422 generated by DCN 400 is likely incorrect. Thus, the error between output 422 and the target output can be calculated. The target output is the true value of image 426 (e.g., "logo" and "60"). The weights of DCN 400 can then be adjusted to more closely align output 422 of DCN 400 with the target output.

[0076] To adjust the weights, the learning algorithm can calculate a gradient vector for the weights. This gradient can indicate the amount by which the error will increase or decrease if the weights are adjusted. At the top layer, this gradient can directly correspond to the value of the weight connecting the activated neurons in the penultimate layer to the neurons in the output layer. In lower layers, this gradient can depend on the value of the weights and the calculated error gradient for the higher layers. The weights can then be adjusted to reduce the error. This method of adjusting weights can be called "backpropagation" because it involves a "backward pass" through the neural network.

[0077] In practice, the error gradient of the weights may be calculated over a small number of examples so that the calculated gradient approximates the true error gradient. This approximation method may be called stochastic gradient descent. Stochastic gradient descent may be repeated until the error rate achievable by the entire system has stopped decreasing or until the error rate has reached a target level. After learning, a new image (e.g., image 426 of a speed limit sign) may be presented to the DCN and a forward pass through the network may produce output 422, which may be considered an inference or prediction of the DCN.

[0078] Deep Belief Network (DBN) is a probabilistic model comprising multiple layers of hidden nodes. DBN can be used to extract hierarchical representations of training data sets. DBN can be obtained by stacking multiple layers of restricted Boltzmann machines (RBM). RBM is a type of artificial neural network that can learn probability distributions on input sets. Since RBM can learn probability distributions without information about which class each input should be classified into, RBM is often used in unsupervised learning. Using a hybrid unsupervised and supervised paradigm, the bottom RBM of the DBN can be trained in an unsupervised manner and can be used as a feature extractor, while the top RBM can be trained in a supervised manner (on the joint distribution of the input and target class from the previous layer) and can be used as a classifier.

[0079] A deep convolutional network (DCN) is a network of convolutional networks with additional pooling and normalization layers. DCNs have achieved state-of-the-art performance on many tasks. DCNs can be trained using supervised learning, where both the input and output targets are known for many examples and are used to modify the network weights using gradient descent.

[0080] The DCN can be a feedforward network. In addition, as described above, the connections from neurons in the first layer of the DCN to the neuron groups in the next higher layer are shared across the neurons in the first layer. The feedforward and shared connections of the DCN can be used to perform fast processing. The computational burden of the DCN can be much smaller than, for example, the computational burden of a similarly sized neural network that includes recurrent or feedback connections.

[0081] The processing of each layer of the convolutional network can be thought of as a spatially invariant template or basis projection. If the input is first decomposed into multiple channels, such as the red, green, and blue channels of a color image, the convolutional network trained on this input can be thought of as three-dimensional, with two spatial dimensions along the axes of the image and a third dimension that captures color information. The output of the convolutional connection can be thought of as forming a feature map in the subsequent layer, each element in which receives input from a range of neurons in the previous layer (e.g., feature map 218) and from each of the multiple channels. The values ​​in the feature map can be further processed with nonlinearities (such as rectification, max(0,x)). The values ​​from adjacent neurons can be further pooled (which corresponds to downsampling) and can provide additional local invariance and dimensionality reduction. Normalization can also be applied by lateral inhibition between neurons in the feature map, which corresponds to whitening.

[0082] The performance of deep learning architectures improves as more labeled data points become available or as computing power increases. Modern deep neural networks are routinely trained using thousands of times more computing resources than were available to typical researchers just fifteen years ago. New architectures and training paradigms can further boost deep learning performance. Rectified linear units can alleviate the training problem known as vanishing gradients. New training techniques can reduce overfitting and thus enable larger models to achieve better generalization. Encapsulation techniques can abstract the data within a given receptive field and further improve overall performance.

[0083] Figure 5 is a block diagram illustrating a deep convolutional network 550. The deep convolutional network 550 may include multiple layers of different types based on connectivity and weight sharing. Figure 5 As shown in FIG, a deep convolutional network 550 includes convolution blocks 554A and 554B. Each of the convolution blocks 554A and 554B may be configured with a convolution layer (CONV) 356, a normalization layer (LNorm) 558, and a maximum pooling layer (MAX POOL) 560.

[0084] The convolution layer 556 may include one or more convolution filters that can be applied to the input data to generate a feature map. Although only two convolution blocks 554A and 554B are shown, the present disclosure is not limited thereto, and instead any number of convolution blocks 554A and 554B may be included in the deep convolutional network 550 according to design preferences. The normalization layer 558 may normalize the output of the convolution filter. For example, the normalization layer 558 may provide whitening or lateral suppression. The maximum pooling layer 560 may provide spatial downsampling aggregation to achieve local invariance and dimensionality reduction.

[0085] For example, the parallel filter banks of the deep convolutional network can be offloaded onto the CPU 302 or GPU 304 of the SOC 300 to achieve high performance and low power consumption. In alternative embodiments, the parallel filter banks can be offloaded onto the DSP 306 or ISP 316 of the SOC 300. In addition, the deep convolutional network 550 can access other processing blocks that may be present on the SOC 300, such as the sensor processor 314 and the navigation module 320 dedicated to sensors and navigation, respectively.

[0086] The deep convolutional network 550 may also include one or more fully connected layers 562 (FC1 and FC2). The deep convolutional network 550 may further include a logistic regression (LR) layer 564. Between each layer 556, 558, 560, 562, 564 of the deep convolutional network 550 are weights (not shown) to be updated. The output of each layer (e.g., 556, 558, 560, 562, 564) can be used as input to a subsequent layer (e.g., 556, 558, 560, 562, 564) in the deep convolutional network 550 to learn a hierarchical feature representation from the input data 552 (e.g., image, audio, video, sensor data, and / or other input data) supplied from the first convolutional block 554A. The output of the deep convolutional network 550 is a classification score 566 for the input data 552. The classification score 566 can be a set of probabilities, where each probability is a probability that the input data includes a feature from a feature set.

[0087] As indicated above, Figure 3-Figure 5 are provided as examples. Other examples may differ from those described in Figure 3-Figure 5 Examples described.

[0088] As mentioned above, standard machine learning approaches centralize the training data on a single machine or in a data center. In contrast, federated learning is a process in which a group of UEs receive a machine learning model from a base station and work together to train the model. More specifically, each UE trains the model locally and sends back updated neural network model weights or gradient updates based on, for example, a stochastic gradient descent process performed locally. The base station receives updates from all UEs in the group and aggregates the updates, for example by averaging them, to obtain updated global weights for the neural network. The base station sends the updated model to each UE, and the process repeats round after round until the desired performance level of the global model is achieved.

[0089] In each round of the federated learning process, a group of UEs sends back weight or gradient updates within a given time interval after they receive the model from the base station. If a UE misses the deadline for sending an update, the weights or gradients will become stale, and the base station will not incorporate the updates into the weight or gradient aggregation for that round of the federated learning process.

[0090] According to various aspects of the present disclosure, a UE reports its machine learning processing capabilities to a base station. In some aspects, the report may indicate the machine learning hardware capabilities. In other aspects, the report indicates the approximate turnaround time for calculating gradients or weight updates in each round of federated learning. In still other aspects of the present disclosure, the UE reports the approximate turnaround time for calculating gradients or weights, for example, as a function of the UE's battery status.

[0091] Figure 6 6 is a block diagram illustrating a federated learning system 600 according to various aspects of the present disclosure. In some configurations, a base station 610 (e.g., a gNB) shares a global federated learning model 630 with a group of user equipments (UEs) 620 (e.g., 620a, 620b, 620c) participating in a federated learning process. In these configurations, the model parameters are optimized by the federated learning system 600. The model parameters w (n) represents the bias and weight of the global joint learning model 630, g (n) represents the gradient estimate, where n is the index of the joint learning round. The initial model parameters are specified as w (0) .

[0092] In these configurations, each UE 620 includes a local data set 640 (e.g., 640a, 640b, 640c), a gradient calculation block 624, and a gradient compression block 622. In this example, the gradient calculation block 624 of the second UE 620b is configured to perform local updates via decentralized stochastic gradient descent (SGD). Each UE 620 performs some type of training iteration, such as a single stochastic gradient descent step or multiple stochastic gradient descent steps as seen in equation (1):

[0093]

[0094] Among them F k (w (n) ) represents the local loss function of the weight w for the nth round of joint learning, and g k (n) represents the local gradient used for the nth round of federated learning.

[0095] The local update has been completed on UE 620 Then, the gradient compression block 622 may compress the calculated gradient vector as seen in equation (2) To obtain the compressed value (e.g., 632a, 632b, 632c), where q() represents a compression function:

[0096]

[0097] UE 620 feeds back the calculated compressed gradient vector to base station 610 (e.g., 632a, 632b, 632c). The joint learning process includes converting the calculated compressed gradient vector 632 (eg, 632a, 632b, 632c) are transmitted from all UEs 620 to the base station 610.

[0098] In these configurations, the base station 610 includes a gradient averaging block 612 configured to average the calculated compressed gradient vectors 632. Although averaging is shown, other types of aggregation are also contemplated. Additionally, the model update block 614 is configured to update the parameters of the global federated learning model 630. The updated model is then sent to all UEs 620. This process repeats until a global federated learning accuracy specification is met (e.g., until the global federated learning algorithm converges). The accuracy specification may refer to a desired level of accuracy for the local training. For example, the accuracy specification may indicate that the local training loss in each iteration of the federated learning process should drop below a threshold.

[0099] The global joint learning algorithm is based on the local loss function F as shown in Equation (3) k (w):

[0100]

[0101] where x j Represents the input vector of the model, y j represents the output scalar of the model, w is the weight vector of the global joint learning model, and D k Denotes the dataset size at the kth UE. For example, the input may be a vectorized image, and the output may be a detected number (e.g., a single scalar).

[0102] The global joint learning algorithm is also based on the global loss function F(w) as shown in Equation (4) (assuming |D k |=D):

[0103]

[0104] The overall goal of this joint learning process is to obtain the optimal parameters w* of the neural network that minimize the global loss function F(w):

[0105] w*=argmin F(w). (5)

[0106] In this joint learning process, the calculated compressed gradient vector The local computations of 632 (e.g., for updating the global federated learning model 630) are collected from each UE 620, and the average is calculated by the gradient averaging block 612 (or another type of aggregated estimation) as follows:

[0107]

[0108] Based on the average gradient g )n) , the updated model parameters are transmitted (eg, broadcast) from the base station 610 to each UE 620. In addition, the model update block 614 of the base station 610 performs the model update as seen in equation (6):

[0109] w (n+1) =w (n) -η.g (n) ,(6)where η represents the learning rate, which is a parameter of the global joint learning model 630.

[0110] In each round of the federated learning process, a group of UEs sends back weight or gradient updates within a given time interval after receiving the model from the base station. In one configuration, the group size is ten to twenty UEs. If a UE misses the deadline for sending an update, the weights or gradients will become stale, and the base station will not incorporate the gradient updates from that UE in that round of the federated learning process.

[0111] If the base station knows the machine learning capabilities of the UEs participating in the joint learning process, this information can be useful to the base station. For example, the base station can group the UEs according to their machine learning capabilities for different rounds of joint learning. If slower UEs are grouped with faster UEs, the slower UEs will be a bottleneck in the training process, thereby adversely affecting the convergence time of the joint learning process. Thus, slower UEs can be grouped with other slower UEs, while fast UEs are grouped with other fast UEs. In addition, different UEs can be paired together for different rounds of the joint learning training process.

[0112] According to various aspects of the present disclosure, a UE reports its machine learning processing capabilities to a base station. The machine learning processing capability report may have a standardized format. For example, the report may be added to the UE capability report defined in 3GPP TS 38.306. The standardized format may indicate the machine learning hardware capabilities of the UE, such as the capabilities of a GPU, NPU, etc.

[0113] In various aspects of the present disclosure, the report indicates the machine learning hardware capabilities in the form of standard metrics of machine learning hardware capabilities. For example, the report may indicate the number of operations per second or the number of multiply-accumulate (MAC) operations per second, etc. These metrics are basic hardware characteristics of the UE and do not change over time.

[0114] Hardware characteristics may reflect a best-case scenario. Thus, the report may indicate manufacturer specifications such as tera operations per second (TOP / s) or tera multiply-accumulate operations per second (TMAC / s). Manufacturer-specified hardware capabilities may be closer to real-world performance.

[0115] In any case, the reported machine learning hardware capabilities provide the base station with an approximate training time on the UE side to prepare gradient or weight updates. For example, the base station can determine whether the reporting UE is a fast UE or a slow UE based on the reported machine learning hardware capabilities. The base station can schedule UEs according to speed ranges. For example, UEs with a first processing capability range can be included in a first group, while UEs with a second processing capability range can be included in a second group. In some implementations, the processing capability can be a machine learning processing capability.

[0116] In other aspects of the present disclosure, the report indicates an approximate or estimated turnaround time for computing gradients or weight updates in each round of federated learning. The report may indicate, for example, a quantized time or an approximate time.

[0117] The turnaround time is a function of the machine learning hardware capabilities of the UE. The turnaround time is also a function of parameters such as the type of federated learning process employed or the application associated with a particular federated learning process. The turnaround time may be a function of other parameters such as the desired level of accuracy of the machine learning model and / or the actual type of machine learning model trained. Other parameters that affect the turnaround time include the learning rate for local training and / or the number of iterations (e.g., stochastic gradient descent iterations) required before an update is derived and sent.

[0118] The batch size used for local training at the UE also affects the turnaround time. For example, smaller batches of training data take less time to process than larger batches of training data. Note that smaller batch sizes increase the number of iterations.

[0119] According to various aspects of the present disclosure, a base station may configure a UE with the above-mentioned parameters for a particular federated learning process. The UE may then use knowledge of these parameters to estimate the amount of time used to calculate weight or gradient updates and report (approximate) turnaround times. For this option, the UE refrains from sending updated reports as long as the above-mentioned parameters are fixed for a given federated learning process. The UE sends updated reports when the parameters are reconfigured.

[0120] In other aspects of the present disclosure, a UE reports an approximate turnaround time for computing gradients or weights, which may vary, for example, depending on the UE's battery status. For example, if the UE is in power saving mode, the UE may decide not to participate in federated learning. Lack of participation may be achieved, for example, by setting the turnaround time to infinity. In other aspects, the turnaround time may be set to a larger value.

[0121] Note that reporting machine learning hardware capabilities can be less dynamic than reporting turnaround time. Furthermore, reporting turnaround time that varies with battery state can be more dynamic than reporting a more general turnaround time.

[0122] Figure 7 is a timing diagram illustrating reporting machine learning capabilities according to various aspects of the present disclosure. At time t1, base station 610 receives a machine learning (ML) capability report from a first UE 620a. At time t2, base station 610 receives a machine learning (ML) capability report from a second UE 620b. At time t3, base station 610 receives a machine learning (ML) capability report from a third UE 620c. The machine learning capability report may indicate machine learning hardware capabilities or machine learning turnaround time, as previously described.

[0123] Based on the received machine learning capability reports, the base station 610 groups the UEs 620 at time t4 and schedules the UEs 620 according to the grouping at time t5. In this example, the first UE 620a and the third UE 620c are grouped together as faster UEs. The second UE 620b is in its own group. Accordingly, at time t6, the first UE 620a and the third UE 620c send their updates to the machine learning (ML) model. These updates are calculated locally at each UE 620 before transmission and will be aggregated at the base station 610 for each round of federated learning. At time t7, the second UE 620b sends its updates to the UEs to be included in that round of federated learning. Due to the grouping of the UEs 620, the slower UEs may not miss the deadline for their federated learning update rounds. Accordingly, the base station considers a more complete set of updates and can train the model more quickly and accurately.

[0124] Figure 8 is a flow diagram illustrating an example process 800, performed, for example, by a UE, in accordance with various aspects of the present disclosure. The example process 800 is an example of user equipment (UE) capability reporting for machine learning applications.

[0125] like Figure 8 As shown in FIG, in some aspects, process 800 may include receiving a machine learning model from a base station (block 802). For example, a UE (e.g., using antenna 252, DEMOD / MOD 254, MIMO detector 256, receive processor 258, controller / processor 280, and / or memory 282) may receive the machine learning model. The machine learning model may be trained in a federated learning process.

[0126] Process 800 may also include reporting machine learning processing capabilities to the base station (block 804). For example, the UE (e.g., using antenna 252, DEMOD / MOD 254, TX MIMO processor 266, transmit processor 264, controller / processor 280, and / or memory 282) may report machine learning processing capabilities to the base station. In some aspects of the present disclosure, the report may indicate the machine learning hardware capabilities. In other aspects, the report indicates the approximate turnaround time for calculating gradients or weight updates in each round of federated learning. In still other aspects, the UE reports the approximate turnaround time for calculating gradients or weights, for example, as a function of the UE's battery state. The machine learning processing capability report may have a standardized format.

[0127] Process 800 may further include transmitting gradient updates or weight updates to the machine learning model to the base station (block 806). For example, the UE (e.g., using antennas 252, DEMOD / MOD 254, TX MIMO processor 266, transmit processor 264, controller / processor 280, and / or memory 282) may transmit gradient updates or weight updates to the base station. These updates may be calculated locally as part of the federated learning process.

[0128] Figure 9 is a flow diagram illustrating an example process 900, performed, for example, by a base station, in accordance with various aspects of the present disclosure. The example process 900 is an example of user equipment (UE) capability reporting for machine learning applications.

[0129] like Figure 9 As shown in FIG, in some aspects, process 900 may include transmitting a machine learning model to a plurality of user equipments (UEs) (block 902). For example, a base station (e.g., using antennas 234, MOD / DEMOD 232, TX MIMO processor 230, transmit processor 220, controller / processor 240, and / or memory 242) may transmit the machine learning model. The machine learning model may be trained in a federated learning process.

[0130] Process 900 may include receiving a machine learning processing capability report from each of the plurality of UEs (block 904). For example, a base station (e.g., using antenna 234, MOD / DEMOD 232, MIMO detector 236, receive processor 238, controller / processor 240, and / or memory 242) may receive a machine learning processing capability report from each of the plurality of UEs. In some aspects of the present disclosure, the report may indicate the machine learning hardware capabilities. In other aspects, the report indicates an approximate turnaround time for computing gradients or weight updates in each round of federated learning. In still other aspects, the UE reports the approximate turnaround time for computing gradients or weights as a function of, for example, the battery state of the UE. The machine learning processing capability report may have a standardized format.

[0131] Process 900 may further include grouping UEs for receiving gradient updates to the machine learning model based on the machine learning processing capability reports (block 906). For example, the base station (e.g., using antennas 234, MOD / DEMOD 232, MIMO detector 236, TX MIMO processor 230, receive processor 238, transmit processor 220, controller / processor 240, and / or memory 242) may group the plurality of UEs. For example, the base station may determine whether the reporting UE is a fast UE or a slow UE based on the reported machine learning hardware capabilities. The base station may schedule UEs based on speed ranges. A group of higher speed UEs may be grouped together, while a group of lower speed UEs may be grouped together.

[0132] Example aspects

[0133] Aspect 1: A method for wireless communication by a user equipment (UE), comprising: receiving a machine learning model from a base station; reporting machine learning processing capabilities to the base station; and transmitting gradient updates or weight updates to the machine learning model to the base station.

[0134] Aspect 2: The method of Aspect 1, wherein the machine learning processing capability includes machine learning hardware capability.

[0135] Aspect 3: The method of Aspect 1 or 2, wherein the machine learning hardware capabilities include manufacturer-specified hardware capabilities.

[0136] Aspect 4: The method of any of the preceding aspects, wherein the machine learning processing capability includes an estimated turnaround time for computing gradients.

[0137] Aspect 5: A method as in any of the preceding aspects, wherein the estimated turnaround time is based on the federated learning process, the federated learning application, the expected accuracy level of the local training, the type of the machine learning model, the number of local training rounds, the batch size configured for UE training, and / or the learning rate of the local training.

[0138] Aspect 6: The method of any of the preceding aspects, further comprising: receiving from the base station parameters for the federated learning process, the federated learning application, the desired accuracy level, the type of machine learning model, the number of local training rounds, the batch size, and / or the learning rate.

[0139] Aspect 7: The method of any of the preceding aspects, further comprising: reporting updated machine learning processing capabilities in response to a change in at least one parameter.

[0140] Aspect 8: The method of any one of the preceding aspects, wherein the estimated turnaround time is based on a battery status of the UE.

[0141] Aspect 9: The method of any of the preceding aspects, wherein the estimated turnaround time is set to infinity in response to the UE operating in a power save mode.

[0142] Aspect 10: A method for wireless communication by a base station, comprising: transmitting a machine learning model to a plurality of user equipments (UEs); receiving a machine learning processing capability report from each of the plurality of UEs; and grouping the plurality of UEs according to the machine learning processing capability report from each of the plurality of UEs for receiving gradient updates to the machine learning model.

[0143] Aspect 11: The method of aspect 10, wherein the grouping further comprises: scheduling UEs with a first processing capability to transmit gradient updates in a first time period and scheduling UEs with a second processing capability to transmit gradient updates in a second time period.

[0144] Aspect 12: An apparatus for wireless communication by a user equipment (UE), comprising: a processor; a memory coupled to the processor; and instructions stored in the memory and operable, when executed by the processor, to cause the apparatus to: receive a machine learning model from a base station; report machine learning processing capabilities to the base station; and transmit gradient updates or weight updates to the machine learning model to the base station.

[0145] Aspect 13: The apparatus of Aspect 12, wherein the machine learning processing capability comprises machine learning hardware capability.

[0146] Aspect 14: The apparatus of Aspect 12 or 13, wherein the machine learning hardware capabilities include manufacturer-specified hardware capabilities.

[0147] Aspect 15: The apparatus of any one of aspects 12-14, wherein the machine learning processing capability comprises an estimated turnaround time for computing gradients.

[0148] Aspect 16: An apparatus as in any of Aspects 12-15, wherein the estimated turnaround time is based on the federated learning process, the federated learning application, the desired accuracy level of the local training, the type of machine learning model, the number of local training rounds, the batch size configured for UE training, and / or the learning rate of the local training.

[0149] Aspect 17: An apparatus as in any of Aspects 12-16, wherein the processor causes the apparatus to receive from the base station parameters for a federated learning process, a federated learning application, a desired level of accuracy, a type of machine learning model, a number of local training rounds, a batch size, and / or a learning rate.

[0150] Aspect 18: The apparatus of any of Aspects 12-17, wherein the processor causes the apparatus to report updated machine learning processing capabilities in response to a change in at least one parameter.

[0151] Aspect 19: The apparatus of any one of aspects 12-18, wherein the estimated turnaround time is based on a battery status of the UE.

[0152] Aspect 20: The apparatus of any one of aspects 12-19, wherein the estimated turnaround time is set to infinity in response to the UE operating in a power save mode.

[0153] Aspect 21: A user equipment (UE) for wireless communication, comprising: a device for receiving a machine learning model from a base station; a device for reporting machine learning processing capabilities to the base station; and a device for transmitting gradient updates or weight updates to the machine learning model to the base station.

[0154] Aspect 22: The UE of Aspect 21, wherein the machine learning processing capability includes machine learning hardware capability.

[0155] Aspect 23: The UE of Aspect 21 or 22, wherein the machine learning hardware capabilities include manufacturer-specified hardware capabilities.

[0156] Aspect 24: The UE of any one of Aspects 21-23, wherein the machine learning processing capability includes an estimated turnaround time for computing gradients.

[0157] Aspect 25: A UE as in any of Aspects 21-24, wherein the estimated turnaround time is based on the federated learning process, the federated learning application, the expected accuracy level of the local training, the type of machine learning model, the number of local training rounds, the batch size configured for UE training, and / or the learning rate of the local training.

[0158] Aspect 26: A UE as in any one of Aspects 21-25, further comprising: means for receiving parameters for a federated learning process, a federated learning application, a desired accuracy level, a type of machine learning model, a number of local training rounds, a batch size, and / or a learning rate from the base station.

[0159] Aspect 27: The UE of any one of aspects 21-26, further comprising: means for reporting updated machine learning processing capabilities in response to a change in at least one parameter.

[0160] Aspect 28: The UE according to any one of aspects 21-27, wherein the estimated turnaround time is based on a battery status of the UE.

[0161] Aspect 29: The UE of any one of aspects 21-28, wherein the estimated turnaround time is set to infinity in response to the UE operating in a power saving mode.

[0162] The foregoing disclosure provides illustration and description, but is not intended to be exhaustive or to limit the aspects to the precise form disclosed. Modifications and variations may be made in light of the above disclosure or may be acquired from practice of the aspects.

[0163] As used, the term "component" is intended to be broadly interpreted as hardware, firmware, and / or a combination of hardware and software. As used, a processor is implemented in hardware, firmware, and / or a combination of hardware and software.

[0164] Some aspects are described in conjunction with thresholds. As used, satisfying a threshold may refer to a value being greater than a threshold, greater than or equal to a threshold, less than a threshold, less than or equal to a threshold, equal to a threshold, not equal to a threshold, etc., depending on the context.

[0165] It will be apparent that the described systems and / or methods can be implemented in various forms of hardware, firmware, and / or a combination of hardware and software. The actual dedicated control hardware or software code used to implement these systems and / or methods does not limit the aspects. Thus, the operation and behavior of these systems and / or methods are described without reference to specific software code - it is understood that software and hardware can be designed to implement these systems and / or methods based at least in part on this description.

[0166] Although specific feature combinations are described in the claims and / or disclosed in the specification, these combinations are not intended to limit the disclosure of various aspects. In fact, many of these features can be combined in a manner not specifically described in the claims and / or not disclosed in the specification. Although each dependent claim listed below can be directly subordinate to only one claim, the disclosure of various aspects includes that each dependent claim is combined with each other claim in this group of claims. The phrase quoting "at least one of" a column of items refers to any combination of these items, including single members. As an example, "at least one of a, b or c" is intended to encompass: a, b, c, ab, ac, bc, and abc, as well as any combination with multiple identical elements (e.g., aa, aaa, aab, aac, abb, acc, bb, bbb, bbc, cc, and ccc, or any other arrangement of a, b and c).

[0167] The elements, actions or instructions used should not be interpreted as critical or necessary unless expressly described as such. Moreover, as used, the articles "a" and "a" are intended to include one or more items and can be used interchangeably with "one or more". Furthermore, as used, the terms "set" and "group" are intended to include one or more items (e.g., related items, non-related items, a combination of related and non-related items, etc.) and can be used interchangeably with "one or more". Where only one item is intended, the phrase "only one" or similar language is used. Furthermore, as used, the terms "having", "containing", "including", etc. are intended to be open terms. Furthermore, the phrase "based on" is intended to mean "based at least in part on", unless otherwise expressly stated.

Claims

1. A method for wireless communication by a user equipment (UE), comprising: Receive machine learning models from network nodes; reporting a machine learning processing capability to the network node, the machine learning processing capability indicating a turnaround time for the UE to calculate a gradient for the machine learning model; calculating a gradient update or a weight update to the machine learning model based on the turnaround time; as well as The gradient update or the weight update to the machine learning model is transmitted to the network node.

2. The method according to claim 1, wherein The machine learning processing capabilities include machine learning hardware capabilities.

3. The method according to claim 2, wherein: The machine learning hardware capabilities include manufacturer-specified hardware capabilities.

4. The method according to claim 1, wherein The machine learning processing capability includes an estimated turnaround time for calculating the gradient.

5. The method according to claim 4, wherein: The estimated turnaround time is based on the federated learning process, the federated learning application, the desired accuracy level of the local training, the type of the machine learning model, the number of local training rounds, the batch size configured for UE training, and / or the learning rate of the local training.

6. The method of claim 5, further comprising: Parameters for the federated learning process, the federated learning application, the desired accuracy level, the type of the machine learning model, the number of local training rounds, the batch size, and / or the learning rate are received from the network node.

7. The method of claim 6, further comprising: Reporting an updated machine learning processing capability in response to a change in at least one of the parameters.

8. The method of claim 4, wherein: The estimated turnaround time is based on a battery status of the UE.

9. The method of claim 8, wherein: The estimated turnaround time is set to infinity in response to the UE operating in a power save mode.

10. A method for wireless communication by a network node, comprising: transmitting a machine learning model to a plurality of user equipments (UEs); receiving a machine learning processing capability report from each of the plurality of UEs, the machine learning processing capability report indicating a turnaround time for each of the plurality of UEs to calculate a gradient for the machine learning model; as well as The multiple UEs are grouped according to a machine learning processing capability report from each of the multiple UEs for receiving gradient updates to the machine learning model.

11. The method according to claim 10, wherein: The grouping further includes scheduling UEs having a first processing capability to transmit the gradient update in a first time period and scheduling UEs having a second processing capability to transmit the gradient update in a second time period.

12. An apparatus for wireless communication by a user equipment (UE), comprising: at least one processor; a memory coupled to the at least one processor; as well as instructions stored in the memory and operable, when executed by the at least one processor, to cause the apparatus to: Receive machine learning models from network nodes; reporting a machine learning processing capability to the network node, the machine learning processing capability indicating a turnaround time for the UE to calculate a gradient for the machine learning model; calculating a gradient update or a weight update to the machine learning model based on the turnaround time; as well as The gradient update or the weight update to the machine learning model is transmitted to the network node.

13. The device of claim 12, wherein: The machine learning processing capabilities include machine learning hardware capabilities.

14. The apparatus of claim 13, wherein: The machine learning hardware capabilities include manufacturer-specified hardware capabilities.

15. The apparatus of claim 12, wherein: The machine learning processing capability includes an estimated turnaround time for calculating the gradient.

16. The apparatus of claim 15, wherein: The estimated turnaround time is based on the federated learning process, the federated learning application, the desired accuracy level of the local training, the type of the machine learning model, the number of local training rounds, the batch size configured for UE training, and / or the learning rate of the local training.

17. The apparatus of claim 16, wherein: The at least one processor causes the apparatus to receive, from the network node, parameters for the federated learning process, the federated learning application, the desired level of accuracy, the type of the machine learning model, the number of local training rounds, the batch size, and / or the learning rate.

18. The apparatus of claim 17, wherein: The at least one processor causes the device to report updated machine learning processing capabilities in response to a change in at least one of the parameters.

19. The apparatus of claim 15, wherein: The estimated turnaround time is based on a battery status of the UE.

20. The apparatus of claim 19, wherein: The estimated turnaround time is set to infinity in response to the UE operating in a power save mode.

21. An apparatus for wireless communication by a network node, comprising: at least one processor; a memory coupled to the at least one processor; as well as instructions stored in the memory and operable, when executed by the at least one processor, to cause the apparatus to: transmitting a machine learning model to a plurality of user equipments (UEs); receiving a machine learning processing capability report from each of the plurality of UEs, the machine learning processing capability report indicating a turnaround time for each of the plurality of UEs to calculate a gradient for the machine learning model; as well as The multiple UEs are grouped according to a machine learning processing capability report from each of the multiple UEs for receiving gradient updates to the machine learning model.

22. The apparatus of claim 21, wherein: The at least one processor causes the apparatus to schedule UEs having a first processing capability to transmit the gradient update in a first time period and to schedule UEs having a second processing capability to transmit the gradient update in a second time period.

23. A user equipment (UE) for wireless communication, comprising: means for receiving a machine learning model from a network node; means for reporting a machine learning processing capability to the network node, the machine learning processing capability indicating a turnaround time for the UE to calculate gradients for the machine learning model; means for computing gradient updates or weight updates to the machine learning model based on the turnaround time; as well as Means for transmitting gradient updates or weight updates to the machine learning model to the network nodes.

24. The UE according to claim 23, wherein: The machine learning processing capabilities include machine learning hardware capabilities.

25. The UE according to claim 24, wherein: The machine learning hardware capabilities include manufacturer-specified hardware capabilities.

26. The UE according to claim 23, wherein: The machine learning processing capability includes an estimated turnaround time for calculating the gradient.

27. The UE according to claim 26, wherein: The estimated turnaround time is based on the federated learning process, the federated learning application, the desired accuracy level of the local training, the type of the machine learning model, the number of local training rounds, the batch size configured for UE training, and / or the learning rate of the local training.

28. The UE according to claim 27, further comprising: Means for receiving, from the network node, parameters for the federated learning process, the federated learning application, the desired accuracy level, the type of the machine learning model, the number of local training rounds, the batch size, and / or the learning rate.

29. The UE of claim 28, further comprising: Means for reporting updated machine learning processing capabilities in response to a change in at least one of the parameters.

30. The UE according to claim 26, wherein: The estimated turnaround time is based on a battery status of the UE.

31. The UE according to claim 30, wherein: The estimated turnaround time is set to infinity in response to the UE operating in a power save mode.

32. A network node for wireless communication, comprising: means for transmitting a machine learning model to a plurality of user equipments (UEs); means for receiving a machine learning processing capability report from each of the plurality of UEs, the machine learning processing capability report indicating a turnaround time for each of the plurality of UEs to calculate gradients for the machine learning model; as well as Means for grouping the plurality of UEs for receiving gradient updates to the machine learning model based on a machine learning processing capability report from each of the plurality of UEs.

33. The network node of claim 32, further comprising: Means for scheduling UEs having a first processing capability to transmit the gradient update in a first time period and scheduling UEs having a second processing capability to transmit the gradient update in a second time period.

34. A non-transitory computer-readable medium having program code stored thereon, wherein the program code, when executed by at least one processor, causes the at least one processor to perform the method of any one of claims 1-9.

35. A non-transitory computer-readable medium having program code stored thereon, wherein the program code, when executed by at least one processor, causes the at least one processor to perform the method of any one of claims 10-11.

36. A system comprising a user equipment according to any one of claims 23-31 and a network node according to any one of claims 32-33.

Images