Learning model selection at a device

Abstract

Various aspects of the present disclosure relate to learning model selection at a device. A first device, such as a user equipment (UE) and / or a network equipment (NE), obtains a set of data samples to provide as input to a set of learning models stored at the first device. The first device generates respective confidence scores associated with outputs from the set of learning models by providing the set of data samples as input to the set of learning models. The first device selects a learning model of the set of learning models for executing at the first device or at a second device. For example, the first device selects a learning model with a confidence score generated using an output from the learning model that satisfies a selection criterion.

Description

LEARNING MODEL SELECTION AT A DEVICERELATED APPLICATION

[0001] This application claims priority to U.S. Provisional Application Serial No. 63 / 633,524 filed April 12, 2024, entitled “LEARNING MODEL SELECTION AT A DEVICE,” the disclosure of which is incorporated by reference herein in its entirety.TECHNICAL FIELD

[0002] The present disclosure relates to wireless communications, and more specifically to learning model techniques for classification.BACKGROUND

[0003] A wireless communications system may include one or multiple network communication devices, which may be otherwise known as network equipment (NE), supporting wireless communications for one or multiple user communication devices, which may be otherwise known as user equipment (UE), or other suitable terminology. The wireless communications system may support wireless communications with one or multiple user communication devices by utilizing resources of the wireless communication system (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers, or the like)). Additionally, the wireless communications system may support wireless communications across various radio access technologies including third generation (3G) radio access technology, fourth generation (4G) radio access technology, fifth generation (5G) radio access technology, among other suitable radio access technologies beyond 5G (e.g., sixth generation (6G)).SUMMARY

[0004] An article “a” before an element is unrestricted and understood to refer to “at least one” of those elements or “one or more” of those elements. The terms “a,” “at least one,” “one or more,” and “at least one of one or more” may be interchangeable. As used herein, including in the claims, “or” as used in a list of items (e.g., a list of items prefaced by a phrase such as “at least one of’ or “one or more of’ or “one or both of’) indicates an inclusive list such that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (i.e., A and B and C). Also,as used herein, the phrase “based on” shall not be construed as a reference to a closed set of conditions. For example, an example step that is described as “based on condition A” may be based on both a condition A and a condition B without departing from the scope of the present disclosure. In other words, as used herein, the phrase “based on” shall be construed in the same manner as the phrase “based at least in part on.” Further, as used herein, including in the claims, a “set” may include one or more elements.

[0005] A first device for wireless communication is described. The first device may be configured to, capable of, or operable to perform one or more operations as described herein. For example, the first device may be configured to, capable of, or operable to obtain a set of data samples to provide as input to a set of learning models stored at the first device, generate, based on providing the set of data samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models, and select, for executing at the first device or at a second device, a learning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion.

[0006] A processor (e.g., a standalone processor chipset, or a component of a first device) for wireless communication is described. The processor may be configured to, capable of, or operable to perform one or more operations as described herein. For example, the processor may be configured to, capable of, or operable to obtain a set of data samples to provide as input to a set of learning models stored at the first device, generate, based on providing the set of data samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models, and select, for executing at the first device or at a second device, a learning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion.

[0007] A method performed or performable by a first device for wireless communication is described. The method may include obtaining a set of data samples to provide as input to a set of learning models stored at the first device, generating, based on providing the set of data samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models, and selecting, for executing at the first device or at a second device, a learning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion.

[0008] In some implementations of the first device, the processor, and the method described herein, the first device, the processor, and the method may further be configured to, capable of, or operable to transmit, to the second device, a message that indicates the selected learning model. In some implementations of the first device, the processor, and the method described herein, to obtain the set of data samples, the first device, the processor, and the method may further be configured to, capable of, or operable to receive, from the second device, a message that indicates the set of data samples. In some implementations of the first device, the processor, and the method described herein, to obtain the set of data samples, the first device, the processor, and the method may further be configured to, capable of, or operable to receive, from the second device, a set of reference signals, and obtains information associated with the set of reference signals, where the information includes one or more of a reference signal received power (RSRP) measurement, a reference signal received quality (RSRQ) measurement, a channel quality indicator (CQI), a precoding matrix indicator (PMI), a modulation and coding scheme (MCS) index, or a rank indicator (RI). In some implementations of the first device, the processor, and the method described herein, the first device, the processor, and the method may further be configured to, capable of, or operable to receive, from the second device, a message that triggers selection of the learning model. In some implementations of the first device, the processor, and the method described herein, to generate the respective confidence scores, the first device, the processor, and the method may further be configured to, capable of, or operable to generate respective measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a minimum measure of uncertainty of the respective measures of uncertainty. In some implementations of the first device, the processor, and the method described herein, the respective measures of uncertainty include at least one of a marginal entropy of the set of outputs or a conditional entropy of the set of outputs based on the set of data samples.

[0009] In some implementations of the first device, the processor, and the method described herein, to generate the respective confidence scores, the first device, the processor, and the method may further be configured to, capable of, or operable to generate respective first measures of uncertainty associated with the set of outputs from the set of learning models and respective second measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a maximum difference between therespective first measures of uncertainty and the respective second measures of uncertainty. In some implementations of the first device, the processor, and the method described herein, to generate the respective confidence scores, the first device, the processor, and the method may further be configured to, capable of, or operable to generate respective first weighted measures of uncertainty associated with the set of outputs from the set of learning models and respective second weighted measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the respective first weighted measures of uncertainty and the respective second weighted measures of uncertainty are weighted based on a value between zero and one. In some implementations of the first device, the processor, and the method described herein, to generate the respective confidence scores, the first device, the processor, and the method may further be configured to, capable of, or operable to generate an information gain associated with the set of outputs from the set of learning models based on the set of data samples.

[0010] In some implementations of the first device, the processor, and the method described herein, to execute the learning model, the first device, the processor, and the method may further be configured to, capable of, or operable to generate, as the output from the learning model, a prediction of at least one of a message, a set of symbols, or a set of bits based on providing the set of data samples as input to the selected learning model, where the set of data samples include one or more channel output signals, generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling in a set of future time slots based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, or generates, as the output from the learning model, a prediction of channel state information (CSI) based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of reference signal measurements or a set of channel vectors, and where the CSI includes at least one of a CQI associated with a set of time-frequency resources, a PMI, an MCS index, or an RI. In some implementations of the first device, the processor, and the methoddescribed herein, the set of channel vectors are associated with a first subset of time-frequency resources of a set of time-frequency resources, and where the output from the learning model is associated with a second subset of time-frequency resources of the set of time-frequency resources or the set of time-frequency resources. In some implementations of the first device, the processor, and the method described herein, the first device is an NE. In some implementations of the first device, the processor, and the method described herein, the first device is a UE.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figures 1 and 2 illustrate examples of wireless communications systems in accordance with aspects of the present disclosure.

[0012] Figure 3 illustrates an example of a signaling diagram, in accordance with aspects of the present disclosure.

[0013] Figure 4 illustrates an example of a UE in accordance with aspects of the present disclosure.

[0014] Figure 5 illustrates an example of a processor in accordance with aspects of the present disclosure.

[0015] Figure 6 illustrates an example of an NE in accordance with aspects of the present disclosure.

[0016] Figure 7 illustrates a flowchart of a method performed by a device (e.g., a UE and / or an NE) in accordance with aspects of the present disclosure.DETAILED DESCRIPTION

[0017] A wireless communications system may include one or more devices, such as UEs and NEs, among other devices that transmit and receive signaling. The devices can implement one or more learning models, which may also be referred to as machine learning (ML) models and / or artificial intelligence (Al) models, to perform classification tasks related to the signaling. The learning model can identify patterns in input data that are indicative of different classes or labels. Example classification tasks related to signaling include, but are not limited to, predicting a type of message from a received signal, predicting one or more beam indices with a greatest signal strengthand / or quality from one or more beam measurements, and predicting information (e.g., a CQI, a PMI, an RI, or an MCS index) related to the signaling, among other classification tasks. In some examples, an NE or other device trains the learning models by determining one or more parameters of the learning models, which can include weights, biases, and / or other parameters that define respective layers of the learning models. During the training of the learning model, parameters are selected, such that the learning model maps (e.g., associates) an input value to an output value (e.g., a class or label of a discrete numerical quantity of classes and labels). The learning models are executed (e.g., implemented) by one or more devices using the determined or selected parameters.

[0018] In some variations, the devices can input data to the learning models that has different statistical characteristics than the dataset over which the learning models are trained. In some cases, the NE can train multiple different learning models for respective domains over which the learning models make inferences or predictions, where a domain represents data for a network scenario, a network configuration, one or more parameters of a network, a physical propagation medium, and / or a device behavior. For example, a domain can include a dataset for a network configuration (e.g., a defined beam codebook, a defined set of multiple input multiple output (MIMO) configuration parameters, a type of scheduling device, different types of link adaptation and power control), for a device behavior (e.g., a device mobility, an orientation of the device, a quality of service (QoS) to support communications at the device), for different traffic patterns, for physical conditions of the propagation medium (e.g., channels with rich multipath vs. sparse channels, indoor vs. outdoor), among other examples, at the time of data collection or at the time of a simulation to generate the data in the dataset. The NE can store the different learning models for transmission to one or more other devices (e.g., UEs with reduced memory storage capabilities when compared with the NE). Additionally, or alternatively, the one or more other devices can store the different learning models.

[0019] A device (e.g., the other devices and / or the NE) can use respective learning models for the different domains. However, conventional techniques for the device to select a stored learning model to use can result in additional signaling overhead and / or processing. For example, the device can exchange signaling that includes labeled data samples for a data domain experienced by the device to provide as input to the stored learning models. The device can compare an output from the learning models to an expected output (e.g., the label of the data samples) and can select a learningmodel with a closest output. The device exchanging signaling that includes the labeled data samples results in an increase in signaling overhead. Additionally, or alternatively, the device can obtain data samples (e.g., by measurement and / or receiving the data samples from a device) for a data domain experienced by the device, and the device can determine the statistical characteristics for the data samples. The device can select a learning model that was trained with training data that has similar statistical characteristics as the data samples. However, processing the data samples to determine the statistical characteristics and / or obtaining the statistical characteristics of the training data can lead to increased processing, delays, memory usage, and signaling overhead.

[0020] As described herein, to reduce signaling overhead and processing related to selection of learning models for different domains, a device can autonomously select a learning model, which includes selecting a learning model without receiving or transmitting signaling from or to another device. For example, the device can generate confidence scores for different learning models using data samples obtained for input to a learning model. The device can generate the confidence score as a function of respective conditional entropies for the different learning models given the data samples and respective marginal entropies for the different learning models. Entropy refers to a measure of uncertainty or disorder in a set of data samples. A conditional entropy for a learning model is an indication of a confidence in individual predictions made by the learning model, where a lower conditional entropy reflects lower uncertainty in the predictions and correspondingly a higher confidence score for the learning model. A marginal entropy for a learning model is an indication of a confidence in a marginal distribution of one or more outputs from the learning model (e.g., predicted labels for input data samples) being a uniform distribution, where a marginal distribution refers to a distribution of a single variable or the predictions of the learning model on a single variable without considering the values of other variables or features.

[0021] In some variations, the confidence score for a learning model is a difference between the conditional entropy for the learning model and the marginal entropy for the learning model and / or a difference between a weighted conditional entropy for the learning model and a weighted marginal entropy for the learning model. The device compares the confidence score to a selection criterion and selects a learning model that satisfies the selection criterion. Example selection criterion includes, but is not limited to, a confidence score of a learning model being a greatest (e.g., highest) confidence score for a set of learning models stored at the device, a difference between theconditional entropy and the marginal entropy for a learning model being a greatest difference for the set of learning models, or a conditional entropy for a learning model being a lowest (e.g., minimum) for the set of learning models, among other selection criterion. In some examples, the device transmits an indication of the selected learning model to another device (e.g., to a device that implements or executes the learning model). In some other examples, the device executes the selected learning model to obtain one or more predictions for input data samples.

[0022] Although the learning models are described as being trained and implemented for wireless communications, the learning models can additionally, or alternatively, be trained and implemented for any task that uses classification. Example tasks that use classification include, but are not limited to, wireless communications, image recognition, natural language processing, and bio-medical imaging, among other tasks. The learning models can be switched or communicated from one device to another device for implementation to perform the task.

[0023] Reference is made herein to receiving, transmitting, or communicating data or information, such as signaling communication resources and / or communications that are transmitted or received between devices. It is to be appreciated that other terms may be used interchangeably with communicating, such as signaling, transmitting, receiving, outputting, forwarding, retrieving, obtaining, and so forth. Similarly, other terms may be used interchangeably with transmitting (e.g., communicating, signaling, outputting, forwarding, and so forth), and other terms may be used interchangeably with receiving (e.g., communicating, retrieving, obtaining, and so forth).

[0024] Aspects of the present disclosure are described in the context of a wireless communications system.

[0025] Figure 1 illustrates an example of a wireless communications system 100 in accordance with aspects of the present disclosure. The wireless communications system 100 may include one or more NE 102, one or more UE 104, and a core network (CN) 106. The wireless communications system 100 may support various radio access technologies. In some implementations, the wireless communications system 100 may be a 4G network, such as an LTE network or an LTE- Advanced (LTE-A) network. In some other implementations, the wireless communications system 100 may be a NR network, such as a 5G network, a 5G-Advanced (5G-A) network, or a 5G ultrawideband (5G-UWB) network. In other implementations, the wireless communications system 100 may be acombination of a 4G network and a 5G network, or other suitable radio access technology including Institute of Electrical and Electronics Engineers (IEEE) 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802.20. The wireless communications system 100 may support radio access technologies beyond 5G, for example, 6G. Additionally, the wireless communications system 100 may support technologies, such as time division multiple access (TDMA), frequency division multiple access (FDMA), or code division multiple access (CDMA), etc.

[0026] The one or more NE 102 may be dispersed throughout a geographic region to form the wireless communications system 100. One or more of the NE 102 described herein may be or include or may be referred to as a network node, a base station, an access point (AP), a network element, a network function, a network entity, network infrastructure (or infrastructure), a radio access network (RAN), a NodeB, an eNodeB (eNB), a next-generation NodeB (gNB), or other suitable terminology. An NE 102 and a UE 104 may communicate via a communications link, which may be a wireless or wired connection. For example, an NE 102 and a UE 104 may perform wireless communication (e.g., receive signaling, transmit signaling) over a Uu interface.

[0027] An NE 102 may provide a geographic coverage area for which the NE 102 may support services for one or more UEs 104 within the geographic coverage area. For example, an NE 102 and a UE 104 may support wireless communication of signals related to services (voice, video, packet data, messaging, broadcast, etc.) according to one or multiple radio access technologies. In some implementations, an NE 102 may be moveable, for example, a satellite associated with a non-terrestrial network (NTN). In some implementations, different geographic coverage areas associated with the same or different radio access technologies may overlap, but the different geographic coverage areas may be associated with different NE 102.

[0028] The one or more UEs 104 may be dispersed throughout a geographic region of the wireless communications system 100. A UE 104 may include or may be referred to as a remote unit, a mobile device, a wireless device, a remote device, a subscriber device, a transmitter device, a receiver device, or some other suitable terminology. In some implementations, the UE 104 may be referred to as a unit, a station, a terminal, or a client, among other examples. Additionally, or alternatively, the UE 104 may be referred to as an Internet-of-Things (loT) device, an Internet-of- Everything (loE) device, or machine-type communication (MTC) device, among other examples.

[0029] A UE 104 may be able to support wireless communication directly with other UEs 104 over a communications link. For example, a UE 104 may support wireless communication directly with another UE 104 over a device-to-device (D2D) communications link. In some implementations, such as vehicle-to-vehicle (V2V) deployments, vehicle-to-everything (V2X) deployments, or cellular-V2X deployments, the communications link may be referred to as a sidelink. For example, a UE 104 may support wireless communication directly with another UE 104 over a PC5 interface.

[0030] An NE 102 may support communications with the CN 106, or with another NE 102, or both. For example, an NE 102 may interface with other NE 102 or the CN 106 through one or more backhaul links (e.g., SI, N2, N6, or other network interface). In some implementations, the NE 102 may communicate with each other directly. In some other implementations, the NE 102 may communicate with each other indirectly (e.g., via the CN 106). In some implementations, one or more NE 102 may include subcomponents, such as an access network entity, which may be an example of an access node controller (ANC). An ANC may communicate with the one or more UEs 104 through one or more other access network transmission entities, which may be referred to as a radio heads, smart radio heads, or transmission-reception points (TRPs).

[0031] The CN 106 may support user authentication, access authorization, tracking, connectivity, and other access, routing, or mobility functions. The CN 106 may be an evolved packet core (EPC), or a 5G core (5GC), which may include a control plane entity that manages access and mobility (e.g., a mobility management entity (MME), an access and mobility management functions (AMF)) and a user plane entity that routes packets or interconnects to external networks (e.g., a serving gateway (S-GW), a packet data network (PDN) gateway (P-GW), or a user plane function (UPF)). In some implementations, the control plane entity may manage non-access stratum (NAS) functions, such as mobility, authentication, and bearer management (data bearers, signal bearers, etc.) for the one or more UEs 104 served by the one or more NE 102 associated with the CN 106.

[0032] The CN 106 may communicate with a packet data network over one or more backhaul links (e.g., via an SI, N2, N6, or other network interface). The packet data network may include an application server. In some implementations, one or more UEs 104 may communicate with the application server. A UE 104 may establish a session (e.g., a protocol data unit (PDU) session, orthe like) with the CN 106 via an NE 102. The CN 106 may route traffic (e.g., control information, data, and the like) between the UE 104 and the application server using the established session (e.g., the established PDU session). The PDU session may be an example of a logical connection between the UE 104 and the CN 106 (e.g., one or more network functions of the CN 106).

[0033] In the wireless communications system 100, the NEs 102 and the UEs 104 may use resources of the wireless communications system 100 (e.g., time resources (e.g., symbols, slots, subframes, frames, or the like) or frequency resources (e.g., subcarriers, carriers)) to perform various operations (e.g., wireless communications). In some implementations, the NEs 102 and the UEs 104 may support different resource structures. For example, the NEs 102 and the UEs 104 may support different frame structures. In some implementations, such as in 4G, the NEs 102 and the UEs 104 may support a single frame structure. In some other implementations, such as in 5G and among other suitable radio access technologies, the NEs 102 and the UEs 104 may support various frame structures (i.e., multiple frame structures). The NEs 102 and the UEs 104 may support various frame structures based on one or more numerologies.

[0034] One or more numerologies may be supported in the wireless communications system 100, and a numerology may include a subcarrier spacing and a cyclic prefix. A first numerology (e.g., / r=0) may be associated with a first subcarrier spacing (e.g., 15 kHz) and a normal cyclic prefix. In some implementations, the first numerology (e.g., / r=0) associated with the first subcarrier spacing (e.g., 15 kHz) may utilize one slot per subframe. A second numerology (e.g., / r=l) may be associated with a second subcarrier spacing (e.g., 30 kHz) and a normal cyclic prefix. A third numerology (e.g., / r=2) may be associated with a third subcarrier spacing (e.g., 60 kHz) and a normal cyclic prefix or an extended cyclic prefix. A fourth numerology (e.g., / r=3) may be associated with a fourth subcarrier spacing (e.g., 120 kHz) and a normal cyclic prefix. A fifth numerology (e.g., / r=4) may be associated with a fifth subcarrier spacing (e.g., 240 kHz) and a normal cyclic prefix.

[0035] A time interval of a resource (e.g., a communication resource) may be organized according to frames (also referred to as radio frames). Each frame may have a duration, for example, a 10 millisecond (ms) duration. In some implementations, each frame may include multiple subframes. For example, each frame may include 10 subframes, and each subframe mayhave a duration, for example, a 1 ms duration. In some implementations, each frame may have the same duration. In some implementations, each subframe of a frame may have the same duration.

[0036] Additionally, or alternatively, a time interval of a resource (e.g., a communication resource) may be organized according to slots. For example, a subframe may include a number (e.g., quantity) of slots. The number of slots in each subframe may also depend on the one or more numerologies supported in the wireless communications system 100. For instance, the first, second, third, fourth, and fifth numerologies (i.e., / r=0, jU=l , / r=2, / r=3, / r=4) associated with respective subcarrier spacings of 15 kHz, 30 kHz, 60 kHz, 120 kHz, and 240 kHz may utilize a single slot per subframe, two slots per subframe, four slots per subframe, eight slots per subframe, and 16 slots per subframe, respectively. Each slot may include a number (e.g., quantity) of symbols (e.g., OFDM symbols). In some implementations, the number (e.g., quantity) of slots for a subframe may depend on a numerology. For a normal cyclic prefix, a slot may include 14 symbols. For an extended cyclic prefix (e.g., applicable for 60 kHz subcarrier spacing), a slot may include 12 symbols. The relationship between the number of symbols per slot, the number of slots per subframe, and the number of slots per frame for a normal cyclic prefix and an extended cyclic prefix may depend on a numerology. It should be understood that reference to a first numerology (e.g., / r=0) associated with a first subcarrier spacing (e.g., 15 kHz) may be used interchangeably between subframes and slots.

[0037] In the wireless communications system 100, an electromagnetic (EM) spectrum may be split, based on frequency or wavelength, into various classes, frequency bands, frequency channels, etc. By way of example, the wireless communications system 100 may support one or multiple operating frequency bands, such as frequency range designations FR1 (410 MHz - 7.125 GHz), FR2 (24.25 GHz - 52.6 GHz), FR3 (7.125 GHz - 24.25 GHz), FR4 (52.6 GHz - 114.25 GHz), FR4a or FR4-1 (52.6 GHz - 71 GHz), and FR5 (114.25 GHz - 300 GHz). In some implementations, the NEs 102 and the UEs 104 may perform wireless communications over one or more of the operating frequency bands. In some implementations, FR1 may be used by the NEs 102 and the UEs 104, among other equipment or devices for cellular communications traffic (e.g., control information, data). In some implementations, FR2 may be used by the NEs 102 and the UEs 104, among other equipment or devices for short-range, high data rate capabilities.

[0038] FR1 may be associated with one or multiple numerologies (e.g., at least three numerologies). For example, FR1 may be associated with a first numerology (e.g., / r=0), whichincludes 15 kHz subcarrier spacing; a second numerology (e.g., / r=l), which includes 30 kHz subcarrier spacing; and a third numerology (e.g., / r=2), which includes 60 kHz subcarrier spacing. FR2 may be associated with one or multiple numerologies (e.g., at least 2 numerologies). For example, FR2 may be associated with a third numerology (e.g., / r=2), which includes 60 kHz subcarrier spacing; and a fourth numerology (e.g., / r=3), which includes 120 kHz subcarrier spacing.

[0039] In some examples, a device (e.g., an NE 102 and / or a UE 104) can implement one or more learning models for classification tasks. Example classification tasks include, but are not limited to, predicting a type of message from a received signal, predicting one or more beam indices with a greatest signal strength and / or quality from one or more beam measurements, and predicting information (e.g., a CQI, PMI, RI, or a MCS index) related to the signaling, among other classification tasks. A classifier learning model and / or a learning model implemented (e.g., executed) for a classification task can identify patterns in input data that are indicative of at least one class or label from a finite set of classes or labels.

[0040] In some cases, a device (e.g., an NE 102 and / or a UE) is equipped with multiple antennas, such as in the form of one or more antenna array panels with respective antenna elements. In some variations, the devices can communicate using one or more beams formed using multiple antenna elements available at the devices. The beams are directional in nature (e.g., providing directional gain) and can have a narrow beam width (e.g., a beam width that is less than a threshold value). For achieving a reliable communications link between the devices, one or more of the devices can select a beam to use. That is, the devices can select a beam pair that includes a receive beam for a receiving device to use for receiving signaling and a transmit beam for a transmitting device to use for transmitting signaling, such that the beam pair results in a signal strength at the receiving device that satisfies a threshold value.

[0041] In some examples, the process of selecting a beam pair can be referred to as a beam search procedure or a beam selection procedure. For signaling from a UE 104 to an NE 102 (e.g., downlink signaling), the beam selection includes selecting a transmit beam at the NE 102 and a corresponding receive beam at the UE 104. For signaling from the NE 102 to the UE 104 (e.g., uplink signaling), the beam selection includes selecting a transmit beam at the UE 104 and a corresponding receive beam at the NE 102. A beam corresponds to a non-zero power (NZP) CSI-reference signal (CSI-RS) resource of at least one NZP CSI-RS resource set, where the NZP CSI- RS resource set is configured with a value of a higher-layer parameter set to repetition.

[0042] Conventionally, for beam selection, a device (e.g., the UE 104 and / or the NE 102) can search over a set of possible beams and select a beam with a maximum signal strength, which is referred to as an exhaustive search. In some examples, such as for downlink signaling, an NE 102 can send one or more reference signals using a set of possible transmit beams. A UE 104 can perform one or more measurements to obtain a received signal strength (e.g., an RSRP, an RSRQ, and / or a signal-to-interference -plus-noise ratio (SINR)) for the beams in the set of possible transmit beams. The UE 104 can select a transmit beam that provides a highest RSRP and / or SINR. The UE 104 can transmit a message to the NE 102 that includes an indication of a beam index for the selected transmit beam. Additionally, or alternatively, the UE 104 can select a receive beam by fixing the transmit beam at the NE 102 and measuring the received signal strength for a set of possible receive beams (e.g., sweeping the receive beams at the UE 104). Thus, an exhaustive search includes selecting a beam (e.g., a best beam, a beam with a highest signal strength), but increases latency and signaling overhead related to transmitting and / or receiving signaling on the set of possible beams for performing measurements.

[0043] In some examples, a device (e.g., the UE 104 and / or the NE 102) can implement one or more learning models for beam prediction. For example, the device can train a learning model to output a prediction of a transmit and / or receive beam to use for reliable transmission or reception of signaling. The device can input one or more beam measurements to the learning model to obtain the prediction of the transmit and / or receive beam. Example beam measurements include, but are not limited to, beamforming gain measurements, beamwidth measurements, RSRP measurements, and SINR measurements, among other measurements. The beam prediction can include spatial beam prediction, in which the device implements the learning model to determine a beam of a set of available or candidate beams in a wireless network using a relatively small numerical quantity (e.g., less than a threshold numerical quantity) of beam measurements. For example, the numerical quantity of beam measurements can be less than a numerical quantity of available or candidate beams. Additionally, or alternatively, the beam prediction can include temporal beam prediction, in which the device implements the learning model to determine a beam to use for transmission during one or more subsequent time slots using a relatively small numerical quantity (e.g., less than athreshold numerical quantity) of historical beam measurements or identifiers of beams that are historically selected for transmitting and / or receiving signaling.

[0044] In some variations, the device implements the learning model to predicting information (e.g., a CQI, a PMI, a RI, or a MCS index). For example, the device implements the learning model to estimate CSI across a space-time grid based on pilot and / or reference signals sent over a relatively small numerical quantity (e.g., less than a threshold numerical quantity) of resource elements in the space-time grid. The device implements the learning model to perform CSI prediction, such that the learning model outputs a prediction of CSI (e.g., the CQI, the PMI, the RI, or the MCS index) for one or more future time slots based on the estimated CSI and / or additional information for one or more previous time slots. Additionally, or alternatively, the UE 104 and / or the NE 102 implements the learning model to function as a receiver. For example, the learning model outputs a prediction of a transmitted message, symbol, or bit based on one or more signals received as output from a wireless channel. In some examples, the device can implement one or more learning models for positioning procedures. For example, the device implements the learning model to determining a position of a node (e.g., a UE 104, an NE 102, or any other device in the wireless communications system 100) based on a channel impulse response (CIR) experienced by the node. A CIR defines how a signal propagates through the channel over time.

[0045] A device can train the learning models using a training dataset obtained either through simulations or through measurements (e.g., considering either one or, at most, a finite set of physical cell-sites, network configurations, and / or channel characteristics). When a dataset includes both an input data sample and a corresponding output data sample for data samples in the dataset, the dataset is referred to as a labeled dataset. When the dataset includes the input data sample without the corresponding output data sample, the dataset is referred to as an unlabeled dataset. In some examples, a device can implement supervised learning techniques to train the learning models. For example, the device can train the learning models to map input data to output labels based on example input-output pairs provided during training. In supervised learning, the learning model computes a mapping function from input features to output labels by observing a dataset that includes labeled examples, such that the learning models can generalize the mapping to make accurate predictions on new, unseen data. Although supervised learning techniques are described, the device can additionally, or alternatively, implement any other type of training techniques to trainthe learning models, including, but not limited to, unsupervised learning techniques and semisupervised learning techniques, among others. In unsupervised learning, the learning models detects patterns, structures, or relationships within training data. Unlike supervised learning, there are no explicit output labels provided during training. Semi-supervised learning leverages both labeled and unlabeled data during a training procedure. The learning models use the labeled examples, while also using the structure of the unlabeled data to improve a performance of the learning models.

[0046] In some examples xLE X and y;G y denote an input sample and a corresponding label or desired output data sample from a learning model for an input x(. In some variations, y(is referred to as a prediction and / or inference for input x(. In some cases, x(is a scalar or a one or multi-dimensional vector and y^ is a scalar or a one or multi-dimensional vector. X and y denote the input sample space and the output sample space (e.g., label space), respectively. If y(assumes finitely many values (e.g., when \y\, the cardinality of the set y, is finite), then the learning model is referred to as a classifier learning model. If y(assumes continuous values (e.g., when y £ K. and \y | = oo), then the learning model is referred to as a regression model.

[0047] A learning model defines a mapping or a function, / e, where fe- X - y. Here, 0 6 0 denotes a set of learning model parameters that are learned during the process of training the learning model. Example learning model parameters include, but are not limited to, weights, biases, and activation function parameters, among other parameters. Weight parameters represent a strength of connections between neurons in different layers of the learning model. Biases are additional parameters added to neurons in the learning model that provide for the learning model to capture offset or bias in input data. Activation function parameters can include slope parameters or parameters defining a shape of an activation function in a parametric activation function.Determining one or more values of model parameters for a defined use case is referred to as training the model or learning the model. The general procedure of developing a learning model includes updating one or more parameters to minimize a cost function based on a training dataset that includes either labeled samples or unlabeled samples, resulting in supervised learning or training or unsupervised learning or training, respectively. For a classifier learning model, a label (e.g., desired output, prediction, and / or inference) of the learning model is discrete and belongs to a set of finite cardinalitythe

[0048] In some examples, a classifier learning model can be probabilistic, where the learning model generates a probability distribution over one or more predictions. The probability distribution can be conditioned on input data (e.g., data samples provided as input to the learning model) and parameterized by 0 to obtain a conditional distribution that is differentiable in 0. Thus, training the classification learning model (e.g., on one or more source domains) generates a probability distribution P(y|x; 0). As the parameters of the learning model, 0, are fixed at the end of training, the probability distribution over the predictions generated by the learning model can be represented as (y|x) (e.g., without 0) and is conditioned on input data. As y is a discrete variable with y G {1, , M], P(y|x) is the probability that y is the predicted label for x when x is provided as input to a learning model (e.g., using learning model parameters, 0, that are fixed at the end of training the learning model). That is, P(y xj) is the probability that ycis the label for the given input Xj, as per a prediction and / or inference made by the learning model f . In some variations,

[0049] In some examples, a device can implement or deploy a trained learning model, such as to make predictions or inferences. The learning model can be expected to perform with a same level of accuracy and / or precision by providing desired inferences or predictions as seen during the training and testing phase of the learning models before deployment. However, in a real-world wireless network (e.g., in the wireless communications system 100), a learning model can make predictions or inferences using input data that has different statistical characteristics than the dataset over which the learning model is trained. A learning model that uses input data with different statistical characteristics than the dataset over which the learning model is trained can output erroneous (e.g., incorrect) inferences or predictions. To improve accuracy of the inferences and / or the predictions, a device can train a learning model with a generalized ability to make inferences and / or predictions over many different data distributions (e.g., data distributions that are different from training data distributions and encountered at a time of inference). However, developing a learning model that generalizes the possible domains and outputs a desired performance across the domains may be difficult for the device, especially, in the context of wireless networks with varying statistics of data distributions.

[0050] In some examples, there are T > 1 distinct data distributions over which the learning model provides inferences and / or predictions with a performance (e.g., precision and / or accuracy)that satisfies a threshold performance. A set of data samples, Dk=according to a statistical distribution, ~ PkY, is also referred to as a data domain or a domain. Thus, the term domain implies data with a defined distribution. For example, )k= {(x^, y )} ~ PkYcan be referred to as a kthdata domain or kthdomain. A source domain is a domain over which the learning model is trained, and a target domain is a domain over which the learning model is executed to provide inferences and / or predictions. In some examples, there can be different target domains for different network conditions, device parameters, channel characteristics, or any other factor that impacts the data distribution of one or more data samples. In some cases, a device can train a respective learning model for the T data distributions (e.g., T target domains). A device can train or develop T different learning models for a task (e.g., beam prediction, CSI prediction, and / or positioning based on CSI, among other classification tasks). The device can store the learning models and / or can transmit an indication of the learning models to another device for storage. At any instant of time, the device and / or the other device can select and execute one of the T models for making inferences and / or predictions.

[0051] However, conventional techniques for the device to select a stored learning model to use can result in additional signaling overhead and / or processing. The device can exchange signaling that includes labeled data samples for a data domain experienced by the device to provide as input to the stored learning models. For example, a device hosting multiple learning models obtains a labeled data sample (x, y), where x is an input data sample and y is a corresponding label. The device can provide x as input to respective learning models stored at the device. The device can select an output using the learning model output. The device can compare an output from the learning models to an expected output (e.g., the label of the data samples) and can select a learning model with a closest output. The device exchanging signaling that includes the labeled data samples results in an increase in signaling overhead. Additionally, or alternatively, the device can obtain data samples (e.g., by measurement and / or receiving the data samples from a device) for a data domain experienced by the device, and the device can determine the distribution and / or statistical characteristics for the data samples. The device can select a learning model that was trained with training data that has a similar distribution and / or statistical characteristics as the data samples. However, processing the data samples to determine the statistical characteristics and / or obtainingthe statistical characteristics of the training data can lead to increased processing, delays, memory usage, and signaling overhead.

[0052] Thus, conventional techniques may lead to inefficiencies (increased signaling overhead, increased processing, etc.) when the device is selecting a learning model from the T learning models with a performance that satisfies a threshold performance for a current data distribution and / or domain. For example, the device can select a learning model from the available (e.g., stored) T learning models that results in a highest inference and / or prediction performance, including, but not limited to, accuracy and precision, for the input data samples obtained by the device. As the data distributions are dynamic (e.g., change relatively frequently) in a wireless communications system 100 due to the time-varying nature of the physical propagation medium and other network parameters and / or settings that change over different time intervals, the task of learning model selection is not a one-time task. That is, the device can perform learning model selection when a network condition and / or scenario, including channel characteristics, changes and / or when a performance of a previously selected learning model degrades beyond a satisfactory level. Example channel characteristics can include, but are not limited to, attenuation, propagation delay, multipath fading, noise and interference, bandwidth, channel capacity, and Doppler shift, among others characteristics. Example network conditions include, but are not limited to, device mobility, environmental factors for a channel (e.g., weather conditions and obstruction of objects, among other environmental factors), and / or changes to antenna configurations at one or more devices.

[0053] According to implementations, one or more of the NEs 102 and the UEs 104 are operable to implement various aspects of the techniques described with reference to the present disclosure. In some examples, the device can implement techniques for autonomous model selection. The device can select a learning model from a set of learning models stored at the device for a target domain without receiving signaling and / or transmitting signaling to another device (e.g., signaling including labeled data samples, distribution and / or other statistical information about a target domain, and / or reference signals, among other information). The learning models stored at the device can be developed and / or trained as classifier learning models (e.g., T classifier learning models). A classifier learning model sorts respective input data sample to one of M classes, where M is a positive integer greater than one. Examples classifier learning models include, but are not limited to, a learning model that is implemented as a receiver, a learning model for beam prediction(e.g., spatial beam prediction and / or temporal beam prediction), a learning model for predicting information about signaling (e.g., a CQI, a PMI, a RI, and / or an MCS index, among other information) as long as the output of the learning model (e.g., the CQI, the PMI, the RI, and / or the MCS index) is discrete valued with a finite number of possible values. The device can select a learning model at a receiver to infer and / or predict transmitted messages, symbols, and / or bits from signals at the output of a channel. Additionally, or alternatively, the device can select a learning model (e.g., for a target domain) for predicting the information about the signaling. Additionally, or alternatively, the device can select a learning model (e.g., for a target domain) to predict one or more beams to use for signaling.

[0054] Figure 2 illustrates an example of a wireless communications system 200 in accordance with aspects of the present disclosure. In some examples, the wireless communications system 200 implements aspects of the wireless communications system 100. For example, the wireless communications system 200 includes a device 202-a and a device 202 -b, which may be examples of one or more UEs 104 and / or one or more NEs 102 as described with reference to Figure 1. In some examples, the devices (e.g., the device 202-a and the device 202-b) may exchange signaling with one another. For example, the device 202-a transmits signaling to the device 202-b via a communications link 204-a, which may be an example of a communications link as described with reference to Figure 1. In some other examples, the device 202-b transmits signaling to the device 202-a via a communications link 204-b, which may be an example of a communications link as described with reference to Figure 1. The signaling between the device 202-a and the device 202-b may include control signaling and / or data transmissions.

[0055] In some variations, the device 202-b is an example of an NE, while the device 202-a is an example of a UE. In some other variations, the device 202-b is an example of a UE, while the device 202-a is an example of an NE. Example UEs can include, but are not limited to, cellular devices, laptop computers, desktop computers, wearable devices, internet-of-things (loT) devices, and / or other equipment that are capable of transmitting and / or receiving communications.

[0056] The devices (e.g., the device 202-a and / or the device 202-b) in the wireless communications system 200 can implement one or more learning models (e.g., one or more of the learning model 206-a, the learning model 206-b, etc., through the learning model 206-T). In some examples, the learning model 206-a through the learning model 206-T can be examples of MLmodels and / or Al models. For example, the learning model 206-a through the learning model 206-T can be examples of neural networks (e.g., a deep neural network (DNN)) with multiple layers. A neural network is a computational model that includes multiple layers of artificial neurons, which may be referred to as nodes or units, organized into an input layer 208, one or more feature extraction layers 210, one or more classification layers 212, and an output layer 214. An input layer 208 passes input data to subsequent layers and may not include learnable parameters. An output layer 214 represents a prediction generated by the neural network and may not include learnable parameters. The feature extraction layers 210 and / or the classification layers 212 can include one or more hidden layers. A node in a hidden layer receives an input signal, performs a mathematical operation on the input data, and produces an output signal, which is then passed on to other nodes in subsequent layers. The connections between nodes are represented by weighted edges, which determine the strength of the connection between nodes. During a training process, the weights are adjusted using input-output pairs from a training dataset, with the goal of minimizing a defined loss or error function. Neural networks are capable of learning complex patterns and representations from data, enabling them to perform a wide range of tasks, including classification, regression, clustering, pattern recognition, and sequence generation. Example neural networks include, but are not limited to, an autoencoder, a convolutional neural network (CNN), a recurrent neural network (RNN), a generative adversarial network (GAN), a long short-term memory (LSTM) network or any other type of neural network.

[0057] In some examples, the learning model 206-a through the learning model 206-T can include site-specific learning models. For example, one or more target domains may correspond to site-specific scenarios, which leads to site-specific learning models and / or device specific learning models for device specific scenarios. For site-specific models, different learning models are developed for different cell-sites or geographic locations, which result in different data distributions due to their distinct physical characteristics. For example, a cell-site in a rural location can have different channel characteristics than a cell-site in a dense urban environment, thereby resulting in two different data distributions. In some examples, the learning models can include device specific learning models. The data distributions can change due to device specific parameters and / or scenarios. For example, the wireless channel experienced by a device that is in an outdoor environment can have different statistical characteristics than that of a device in an indoorenvironment. The mobility of a device and a physical orientation of the device can create different data distributions. Further, different devices having different capabilities, such as a measurement sensitivity of a device, can benefit from using multiple learning models.

[0058] A device can develop or train T learning models (e.g., the learning model 206-a through the learning model 206-T) where respective learning models have a threshold performance (e.g., a threshold accuracy and precision) over at least one target domain for a training dataset. The device 202-a and / or the device 202-b can implement the learning models for a classification task, such as for spatial beam prediction, temporal beam prediction, as a receiver, and / or predicting information for signaling (e.g., a CQI, a RI, a PMI, and / or a MCS index). A classification task can include any task that categorizes or labels input data as a defined class or category based on features or attributes of the input data. A learning model for spatial beam prediction can predict either a single beam with a highest performance or multiple beams with highest performances out of a set of beams using a current set of beam measurements. A learning model for temporal beam prediction can predict either a single beam with a highest performance or multiple beams with highest performances for a set of future time slots using current and past (e.g., historical) data. The current and past data can include current and past beam measurements and / or current and past beam indices with the highest performance. A learning model at a receiver infers or predicts transmitted messages, symbols, and / or bits based on the signals at the output of the wireless channel. A learning model can predict information (e.g., a CQI, a RI, PMI, and / or an MCS index) if the output of the learning model is a discrete value (e.g., where there are a finite number of values).

[0059] In some variations, the learning model 206-a through the learning model 206-T include respective feature components and respective classifier components. The feature components and the classifier components can represent sub-networks of a neural network and can include one or more layers. For example, the feature components can include an input layer 208 and one or more feature extraction layers 210, which may be examples of hidden layers. In some variations, one or more last layers of a learning model (e.g., the output layer 214 and / or a layer prior to an output layer 214) can include at least one classification layer 212. In some cases, a classifier component can include a classification layer 212 and an output layer 214. In some other cases, the classification layer 212 can be part of the output layer 214, such that the classifier component can include an output layer 214 that includes the classification layer 212. That is, the classification layer 212 canbe a specialized component within the output layer 214 that performs a final classification based on the learned features from the feature component. A classification layer 212 maps one or more learned features from one or more preceding layers (e.g., features extracted by the feature extraction layers 210 of the feature component) to one or more output classes or labels of the output layer 214. The output layer 214 can include neurons representing different classes or categories, and the activation of these neurons indicates the likelihood or probability of the input data belonging to a class. In some variations, the output from the feature component is used as input data and is provided to a classifier component of a learning model. The classifier component outputs a classification (e.g., the likelihood or probability of the input data belonging to a class) of input data provided to the input layer 208 of the feature component.

[0060] Although the learning model 206-a through the learning model 206-T are illustrated as including a single classification layer 212, the learning model 206-a through the learning model 206-T can include any numerical quantity of classification layers 212. Similarly, although the learning model 206-a through the learning model 206-T are illustrated as including a classification layer 212 that is separate from the output layer 214, the classification layer 212 and the output layer 214 can be a single layer. Further, the learning model 206-a through the learning model 206-T can include any numerical quantity of feature extraction layers 210. A numerical quantity of layers in the learning model 206-a through the learning model 206-T may be different for respective learning models and / or may be the same for respective learning models.

[0061] In some examples, a device 202-a and / or a device 202-b can train one or more of the learning model 206-a through the learning model 206-T to obtain or generate parameters of the hidden layers of the learning model 206-a through the learning model 206-T. Example parameters of the learning model 206-a through the learning model 206-T include, but are not limited to, weights, biases, and activation function parameters, among other parameters. Weight parameters represent a strength of connections between neurons in different layers of the learning model 206-a through the learning model 206-T. Biases are additional parameters added to neurons in the learning model 206-a through the learning model 206-T that provide for the learning model 206-a through the learning model 206-T to capture offset or bias in input data. Activation function parameters can include slope parameters or parameters defining a shape of an activation function in a parametric activation function.

[0062] In some variations, the device 202-b can train the learning model 206-a through the learning model 206-T, while the device 202-a and / or the device 202-b implement (e.g., execute) at least one of the learning model 206-a through the learning model 206-T. For example, the device 202-b can generate the parameters that define the learning model 206-a through the learning model 206-T and can transmit the parameters to the device 202-a, where the device 202-a implements at least one learning model of the learning model 206-a through the learning model 206-T that includes the parameters. Additionally, or alternatively, the device 202-a can train the learning model 206-a through the learning model 206-T, while the device 202-a and / or the device 202-b implement (e.g., execute) at least one of the learning model 206-a through the learning model 206-T.

[0063] The device 202-a and / or the device 202-b can develop (e.g., train) multiple learning models (e.g., the learning model 206-a through the learning model 206-T) for a defined task, where respective learning models have different parameters for the different layers. The device 202-a and / or the device 202-b can develop a learning model for different target domains for the defined task. The different target domains can represent a specific scenario, configuration, parameter of the network, physical propagation medium, or device behavior. That is, a target domain represents data for a scenario, configuration, parameter of a network, a physical propagation medium, and / or a device behavior. The device 202-a and / or the device 202-b can store parameters of the learning model 206-a through the learning model 206-T for transmission to one or more other devices (e.g., UEs or other devices). Additionally, or alternatively, the one or more other devices can receive and store the parameters of the learning model 206-a through the learning model 206-T. The other devices can use respective learning models for the different target domains. For example, a device can provide data samples from the target domain as input to at least one learning model. The learning model can provide an output that the device uses for a task.

[0064] In some cases, x and y denote an input data sample and an output data sample of the learning model 206-a through the learning model 206-T. As the learning model 206-a through the learning model 206-T are classifier learning models, an output, prediction, and / or inference y is discrete valued. As the learning model 206-a through the learning model 206-T are developed for a same task (e.g., for estimating transmitted messages, symbols, and / or bits, for spatial beam prediction, for temporal beam prediction, or for CSI prediction) the learning model 206-a through the learning model 206-T may have a same output space (e.g., y 6 {1, M], where M is a positiveinteger greater than one). In some variations, the kthclassification learning model, where k =1, T, generates a probability distribution over the predictions, conditioned on the input x, denoted by Pk(y|x). For respective kthclassification learning models, with k = 1, T, the prediction y is a discrete variable with y 6 {1, , M] and Pk(y|x) is the probability that y is the predicted label for x as per the model fQk. That is, Pk(yc|x() is the probability that ycis the label for the given input data sample x;, as per the prediction and / or inference made by the classification learning model fQk, with 0kas learning model parameters. In some examples, S£Li Pk(yc|Xj) = 1 or

[0065] In some examples, at 216, the device 202-b (e.g., and / or the device 202-a) can select a learning model of the learning model 206-a through the learning model 206-T based on a confidence score for the learning model satisfying a selection criterion. The selected learning model can result in a greatest accuracy and / or precision (e.g., inference and / or prediction performance) for one or more incoming data samples. The device 202-b can perform learning model selection without labeled data samples, reference signals, and / or control signals from another node and / or device in the wireless communications system 200 (e.g., or any other side and / or assistance information to aid the learning model selection). The device 202-b uses a set of unlabeled input data samples that are to be classified or, equivalently, the data samples for which labels are to be predicted and / or inferred through one of the learning model 206-a through the learning model 206-T to select a learning model. Thus, the learning model selection may be referred to as autonomous learning model selection. Autonomous learning model selection can additionally, or alternatively, be used in conjunction with labeled data, reference signals, or any other side and / or assistance information (e.g., the characteristics of the input samples) that may be relevant for learning model selection. Thus, the autonomous learning model selection can be used as a standalone method or a method to complement other methods for learning model selection.

[0066] In some variations, autonomous learning model selection may be triggered, initiated, and / or performed by a device (e.g., either a user device, including a UE or other user device, or an NE) hosting the learning model 206-a through the learning model 206-T or another device and / or entity in the wireless communications system 200.

[0067] The device 202 -b can obtain a set of data samples which are to be classified or inferred using a selected learning model from the learning model 206-a through the learning model 206-T. These data samples can be referred to as data samples from the target domain. A data distribution of incoming input data samples at the device 202 -b may change periodically or continuously across a time period, leading to the device 202 -b performing learning model selection often (e.g., greater than a threshold numerical quantity of times over a time period). The data samples from the target domain can be unlabeled samples. Thus, for selecting one of the T learning models (e.g., the learning model 206-a through the learning model 206-T), the device 202-b uses a set of input data samples xf= {xj, X2, , x( from the target domain (e.g., without the corresponding labels yi, i = 1, ... , n, for the input samples x , i = 1, ... , n).

[0068] In some examples, the device 202-b selects a learning model among the T learning models, fQk, for the input data samples x , i = 1, ... , n. A learning model fQk, k = 1, ... , T produces a conditional distribution Pk(y|x). The device 202-b can compute Pk(y|x^) for i = 1, ..., n and k = 1, ..., T. The device 202-b can compute a conditional entropy given the n input data samples xf, according to Equation 1 :where Pk(x) is a probability distribution of the input data samples for the learning model k. In some examples, the probability distribution of the input data samples may be available for the learning model or may be provided to the device 202-b (e.g., a node and / or entity performing the model selection). For example, the probability distribution may be based on a set of training data samples used to train the learning model. If the input data samples are equally likely, then the conditional entropy can be given as the average of the entropy of the kthlearning model predictions (y) given the n input data samples xf, according to Equation 2:H^Cylx')) = iSt, / / ( ’"(ylxf)In some examples, the logarithm (log) may be a base natural log, or a base-2 log. The conditional entropyanindication of the confidence in individualpredictions made by the learning model (e.g., lower entropy implies lower uncertainty in the predictions indicating higher confidence in the learning model).

[0069] The device 202 -b can compute an empirical marginal distribution of predicted labels according to Equation 3:In some examples, the probability distribution of the input data samples, Pk(x), for a learning model k may be available for the learning model or may be provided to the device 202-b (e.g., the node and / or entity performing the model selection). The probability distribution may be based on a set of training data samples used to train the learning model. If the input data samples are equally likely, then the distribution of the data samples can be calculated according to Equation 4:In some examples, the computed distribution Pk(y) is an approximation of Pk(y), which is a true marginal distribution of y. In some examples, instead of computing the marginal distribution of the predicted labels, the device 202-b can otherwise obtain the marginal probability distribution for the learning model, such as by receiving an indication of the marginal probability distribution. For example, the probability distribution of predicted labels may be based on a set of training data samples and training labels used to train the model.

[0070] The device 202-b can compute the entropy of the predicted labels according to Equation 5:In some variations, a higher value of H(Pk(y)) leads to balance across label prediction or indicates that the marginal distribution of the predicted labels is close to a unform distribution (e.g., which is a desired quality with a reasonable number of data samples x-, i = 1, ... , n).

[0071] The device 202-b can assign confidence scores to the respective learning models (e.g., the learning model 206-a through the learning model 206-T). For example, the device 202-b can set a confidence score of a kthmodel, denoted by Gk, according to Equation 6 and Equation 7:In some examples, w1and w2, with 0 < w1(w2< 1, are weights (e.g., an importance) assigned to respective terms in the confidence score. The device 202-b can select and / or determine values for the weights based on use cases for the learning models and / or other available side-information at the device carrying out the model selection procedure. Additionally, or alternatively, the device 202-b can receive an indication (e.g., from anther device or node in the wireless communications system 200) that indicates the weights. If vvx= w2= 1, Gk= / kdenotes the information gain which is a measure of how much the entropy or uncertainty of output of the kthmodel y is reduced when xf, the input of the learning model, is known. That is, the information gain is a measure of how much information about y the device 202-b obtains by knowing xf, Gk= / k(xf; y), where / k(xt; y) denotes the mutual information between xfand y, input and output of the kthlearning model. In some cases, the device 202-b can fix w1(w2based on user provided input that is based on the use case of the learning model, known physics related to the wireless communications system 200, among other factors.

[0072] In some cases, such as for spatial beam prediction, the learning model classifies respective input data samples (e.g., a set of beam measurements) into one of M candidate beams, such that the inferred and / or predicted beam is the beam with a highest performance (e.g., signal quality and / or signal power, among other metrics). A beam with the highest performance for a device (e.g., a UE) that is within a defined geographic location may be part of a subset of a set of available beams. Thus, the learning model output is a value from a subset of the available beam indices, resulting in a lower value of ) . In some other examples, such as for alearning-based receiver, the learning model classifies input data sample received from a wireless channel output into one of a set of possible transmitted messages, symbols, and / or bits. In some variations, the information bits have a uniform distribution and the output of the learning model can include respective different possible symbols (e.g., all of the different possible symbols), even with a relatively small number of data samples 'Xti, i = 1, Thus, a value ocanbe relatively high (e.g., greater than a threshold value). The device 202-b can assign values to theweights, such that w2< w1;when a number of data samples available from the target domain are relatively small in number (e.g., when n is less than a threshold value). When the n is a relatively large value (e.g., greater than a threshold value), then the device 202 -b can assign values to the weights, such that

[0073] The device 202 -b can select a learning model with a highest confidence score (e.g., by comparing confidence scores for the learning model 206-a through the learning model 206-T). For example, the device 202 -b can store T number of learning models fQk, k = 1, ... , T. The device 202- b can obtain unlabeled data samples xf= {xj, x2, x^} from a target domain and values of weights w1(w2. In some examples, the device 202 -b can receive signaling that indicates the data samples 218 from another device (e.g., the device 202-a). For example, the device 202-a can perform one or more measurements (e.g., beam measurements, reference signal measurements, and / or channel measurements, among other measurements) and can transmit the measurements to the device 202 -b to use as the data samples. Additionally, or alternatively, the device 202-b can perform the measurements to obtain the data samples. Example beam measurements can include, but are not limited to, received signal strength and / or quality from a set of beams. Example reference signal measurements and / or channel measurements can include, but are not limited to, channel gain data, phase shift data, frequency response data, delay spread data, and Doppler shift data, among other information.

[0074] The device 202-b obtains an index of a learning model with a highest performance (e.g., accuracy and / or precision), k*, for the data xf= {xj, x2, x^}. For example, for k = 1 -» T, the device 202-b uses xfand fQk to determine Pk(y|x^) for i = 1, n. The device 202-b computes the conditional entropy—The device 202-b computes Gk, the confidence score of kthmodel Gk= The device 202-b selects alearning model that satisfies a selection criterion. For example, the selection criterion can include a learning model with a highest confidence score (e.g., k* = ^max Gk). Additionally, or alternatively, the selection criterion can include a difference between the conditional entropy and the marginal entropy for a learning model being a greatest difference for the set of learning models.Additionally, or alternatively, the selection criterion can include a conditional entropy for a learning model being a lowest (e.g., minimum) for the set of learning models.

[0075] In some examples, the data distribution of input data samples may change relatively quickly (e.g., in less than a threshold duration, on a short scale of time). For example, the distribution of data samples can change after a numerical quantity of data samples, such as every few samples or every couple of samples. The device 202 -b can select values for the weights (e.g., w±and w2) according to a numerical quantity of data samples for a target distribution. For example, the device 202 -b can select values of the weights, such that w2< w1when a numerical quantity of data samples available from the target domain is relatively small (e.g., when n is less than a threshold value). If n = 1, then the device 202-b can select values, such that w2== 1. Additionally, or alternatively, if T > 1 learning models are available, then the device 202-b can compute a confidence score Gkfor respective learning models for incoming data samples that are to be classified. The device 202-b can select the output of the learning model that has a highest confidence score. Thus, the device 202-b can use multiple learning models concurrently, such that the device uses a learning model with a highest confidence score for each incoming data sample that is to be classified (e.g., at every instant of time).

[0076] In some examples, the device 202-b can implement autonomous learning model selection for information prediction for a signal, such as CQI prediction. For example, for CQI prediction, the input of the learning model includes channel vectors over a subset of a set of time and / or frequency resources. The output of the learning model includes CQI for an entire set of time and / or frequency resources. In some cases, the learning model input includes channel estimates for a subset of the set of subbands of a bandwidth part (BWP) for a channel. The learning model output includes subband CQI values for both the subset of the set of subbands and the remainder of subbands in the set of subbands. In some other cases, the learning model input includes channel estimates for a first set of slots (e.g., time resources) prior to a CSI reference resource for a CSI reporting setting. The learning model output includes CQI values for a second set of slots after the CSI reference resource. The device 202-b can be configured with a CSI resource setting, where one or more CSLRSs with a CQI prediction type are transmitted over a subset of resource blocks of BWPs. The subset of the resource blocks of the BWPs are either selected by the device 202-b, configured by the network, or set by a rule. A selection of the subsets of the resource blocks maydepend on the selected learning model. A UE can implement the learning model. In yet other cases, the input of the learning model includes CQI values over a subset of a set of time and / or frequency resources. The output of the learning model includes CQI for a remainder of the time and / or frequency resources. In yet other cases, the learning model input includes subband CQI values for a subset of the set of subbands of a BWP of a channel. The learning model output includes subband CQI values for a remainder of subbands in the set of subbands. In yet other cases, the learning model input includes CQI values for a first set of slots prior to a CSI reference resource for a CSI reporting setting. The learning model output includes CQI values for a second set of slots after the CSI reference resource. The device 202 -b can be configured with a CSI reporting setting, where the CSI report setting defines reporting of a subset of the subband CQI values of the set of subband that includes the BWP. The subset of the set of the subbands may be network configured and can depend on the selected learning model. A NE can implement the learning model.

[0077] In some examples, the device 202-b can transmit an indication of the selected learning model 220 to the device 202-a. For example, the device 202-b can transmit control signaling via the communications link 204-a that indicates the indication of the selected learning model 220. In some variations, the indication of the selected learning model 220 can include an index of the selected learning model and / or can include one or more parameters that define the selected learning model.

[0078] Figure 3 illustrates an example of a signaling diagram 300 in accordance with aspects of the present disclosure. In some examples, the signaling diagram 300 may implement aspects of the wireless communications system 100 and the wireless communications system 200. The signaling diagram 300 may illustrate an example of a device 302-a and / or a device 302 -b performing autonomous model selection. In some cases, the device 302-a may be an example of a device 202-a and the device 302-b may be an example of a device 202-b, as described with reference to Figure 2. For example, the device 302-a and the device 302-b can be examples of an NE and / or a UE. Alternative examples of the following may be implemented, where some processes are performed in a different order than described or are not performed. In some cases, processes may include additional features not mentioned below, or further processes may be added.

[0079] In some examples, at 304, the device 302-a can transmit signaling that triggers selection of a learning model to the device 302-b. For example, the device 302-a can transmit controlsignaling that includes a field (e.g., a one-bit indicator) that triggers the selection of the learning model to the device 302-b.

[0080] In some cases, at 306, the device 302-a can transmit signaling that indicates one or more data samples to the device 302-b. For example, the device 302-a can perform one or more measurements to obtain the data samples. The device 302-a can transmit control signaling, and / or a data transmission that includes the data samples. Example measurements can include, but are not limited to, received signal strength of one or more beams, quality measurements from a set of beams, channel gain data, phase shift data, frequency response data, delay spread data, and Doppler shift data, among other information. In some examples, the signaling that indicates the data samples can also trigger selection of the learning model (e.g., the signaling that includes the data samples implicitly indicates for the device 302-b to trigger the selection of the learning model and / or includes a field that explicitly indicates for the device 302-b to trigger the selection of the learning model).

[0081] At 308, the device 302-b can obtain one or more data samples. The data samples can include one or more data samples (e.g., unlabeled data samples) for input to a learning model. For example, the data samples are classified by the learning model into one or more labels and / or classes. The device 302-b can obtain the data samples via the signaling at 304. Additionally, or alternatively, the device 302-b can perform the measurements to obtain the data samples.

[0082] In some cases, the device 302-b can receive one or more reference signals from another device (e.g., from the device 302-a). The device 302-b can obtain information by performing measurements on the reference signals. The information can include one or more of an RSRP measurement, an RSRQ measurement, a CQI, a PMI, an MCS index, or a RI.

[0083] At 310, the device 302-b can generate confidence scores for a set of learning models. The device can generate the confidence scores by providing the set of data samples as input to the set of learning models. For example, the confidence scores can be a function of conditional entropy and marginal entropy for the learning models. In some cases, the device 302-b generates respective measures of uncertainty for outputs from the learning models based on the set of data samples (e.g., conditional entropy and / or marginal entropy). For example, the respective measures of uncertainty include a marginal entropy of the outputs, or a conditional entropy of the outputs given the input(e.g., the set of data samples). In some examples, the device 302-b can apply one or more weights (e.g., with values between zero and one) to the conditional entropy and the marginal entropy. In some examples, the confidence scores include an information gain for the outputs from the learning models given the input (e.g., the set of data samples). The information gain is a measure of how much information about the output the device 302-b obtains by knowing the input.

[0084] At 312, the device 302-b can select a learning model based on a confidence score for the learning model satisfying a selection criterion. The device 302-b selects the learning model to execute at the device 302-b and / or at the device 302-a. The selection criterion can include a minimum conditional entropy (e.g., measure of uncertainty) of the respective conditional entropies for the different learning models. Additionally, or alternatively, the selection criterion includes a maximum difference between the conditional entropy and the marginal entropy (e.g., or a weighted conditional entropy and weighted marginal entropy). Additionally, or alternatively, the selection criterion includes the highest confidence score of the confidence scores calculated (e.g., generated, obtained) for different learning models.

[0085] In some examples, at 314, the device 302-b can transmit an indication of the selected learning model to the device 302-a. The indication of the selected learning model can include an indication of an index of the learning model and / or can include one or more parameters that define the learning model. The device 302-a can include the indication of the learning model in control signaling and / or in any other transmission to the device 302-a (e.g., signaling dedicated for learning model indications and / or one or more new fields in signaling that is not dedicated for learning model indications).

[0086] In some variations, at 316, the device 302-b can execute (process, implement, deploy, etc.) the learning model. For example, the device 302-b can generate a prediction as the output from the learning model. Additionally, or alternatively, at 318, the device 302-a can execute the learning model. For example, the device 302-a can generate a prediction as the output from the learning model.

[0087] For a learning model implemented as a receiver, the device 302-b and / or the device 302-a can input signals output from a channel (e.g., a wireless channel) to the selected learning model, and the selected learning model can output a prediction of at least one of a message, a set ofsymbols, or a set of bits. For a learning model implemented for beam prediction, the device 302-b and / or the device 302-a can input a set of beam measurements and / or assistance information for one or more beams to the learning model, and the learning model can output a prediction of at least one beam to use for transmitting or receiving signaling (e.g., a beam with a highest signal quality and / or power, among other performance metrics). Additionally, or alternatively, the device 302-b and / or the device 302-a can input a set of beam measurements or assistance information for one or more beams that is current and historical to the learning model, and the learning model can output a prediction of at least one beam to use for transmitting or receiving signaling in a set of future time slots. The beam measurements can include, but are not limited to, RSRP measurements, RSRQ measurements, and / or SINR measurements. The assistance information can include, but is not limited to, information about a channel (e.g., CSI), a quality of service (QoS), and information about a communication environment, among other information. For a learning model implemented to predict information about a signal and / or channel (e.g., CSI, including a CQI, a MPI, a RI, and / or a MCS index), the device 302-b and / or the device 302-a can input a set of reference signal measurements or a set of channel vectors to the learning model, and the learning model can output a prediction of the information about the signal and / or channel. The set of channel vectors define an initial subset of time-frequency resources of a set of time-frequency resources, and the output from the learning model defines another subset of time-frequency resources of the set of time-frequency resources and / or the entire set of time-frequency resources.

[0088] Figure 4 illustrates an example of a UE 400 in accordance with aspects of the present disclosure. The UE 400 may include a processor 402, a memory 404, a controller 406, and a transceiver 408. The processor 402, the memory 404, the controller 406, or the transceiver 408, or various combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0089] The processor 402, the memory 404, the controller 406, or the transceiver 408, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereofconfigured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0090] The processor 402 may include an intelligent hardware device (e.g., a general-purpose processor, a DSP, a CPU, an ASIC, a field-programmable gate array (FPGA), or any combination thereof). In some implementations, the processor 402 may be configured to operate the memory 404. In some other implementations, the memory 404 may be integrated into the processor 402. The processor 402 may be configured to execute computer-readable instructions stored in the memory 404 to cause the UE 400 to perform various functions of the present disclosure.

[0091] The memory 404 may include volatile or non-volatile memory. The memory 404 may store computer-readable, computer-executable code including instructions when executed by the processor 402 cause the UE 400 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such as the memory 404 or another type of memory. Computer-readable media includes both non-transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0092] In some implementations, the processor 402 and the memory 404 coupled with the processor 402 may be configured to or operable to cause the UE 400 to perform one or more of the functions described herein (e.g., executing, by the processor 402, instructions stored in the memory 404). For example, the processor 402 may support wireless communication at the UE 400 in accordance with examples as disclosed herein. The UE 400 may be configured to or operable to support a means for obtaining a set of data samples to provide as input to a set of learning models stored at the first device, generating, based on providing the set of data samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models, and selecting, for executing at the first device or at a second device, a learning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion.

[0093] Additionally, the UE 400 may be configured to or operable to support any one or combination of transmitting, to the second device, a message that indicates the selected learningmodel. Additionally, or alternatively, obtaining the set of data samples includes receiving, from the second device, a message that indicates the set of data samples. Additionally, or alternatively, obtaining the set of data samples includes receiving, from the second device, a set of reference signals, and obtaining information associated with the set of reference signals, where the information includes one or more of an RSRP measurement, an RSRQ measurement, a CQI, a PMI, an MCS index, or an RI. Additionally, or alternatively, the UE 400 may be configured to or operable to support includes receiving, from the second device, a message that triggers selection of the learning model. Additionally, or alternatively, generating the respective confidence scores includes generating respective measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a minimum measure of uncertainty of the respective measures of uncertainty. Additionally, or alternatively, the respective measures of uncertainty include at least one of a marginal entropy of the set of outputs or a conditional entropy of the set of outputs based on the set of data samples.

[0094] Additionally, or alternatively, generating the respective confidence scores includes generating respective first measures of uncertainty associated with the set of outputs from the set of learning models and respective second measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a maximum difference between the respective first measures of uncertainty and the respective second measures of uncertainty. Additionally, or alternatively, generating the respective confidence scores includes generating respective first weighted measures of uncertainty associated with the set of outputs from the set of learning models and respective second weighted measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the respective first weighted measures of uncertainty and the respective second weighted measures of uncertainty are weighted based on a value between zero and one. Additionally, or alternatively, generating the respective confidence scores includes generating an information gain associated with the set of outputs from the set of learning models based on the set of data samples.

[0095] Additionally, or alternatively, executing the learning model includes generating, as the output from the learning model, a prediction of at least one of a message, a set of symbols, or a set of bits based on providing the set of data samples as input to the selected learning model, where theset of data samples include one or more channel output signals, generating, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, generating, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling in a set of future time slots based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, or generating, as the output from the learning model, a prediction of CSI based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of reference signal measurements or a set of channel vectors, and where the CSI includes at least one of a CQI associated with a set of time-frequency resources, a PMI, an MCS index, or an RI. Additionally, or alternatively, the set of channel vectors are associated with a first subset of time-frequency resources of a set of time-frequency resources, and where the output from the learning model is associated with a second subset of time-frequency resources of the set of time-frequency resources or the set of time-frequency resources. Additionally, or alternatively, the first device is an NE. Additionally, or alternatively, the first device is a UE.

[0096] Additionally, or alternatively, the UE 400 may support at least one memory (e.g., the memory 404) and at least one processor (e.g., the processor 402) coupled with the at least one memory and configured to or operable to cause the UE to obtain a set of data samples to provide as input to a set of learning models stored at the first device, generate, based on providing the set of data samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models, and select, for executing at the first device or at a second device, a learning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion.

[0097] Additionally, the UE 400 may be configured to or operable to support any one or combination of the at least one processor is configured to or operable to transmit, to the second device, a message that indicates the selected learning model. Additionally, or alternatively, to obtain the set of data samples, the at least one processor is configured to or operable to receive, from thesecond device, a message that indicates the set of data samples. Additionally, or alternatively, to obtain the set of data samples, the at least one processor is configured to or operable to receive, from the second device, a set of reference signals, and obtain information associated with the set of reference signals, where the information includes one or more of an RSRP measurement, an RSRQ measurement, a CQI, a PMI, an MCS index, or an RI. Additionally, or alternatively, the at least one processor is configured to or operable to receive, from the second device, a message that triggers selection of the learning model. Additionally, or alternatively, to generate the respective confidence scores, the at least one processor is configured to or operable to generate respective measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a minimum measure of uncertainty of the respective measures of uncertainty. Additionally, or alternatively, the respective measures of uncertainty include at least one of a marginal entropy of the set of outputs or a conditional entropy of the set of outputs based on the set of data samples.

[0098] Additionally, or alternatively, to generate the respective confidence scores, the at least one processor is configured to or operable to generate respective first measures of uncertainty associated with the set of outputs from the set of learning models and respective second measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a maximum difference between the respective first measures of uncertainty and the respective second measures of uncertainty. Additionally, or alternatively, to generate the respective confidence scores, the at least one processor is configured to or operable to generate respective first weighted measures of uncertainty associated with the set of outputs from the set of learning models and respective second weighted measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the respective first weighted measures of uncertainty and the respective second weighted measures of uncertainty are weighted based on a value between zero and one. Additionally, or alternatively, to generate the respective confidence scores, the at least one processor is configured to or operable to generate an information gain associated with the set of outputs from the set of learning models based on the set of data samples.

[0099] Additionally, or alternatively, to execute the learning model, the at least one processor is configured to or operable to generate, as the output from the learning model, a prediction of at leastone of a message, a set of symbols, or a set of bits based on providing the set of data samples as input to the selected learning model, where the set of data samples include one or more channel output signals, generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling in a set of future time slots based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, or generate, as the output from the learning model, a prediction of CSI based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of reference signal measurements or a set of channel vectors, and where the CSI includes at least one of a CQI associated with a set of time-frequency resources, a PMI, an MCS index, or an RI. Additionally, or alternatively, the set of channel vectors are associated with a first subset of timefrequency resources of a set of time-frequency resources, and where the output from the learning model is associated with a second subset of time-frequency resources of the set of time-frequency resources or the set of time-frequency resources. Additionally, or alternatively, the first device is an NE. Additionally, or alternatively, the first device is a UE.

[0100] The controller 406 may manage input and output signals for the UE 400. The controller 406 may also manage peripherals not integrated into the UE 400. In some implementations, the controller 406 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or other operating systems. In some implementations, the controller 406 may be implemented as part of the processor 402.

[0101] In some implementations, the UE 400 may include at least one transceiver 408. In some other implementations, the UE 400 may have more than one transceiver 408. The transceiver 408 may represent a wireless transceiver. The transceiver 408 may include one or more receiver chains 410, one or more transmitter chains 412, or a combination thereof.

[0102] A receiver chain 410 may be configured to or operable to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 410 mayinclude one or more antennas to receive a signal over the air or wireless medium. The receiver chain 410 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured to or operable to amplify the received signal. The receiver chain 410 may include at least one demodulator configured to or operable to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 410 may include at least one decoder for decoding the demodulated signal to receive the transmitted data.

[0103] A transmitter chain 412 may be configured to or operable to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 412 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. The at least one modulator may be configured to or operable to support one or more techniques such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase-shift keying (PSK) or quadrature amplitude modulation (QAM).The transmitter chain 412 may also include at least one power amplifier configured to or operable to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 412 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0104] Figure 5 illustrates an example of a processor 500 in accordance with aspects of the present disclosure. The processor 500 may be an example of a processor configured to perform various operations in accordance with examples as described herein. The processor 500 may include a controller 502 configured to perform various operations in accordance with examples as described herein. The processor 500 may optionally include at least one memory 504, which may be, for example, an L1 / L2 / L3 cache. Additionally, or alternatively, the processor 500 may optionally include one or more arithmetic-logic units (ALUs) 506. One or more of these components may be in electronic communication or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces (e.g., buses).

[0105] The processor 500 may be a processor chipset and include a protocol stack (e.g., a software stack) executed by the processor chipset to perform various operations (e.g., receiving, obtaining, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, reading) in accordance with examples as described herein. The processor chipsetmay include one or more cores, one or more caches (e.g., memory local to or included in the processor chipset (e.g., the processor 500) or other memory (e.g., random access memory (RAM), read-only memory (ROM), dynamic RAM (DRAM), synchronous dynamic RAM (SDRAM), static RAM (SRAM), ferroelectric RAM (FeRAM), magnetic RAM (MRAM), resistive RAM (RRAM), flash memory, phase change memory (PCM), and others).

[0106] The controller 502 may be configured to manage and coordinate various operations (e.g., signaling, receiving, obtaining, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, reading) of the processor 500 to cause the processor 500 to support various operations in accordance with examples as described herein. For example, the controller 502 may operate as a control unit of the processor 500, generating control signals that manage the operation of various components of the processor 500. These control signals include enabling or disabling functional units, selecting data paths, initiating memory access, and coordinating timing of operations.

[0107] The controller 502 may be configured to fetch (e.g., obtain, retrieve, receive) instructions from the memory 504 and determine subsequent instruction(s) to be executed to cause the processor 500 to support various operations in accordance with examples as described herein. The controller 502 may be configured to track memory addresses of instructions associated with the memory 504. The controller 502 may be configured to decode instructions to determine the operation to be performed and the operands involved. For example, the controller 502 may be configured to interpret the instruction and determine control signals to be output to other components of the processor 500 to cause the processor 500 to support various operations in accordance with examples as described herein. Additionally, or alternatively, the controller 502 may be configured to manage flow of data within the processor 500. The controller 502 may be configured to control transfer of data between registers, ALUs 506, and other functional units of the processor 500.

[0108] The memory 504 may include one or more caches (e.g., memory local to or included in the processor 500 or other memory, such as RAM, ROM, DRAM, SDRAM, SRAM, MRAM, flash memory, etc. In some implementations, the memory 504 may reside within or on a processor chipset (e.g., local to the processor 500). In some other implementations, the memory 504 may reside external to the processor chipset (e.g., remote to the processor 500).

[0109] The memory 504 may store computer-readable, computer-executable code including instructions that, when executed by the processor 500, cause the processor 500 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such as system memory or another type of memory. The controller 502 and / or the processor 500 may be configured to execute computer-readable instructions stored in the memory 504 to cause the processor 500 to perform various functions. For example, the processor 500 and / or the controller 502 may be coupled with or to the memory 504, the processor 500, and the controller 502, and may be configured to perform various functions described herein. In some examples, the processor 500 may include multiple processors and the memory 504 may include multiple memories. One or more of the multiple processors may be coupled with one or more of the multiple memories, which may, individually or collectively, be configured to perform various functions herein.

[0110] The one or more ALUs 506 may be configured to support various operations in accordance with examples as described herein. In some implementations, the one or more ALUs 506 may reside within or on a processor chipset (e.g., the processor 500). In some other implementations, the one or more ALUs 506 may reside external to the processor chipset (e.g., the processor 500). One or more ALUs 506 may perform one or more computations such as addition, subtraction, multiplication, and division on data. For example, one or more ALUs 506 may receive input operands and an operation code, which determines an operation to be executed. One or more ALUs 506 may be configured with a variety of logical and arithmetic circuits, including adders, subtractors, shifters, and logic gates, to process and manipulate the data according to the operation. Additionally, or alternatively, the one or more ALUs 506 may support logical operations such as AND, OR, exclusive-OR (XOR), not-OR (NOR), and not-AND (NAND), enabling the one or more ALUs 506 to manage conditional operations, comparisons, and bitwise operations.

[0111] The processor 500 may support wireless communication in accordance with examples as disclosed herein. The processor 500 may be configured to or operable to support at least one controller (e.g., the controller 502) coupled with at least one memory (e.g., the memory 504) and configured to or operable to cause the processor to obtain a set of data samples to provide as input to a set of learning models stored at the processor, generate, based on providing the set of data samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models, and select, for executing at the processor or at a device, alearning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion.

[0112] Additionally, the processor 500 may be configured to or operable to support any one or combination of the at least one controller is configured to or operable to cause the processor to transmit, to the device, a message that indicates the selected learning model. Additionally, or alternatively, to obtain the set of data samples, the at least one controller is configured to or operable to cause the processor to receive, from the device, a message that indicates the set of data samples. Additionally, or alternatively, to obtain the set of data samples, the at least one controller is configured to or operable to cause the processor to receive, from the device, a set of reference signals, and obtain information associated with the set of reference signals, where the information includes one or more of an RSRP measurement, an RSRQ measurement, a CQI, a PMI, an MCS index, or an RI. Additionally, or alternatively, the at least one controller is configured to or operable to cause the processor to receive, from the device, a message that triggers selection of the learning model. Additionally, or alternatively, to generate the respective confidence scores, the at least one controller is configured to or operable to cause the processor to generate respective measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a minimum measure of uncertainty of the respective measures of uncertainty. Additionally, or alternatively, the respective measures of uncertainty include at least one of a marginal entropy of the set of outputs or a conditional entropy of the set of outputs based on the set of data samples.

[0113] Additionally, or alternatively, to generate the respective confidence scores, the at least one controller is configured to or operable to cause the processor to generate respective first measures of uncertainty associated with the set of outputs from the set of learning models and respective second measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a maximum difference between the respective first measures of uncertainty and the respective second measures of uncertainty. Additionally, or alternatively, to generate the respective confidence scores, the at least one controller is configured to or operable to cause the processor to generate respective first weighted measures of uncertainty associated with the set of outputs from the set of learning models and respective second weighted measures of uncertainty associated with the set of outputs from theset of learning models based on the set of data samples, where the respective first weighted measures of uncertainty and the respective second weighted measures of uncertainty are weighted based on a value between zero and one. Additionally, or alternatively, to generate the respective confidence scores, the at least one controller is configured to or operable to cause the processor to generate an information gain associated with the set of outputs from the set of learning models based on the set of data samples.

[0114] Additionally, or alternatively, to execute the learning model, the at least one controller is configured to or operable to cause the processor to generate, as the output from the learning model, a prediction of at least one of a message, a set of symbols, or a set of bits based on providing the set of data samples as input to the selected learning model, where the set of data samples include one or more channel output signals, generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling in a set of future time slots based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, or generate, as the output from the learning model, a prediction of CSI based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of reference signal measurements or a set of channel vectors, and where the CSI includes at least one of a CQI associated with a set of time-frequency resources, a PMI, an MCS index, or an RI. Additionally, or alternatively, the set of channel vectors are associated with a first subset of timefrequency resources of a set of time-frequency resources, and where the output from the learning model is associated with a second subset of time-frequency resources of the set of time-frequency resources or the set of time-frequency resources. Additionally, or alternatively, the processor is associated with an NE. Additionally, or alternatively, the processor is associated with a UE.

[0115] Figure 6 illustrates an example of an NE 600 in accordance with aspects of the present disclosure. The NE 600 may include a processor 602, a memory 604, a controller 606, and a transceiver 608. The processor 602, the memory 604, the controller 606, or the transceiver 608, orvarious combinations thereof or various components thereof may be examples of means for performing various aspects of the present disclosure as described herein. These components may be coupled (e.g., operatively, communicatively, functionally, electronically, electrically) via one or more interfaces.

[0116] The processor 602, the memory 604, the controller 606, or the transceiver 608, or various combinations or components thereof may be implemented in hardware (e.g., circuitry). The hardware may include a processor, a DSP, an ASIC, or other programmable logic device, or any combination thereof configured as or otherwise supporting a means for performing the functions described in the present disclosure.

[0117] The processor 602 may include an intelligent hardware device (e.g., a general-purpose processor, a DSP, a CPU, an ASIC, an FPGA, or any combination thereof). In some implementations, the processor 602 may be configured to operate the memory 604. In some other implementations, the memory 604 may be integrated into the processor 602. The processor 602 may be configured to execute computer-readable instructions stored in the memory 604 to cause the NE 600 to perform various functions of the present disclosure.

[0118] The memory 604 may include volatile or non-volatile memory. The memory 604 may store computer-readable, computer-executable code including instructions when executed by the processor 602 cause the NE 600 to perform various functions described herein. The code may be stored in a non-transitory computer-readable medium such as the memory 604 or another type of memory. Computer-readable media includes both non-transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that may be accessed by a general-purpose or special-purpose computer.

[0119] In some implementations, the processor 602 and the memory 604 coupled with the processor 602 may be configured to or operable to cause the NE 600 to perform one or more of the functions described herein (e.g., executing, by the processor 602, instructions stored in the memory 604). For example, the processor 602 may support wireless communication at the NE 600 in accordance with examples as disclosed herein. The NE 600 may be configured to or operable to support a means for obtaining a set of data samples to provide as input to a set of learning modelsstored at the first device, generating, based on providing the set of data samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models, and selecting, for executing at the first device or at a second device, a learning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion.

[0120] Additionally, the NE 600 may be configured to or operable to support any one or combination of the method further comprising transmitting, to the second device, a message that indicates the selected learning model. Additionally, or alternatively, obtaining the set of data samples includes receiving, from the second device, a message that indicates the set of data samples. Additionally, or alternatively, obtaining the set of data samples includes receiving, from the second device, a set of reference signals, and obtaining information associated with the set of reference signals, where the information includes one or more of an RSRP measurement, an RSRQ measurement, a CQI, a PMI, an MCS index, or an RI. Additionally, or alternatively, the NE 600 may be configured to or operable to support receiving, from the second device, a message that triggers selection of the learning model. Additionally, or alternatively, generating the respective confidence scores includes generating respective measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a minimum measure of uncertainty of the respective measures of uncertainty. Additionally, or alternatively, the respective measures of uncertainty include at least one of a marginal entropy of the set of outputs or a conditional entropy of the set of outputs based on the set of data samples.

[0121] Additionally, or alternatively, generating the respective confidence scores includes generating respective first measures of uncertainty associated with the set of outputs from the set of learning models and respective second measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a maximum difference between the respective first measures of uncertainty and the respective second measures of uncertainty. Additionally, or alternatively, generating the respective confidence scores includes generating respective first weighted measures of uncertainty associated with the set of outputs from the set of learning models and respective second weighted measures of uncertainty associated with the set of outputs from the set of learning models based on the set ofdata samples, where the respective first weighted measures of uncertainty and the respective second weighted measures of uncertainty are weighted based on a value between zero and one. Additionally, or alternatively, generating the respective confidence scores includes generating an information gain associated with the set of outputs from the set of learning models based on the set of data samples.

[0122] Additionally, or alternatively, executing the learning model includes generating, as the output from the learning model, a prediction of at least one of a message, a set of symbols, or a set of bits based on providing the set of data samples as input to the selected learning model, where the set of data samples include one or more channel output signals, generating, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, generating, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling in a set of future time slots based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, or generating, as the output from the learning model, a prediction of CSI based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of reference signal measurements or a set of channel vectors, and where the CSI includes at least one of a CQI associated with a set of time-frequency resources, a PMI, an MCS index, or an RI. Additionally, or alternatively, the set of channel vectors are associated with a first subset of time-frequency resources of a set of time-frequency resources, and where the output from the learning model is associated with a second subset of time-frequency resources of the set of time-frequency resources or the set of time-frequency resources.Additionally, or alternatively, the first device is an NE. Additionally, or alternatively, the first device is a UE.

[0123] Additionally, or alternatively, the NE 600 may support at least one memory (e.g., the memory 604) and at least one processor (e.g., the processor 602) coupled with the at least one memory and configured to or operable to cause the NE to obtain a set of data samples to provide as input to a set of learning models stored at the first device, generate, based on providing the set ofdata samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models, and select, for executing at the first device or at a second device, a learning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion.

[0124] Additionally, the NE 600 may be configured to support any one or combination of the at least one processor is configured to or operable to cause the NE to transmit, to the second device, a message that indicates the selected learning model. Additionally, or alternatively, to obtain the set of data samples, the at least one processor is configured to or operable to receive, from the second device, a message that indicates the set of data samples. Additionally, or alternatively, to obtain the set of data samples, the at least one processor is configured to or operable to receive, from the second device, a set of reference signals, and obtain information associated with the set of reference signals, where the information includes one or more of an RSRP measurement, an RSRQ measurement, a CQI, a PMI, an MCS index, or an RI. Additionally, or alternatively, the at least one processor is configured to or operable to receive, from the second device, a message that triggers selection of the learning model. Additionally, or alternatively, to generate the respective confidence scores, the at least one processor is configured to or operable to generate respective measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a minimum measure of uncertainty of the respective measures of uncertainty. Additionally, or alternatively, the respective measures of uncertainty include at least one of a marginal entropy of the set of outputs or a conditional entropy of the set of outputs based on the set of data samples.

[0125] Additionally, or alternatively, to generate the respective confidence scores, the at least one processor is configured to or operable to generate respective first measures of uncertainty associated with the set of outputs from the set of learning models and respective second measures of uncertainty associated with the set of outputs from the set of learning models based on the set of data samples, where the selection criterion includes a maximum difference between the respective first measures of uncertainty and the respective second measures of uncertainty. Additionally, or alternatively, to generate the respective confidence scores, the at least one processor is configured to or operable to generate respective first weighted measures of uncertainty associated with the set of outputs from the set of learning models and respective second weighted measures of uncertaintyassociated with the set of outputs from the set of learning models based on the set of data samples, where the respective first weighted measures of uncertainty and the respective second weighted measures of uncertainty are weighted based on a value between zero and one. Additionally, or alternatively, to generate the respective confidence scores, the at least one processor is configured to or operable to generate an information gain associated with the set of outputs from the set of learning models based on the set of data samples.

[0126] Additionally, or alternatively, to execute the learning model, the at least one processor is configured to or operable to generate, as the output from the learning model, a prediction of at least one of a message, a set of symbols, or a set of bits based on providing the set of data samples as input to the selected learning model, where the set of data samples include one or more channel output signals, generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling in a set of future time slots based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of beam measurements or assistance information associated with one or more beams, or generate, as the output from the learning model, a prediction of CSI based on providing the set of data samples as input to the selected learning model, where the set of data samples include at least one of a set of reference signal measurements or a set of channel vectors, and where the CSI includes at least one of a CQI associated with a set of time-frequency resources, a PMI, an MCS index, or an RI. Additionally, or alternatively, the set of channel vectors are associated with a first subset of timefrequency resources of a set of time-frequency resources, and where the output from the learning model is associated with a second subset of time-frequency resources of the set of time-frequency resources or the set of time-frequency resources. Additionally, or alternatively, the first device is an NE. Additionally, or alternatively, the first device is a UE.

[0127] The controller 606 may manage input and output signals for the NE 600. The controller 606 may also manage peripherals not integrated into the NE 600. In some implementations, the controller 606 may utilize an operating system such as iOS®, ANDROID®, WINDOWS®, or otheroperating systems. In some implementations, the controller 606 may be implemented as part of the processor 602.

[0128] In some implementations, the NE 600 may include at least one transceiver 608. In some other implementations, the NE 600 may have more than one transceiver 608. The transceiver 608 may represent a wireless transceiver. The transceiver 608 may include one or more receiver chains 610, one or more transmitter chains 612, or a combination thereof.

[0129] A receiver chain 610 may be configured to receive signals (e.g., control information, data, packets) over a wireless medium. For example, the receiver chain 610 may include one or more antennas to receive a signal over the air or wireless medium. The receiver chain 610 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured to amplify the received signal. The receiver chain 610 may include at least one demodulator configured to demodulate the receive signal and obtain the transmitted data by reversing the modulation technique applied during transmission of the signal. The receiver chain 610 may include at least one decoder for decoding the demodulated signal to receive the transmitted data.

[0130] A transmitter chain 612 may be configured to generate and transmit signals (e.g., control information, data, packets). The transmitter chain 612 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. The at least one modulator may be configured to support one or more techniques such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase-shift keying (PSK) or quadrature amplitude modulation (QAM). The transmitter chain 612 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over the wireless medium. The transmitter chain 612 may also include one or more antennas for transmitting the amplified signal into the air or wireless medium.

[0131] Figure 7 illustrates a flowchart of a method 700 in accordance with aspects of the present disclosure. The operations of the method may be implemented by a device (e.g., an NE and / or a UE) as described herein. In some implementations, the UE and / or the NE may execute a set of instructions to control the function elements of the UE and / or the NE to perform the described functions. It should be noted that the method described herein describes a possible implementation,and that the operations and the steps may be rearranged or otherwise modified and that other implementations are possible.

[0132] At 702, the method may include obtaining a set of data samples to provide as input to a set of learning models stored at the first device. The operations of 702 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 702 may be performed by a UE and / or the NE as described with reference to Figures 4 and 6.

[0133] At 704, the method may include generating, based on providing the set of data samples as input to the set of learning models, respective confidence scores associated with a set of outputs from the set of learning models. The operations of 704 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 704 may be performed by a UE and / or the NE as described with reference to Figures 4 and 6.

[0134] At 706, the method may include selecting, for executing at the first device or at a second device, a learning model of the set of learning models based on a confidence score associated with an output from the learning model satisfying a selection criterion. The operations of 706 may be performed in accordance with examples as described herein. In some implementations, aspects of the operations of 706 may be performed a UE and / or the NE as described with reference to Figures 4 and 6.

[0135] The description herein is provided to enable a person having ordinary skill in the art to make or use the disclosure. Various modifications to the disclosure will be apparent to a person having ordinary skill in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. Thus, the disclosure is not limited to the examples and designs described herein but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein.

Claims

CLAIMSWhat is claimed is:

1. A first device, comprising: at least one memory; and at least one processor coupled with the at least one memory and operable to cause the first device to: obtain a set of data samples to provide as input to a plurality of learning models stored at the first device; generate, based at least in part on providing the set of data samples as input to the plurality of learning models, respective confidence scores associated with a plurality of outputs from the plurality of learning models; and select, for executing at the first device or at a second device, a learning model of the plurality of learning models based at least in part on a confidence score associated with an output from the learning model satisfying a selection criterion.

2. The first device of claim 1 , wherein the at least one processor is further operable to cause the first device to transmit, to the second device, a message that indicates the learning model.

3. The first device of claim 1, wherein to obtain the set of data samples, the at least one processor is operable to cause the first device to receive, from the second device, a message that indicates the set of data samples.

4. The first device of claim 1, wherein to obtain the set of data samples, the at least one processor is operable to cause the first device to: receive, from the second device, a plurality of reference signals; and obtain information associated with the plurality of reference signals, wherein the information comprises one or more of a reference signal received power (RSRP) measurement, a reference signal received quality (RSRQ) measurement, a channel quality indicator (CQI), a precoding matrix indicator (PMI), a modulation and coding scheme (MCS) index, or a rank indicator.

5. The first device of claim 1, wherein the at least one processor is further operable to cause the first device to receive, from the second device, a message that triggers selection of the learning model.

6. The first device of claim 1, wherein to generate the respective confidence scores, the at least one processor is operable to cause the first device to generate respective measures of uncertainty associated with the plurality of outputs from the plurality of learning models based at least in part on the set of data samples, wherein the selection criterion comprises a minimum measure of uncertainty of the respective measures of uncertainty.

7. The first device of claim 6, wherein the respective measures of uncertainty comprise at least one of a marginal entropy of the plurality of outputs or a conditional entropy of the plurality of outputs based at least in part on the set of data samples.

8. The first device of claim 1, wherein to generate the respective confidence scores, the at least one processor is operable to cause the first device to generate respective first measures of uncertainty associated with the plurality of outputs from the plurality of learning models and respective second measures of uncertainty associated with the plurality of outputs from the plurality of learning models based at least in part on the set of data samples, wherein the selection criterion comprises a maximum difference between the respective first measures of uncertainty and the respective second measures of uncertainty.

9. The first device of claim 1, wherein to generate the respective confidence scores, the at least one processor is operable to cause the first device to generate respective first weighted measures of uncertainty associated with the plurality of outputs from the plurality of learning models and respective second weighted measures of uncertainty associated with the plurality of outputs from the plurality of learning models based at least in part on the set of data samples, wherein the respective first weighted measures of uncertainty and the respective second weighted measures of uncertainty are weighted based at least in part on a value between zero and one.

10. The first device of claim 1, wherein to generate the respective confidence scores, the at least one processor is operable to cause the first device to generate an information gain associatedwith the plurality of outputs from the plurality of learning models based at least in part on the set of data samples.

11. The first device of claim 1 , wherein to execute the learning model, the at least one processor is operable to cause the first device to: generate, as the output from the learning model, a prediction of at least one of a message, a set of symbols, or a set of bits based at least in part on providing the set of data samples as input to the learning model, wherein the set of data samples comprise one or more channel output signals; generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling based at least in part on providing the set of data samples as input to the learning model, wherein the set of data samples comprise at least one of a set of beam measurements or assistance information associated with one or more beams; generate, as the output from the learning model, a prediction of at least one beam to use for transmitting or receiving signaling in a set of future time slots based at least in part on providing the set of data samples as input to the learning model, wherein the set of data samples comprise at least one of a set of beam measurements or assistance information associated with one or more beams; or generate, as the output from the learning model, a prediction of channel state information (CSI) based at least in part on providing the set of data samples as input to the learning model, wherein the set of data samples comprise at least one of a set of reference signal measurements or a set of channel vectors, and wherein the CSI comprises at least one of a channel quality indicator (CQI) associated with a set of time-frequency resources, a precoding matrix indicator (PMI), a modulation and coding scheme (MCS) index, or a rank indicator.

12. The first device of claim 11, wherein the set of channel vectors are associated with a first subset of time-frequency resources of a set of time-frequency resources, and wherein the output from the learning model is associated with a second subset of time-frequency resources of the set of time-frequency resources or the set of time-frequency resources.

13. The first device of claim 1, wherein the first device is a network equipment (NE).

14. The first device of claim 1, wherein the first device is a user equipment (UE).

15. A processor, comprising:at least one controller coupled with at least one memory and operable to cause the processor to: obtain a set of data samples to provide as input to a plurality of learning models stored at the processor; generate, based at least in part on providing the set of data samples as input to the plurality of learning models, respective confidence scores associated with a plurality of outputs from the plurality of learning models; and select, for executing at the processor or at a device, a learning model of the plurality of learning models based at least in part on a confidence score associated with an output from the learning model satisfying a selection criterion.

16. A method performed by a first device, the method comprising: obtaining a set of data samples to provide as input to a plurality of learning models stored at the first device; generating, based at least in part on providing the set of data samples as input to the plurality of learning models, respective confidence scores associated with a plurality of outputs from the plurality of learning models; and selecting, for executing at the first device or at a second device, a learning model of the plurality of learning models based at least in part on a confidence score associated with an output from the learning model satisfying a selection criterion.

17. The method of claim 16, further comprising transmitting, to the second device, a message that indicates the learning model.

18. The method of claim 16, wherein obtaining the set of data samples comprises receiving, from the second device, a message that indicates the set of data samples.

19. The method of claim 16, wherein obtaining the set of data samples comprises: receiving, from the second device, a plurality of reference signals; and obtaining information associated with the plurality of reference signals, wherein the information comprises one or more of a reference signal received power (RSRP) measurement, a reference signal received quality (RSRQ) measurement, a channel quality indicator (CQI), aprecoding matrix indicator (PMI), a modulation and coding scheme (MCS) index, or a rank indicator.

20. The method of claim 16, further comprising receiving, from the second device, a message that triggers selection of the learning model.

Images