DETAILED ACTION
This Office Action is in response to the Applicants' communication filed on December 30, 2025. Claims 1, 8, 21, 29 and 30 are amended. Claim 7 is cancelled. Claim 31 is New. Claims 1-6 and 8-31 are currently pending and have been examined.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Applicant’s arguments/remarks made in an amendment filed December 30, 2025, have been fully considered. In view of the amended claims 1, 8, 21, 29 and 30 and upon further consideration, a new ground(s) of rejection, necessitated by the amendments is made in view of different interpretation of the previously applied references as presented in this Office action. Applicant’s arguments with respect to claim(s) 1-6 and 8-31 are therefore moot.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-6 and 8-31 are rejected under 35 U.S.C. 103 as being unpatentable over US 20220116764 A1 (PEZESHKI et al.)(hereinafter PEZESHKI) in view of OPPO: "FS_AMMT: Annex- General principle of Federated Learning over 5GS", 3GPP TSG-SA WG1 e-Meeting #91, S1-203146, 2020-08-24 - 2020-09-02 (hereinafter OPPO) in view of WO 2024101064 (TANIMURA et al.)(hereinafter TANIMURA) and in further view of US 20250184234 A1 (Tong et al.)(hereinafter Tong).
In re claims 1 and 29, PEZESHKI discloses a method of wireless communication performed by a user equipment ([0117], “A method of wireless communication by a user equipment (UE)”. [0051], “UE 120”), the UE comprising: at least one memory (Fig. 2:282); and at least one processor (Fig. 2: 280) communicatively coupled with the at least one memory ([0009], “In other aspects of the present disclosure, an apparatus for wireless communications at a user equipment (UE) includes a processor and memory coupled with the processor. Instructions stored in the memory are operable, when executed by the processor, to cause the apparatus to transmit, to the base station, gradient updates or weight updates to the machine learning model”), the at least one processor configured to cause the UE to:
transmit, to a network node, an indication of a local training data distribution associated with the UE; receive configuration information that indicates a radio frequency configuration condition associated with a federated learning model; and
transmit, to the network node, local gradient information (Fig. 8: 806, [0007], “The method also transmits, to the base station, gradient updates or weight updates to the machine learning model”. [0032], “In each round of a federated learning process, a group of UEs sends back weights or gradient updates within a given time interval after they receive the model from the base station. If a UE misses the deadline for sending updates, the weights or gradients will become stale, and the base station will not incorporate the update in the weight or gradient aggregation for that local training round of the federated learning process”) for the federated learning model that is based at least in part on the local training data distribution associated with the UE ([0031], “Standard machine learning approaches centralize training data on one machine, or in a data center. A federated learning model supports collaborative learning of a shared prediction model among user equipment (UEs) and a base station (or centralized server). Federated learning is a process where a group of UEs receives a machine learning model from a base station and work together to train the model. More specifically, each UE trains the model locally, and sends back either updated neural network model weights or gradient updates from, for example, a locally performed stochastic gradient descent process. The base station receives the updates from all of the UEs in the group and aggregates them, for example by averaging them, to obtain updated global weights of the neural network. The base station sends the updated model to the UEs, and the process repeats, round after round, until a desired performance level from the global model is obtained”), wherein the local gradient information is transmitted based at least in part on the radio frequency configuration condition being satisfied.
PEZESHKI does not explicitly disclose transmit, to a network node, an indication of a local training data distribution associated with the UE.
OPPO discloses transmit, to a network node, an indication of a local training data distribution associated with the UE (Fig. x.1-1, page 2, lines 11-14, “In distributed learning mode, each computing node trains its own DNN model locally with local data, which preserves private information locally. To obtain the global DNN model by sharing local training improvement, the network will communicate with each other to exchange the local model update”. Page 3, first para, lines 1-6, “when finished the local training, a UE reports its interim training results to the FL server”. Page 3, training loss, lines 1-7, “For federated Learning, only when the valuable local training data can be fully learned in the duration of the iteration and the local training updates can be correctly reported to the cloud server within the target duration, the training loss can be minimized” (transmit to a network node local training data associated with the UE)).
It would have been obvious to one of ordinary skill in the art, at the time the invention was filed, to combine the teachings of PEZESHKI with OPPO to provide a method and apparatus for a machine learning iterative process for federated learning wherein each associated group of UEs train the model locally using stochastic gradient descent process, and sends back gradient updates to the base station which aggregates them and sends updated model to the UEs in a repeat iteration process to achieve the desired performance. The advantage of doing so is to capture the skewed local training data based on various conditions in the NR such as different signal-to-interference-plus-noise ratios (SINRs), environment conditions and scheduling scenarios to improve accuracy of federated learning algorithms trained on neural network architectures.
PEZESHKI and OPPO do not explicitly disclose receive configuration information that indicates a radio frequency configuration condition associated with a federated learning model; wherein the local gradient information is transmitted based at least in part on the radio frequency configuration condition being satisfied.
TANIMURA discloses receive configuration information that indicates a radio frequency configuration condition associated with a federated learning model; wherein the local gradient information is transmitted based at least in part on the radio frequency configuration condition being satisfied (Fig. 12, Page 5, lines 49-51, “In this way, the federated learning (FL) client 200 transmits the data property characteristic information of the above data and the extracted AI model Ext (AI_model_weight_i) to the FL server 100. The FL server 100 performs federated learning using the data received...” (here weights represent the local gradients). Page 5, lines 23-34, “The data property characteristic information is information that represents the characteristic properties of the data...Thereafter, the transmission/reception unit 207 transmits the data property characteristic information of the data to the FL server 100”. Page 6, lines 45-47, “The transmission/reception unit 207 of each federated learning (FL) client 200 transmits the data property characteristic information calculated by the data property characteristic information calculation unit 203 to the FL server 100 (S804)”. Page 10, lines 26-30, “...If the AI model feature learning unit 1201 determines that the common AI model after federated learning by the FL server 100 does not provide performance above a certain level, it learns the hyperparameters including the data extension method and strength distributed to the FL client 200 so that the common AI model satisfies a certain level of performance. The certain level of performance means, for example, satisfying the conditions specified for each FL client 200” (satisfying a certain level of performance is interpreted as condition being satisfied). Page 11, lines 17-19, “The FL server 100 performs federated learning using the data property characteristic information of the above data for each FL client 200 received from the FL client 200 and the extracted AI models”. Page 13, lines 8-11; 25-29, “In this way, each of the multiple client terminals outputs an input/output screen (interface screen 1500, FIG. 15) having an input area (G06) for inputting the specified parameters and a display area (G08) for outputting the determination result of whether or not the specified relationship between the data property characteristic information is satisfied...In response to this, in each of the above embodiments, the statistical properties are modified by applying individual data augmentation processing adjusted for each base to the local data held by each base, and the characteristics of the local data for each base are aligned. As a result, even if the data distribution of raw data for each base is highly heterogeneous, federated learning is possible while maintaining the performance of the joint AI model” (discloses federated learning utilizing data characteristics and transmitting level of performance based on certain conditions being satisfied. Here the conditions may include radio characteristics such as interference and signal quality)).
It would have been obvious to one of ordinary skill in the art, at the time the invention was filed, to combine the teachings of PEZESHKI and OPPO with TANIMURA to provide a method and apparatus for a machine learning iterative process for federated learning wherein each associated group of UEs train the model locally using stochastic gradient descent process, and sends back gradient updates to the base station which aggregates them and sends updated model to the UEs in a repeat iteration process to achieve the desired performance. The advantage of doing so is to capture the skewed local training data based on various conditions in the NR such as different signal-to-interference-plus-noise ratios (SINRs), environment conditions and scheduling scenarios to improve accuracy of federated learning algorithms trained on neural network architectures.
PEZESHKI, OPPO with TANIMURA do not explicitly disclose that the conditions maybe radio frequency conditions.
Tong discloses that the conditions maybe radio frequency conditions (Fig. 1, [0127], “An air interface generally includes a number of components and associated parameters that collectively specify how a transmission is to be sent and/or received over a wireless communications link between two or more communicating devices...The wireless communications link may support a link between a radio access network and user equipment (e.g. a “Uu” link), and/or the wireless communications link may support a link between device and device, such as between two user equipment’s (e.g. a “sidelink”)...” (radio conditions for an air interface). [0005], “However, communications in wireless communications systems, including communications associated with AI training at multiple nodes, typically occur over non-ideal channels. For example, non-ideal conditions such as electromagnetic interference, signal degradation, phase delays, fading, and other non-idealities may attenuate and/or distort a communication signal or may otherwise interfere with or degrade the communications capabilities of the system” (radio parameters for the model). [0157], “The network device 452 is part of a network (e.g. a radio access network 120)”. [0150], “The carrier, the BWP, or the occupied bandwidth may be signaled by a network device (e.g. base station) dynamically...or be determined by the UE as a function of other parameters that are known by the UE” (radio conditions of the UE). [0133], “As another example, a unified air interface may be self-contained in a frequency domain, and a frequency domain self-contained design may support more flexible radio access network (RAN) slicing through channel resource sharing between different services in both frequency and time”. [0165], “A gradient of the loss function is calculated with respect to the parameters of the DNN, and the calculated gradient is used (e.g., using a gradient descent algorithm) to update the parameters with the goal of minimizing the loss function” (discloses radio conditions for a UE and such conditions are taken into account by the federated learning model to transmit performance gradients). [0167], “Backpropagation involves computing a gradient of the loss function with respect to the parameters to be optimized, and a gradient algorithm (e.g., gradient descent) is used to update the parameters to reduce the loss function...After a training condition is satisfied (e.g., the loss function has converged, or a predefined number of training iterations have been performed), the neural network is considered to be trained. The trained neural network may be deployed (or executed) to generate inferred output data from input data”. [0152], “Future generations of networks may also have access to more accurate and/or new information (compared to previous networks) that may form the basis of inputs to AI models, e.g.: the physical speed/velocity at which a device is moving, a link budget of the device, the channel conditions of the device (radio frequency conditions), one or more device capabilities and/or a service type that is to be supported, sensing information, and/or positioning information, etc.”).
It would have been obvious to one of ordinary skill in the art, at the time the invention was filed, to combine the teachings of PEZESHKI, OPPO, TANIMURA with Tong to provide a method and apparatus for a machine learning iterative process for federated learning wherein each associated group of UEs train the model locally using stochastic gradient descent process, and sends back gradient updates to the base station which aggregates them and sends updated model to the UEs in a repeat iteration process to achieve the desired performance. The advantage of doing so is to capture the skewed local training data based on various conditions in the NR such as different signal-to-interference-plus-noise ratios (SINRs), environment conditions and scheduling scenarios to improve accuracy of federated learning algorithms trained on neural network architectures.
In re claim 2, the combination discloses the UE of claim 1, wherein OPPO discloses wherein the indication of the local training data distribution includes an indication of at least one of an input training data distribution of a local dataset associated with the UE or an output training data distribution of the local dataset associated with the UE (page 2, last para – page 3, first para, “...a training device starts training based on received global model and configuration...” (based on configuration and conditions inputs in the algorithm and the model provided an iteration is performed)).
In re claim 3, the combination discloses the UE of claim 1, wherein PEZESHKI discloses wherein the at least one processor is further configured to cause the UE to: receive, from the network node, an indication of a federated learning model index associated with the federated learning model ([0085], “The model parameters w(n) represent biases and weights of the global federated learning model 630, g(n) represents the gradient estimates, where n is a federated learning round index”).
In re claim 4, the combination discloses the UE of claim 1, wherein OPPO discloses wherein the at least one processor is further configured to cause the UE to receive, from the network node, an indication of a reporting resource associated with the federated learning model (page 2, last para-page 3, first para, “The candidate training devices report their computation resource available for the training task to the Federated Learning (FL) server...FL server will send the training configurations at the beginning of each iteration to the selected training devices together with global model for training...”), to cause the UE to transmit the local gradient information, the at least one processor is configured to cause the UE to: transmit the local gradient information in the reporting resource associated with the federated learning model (page 3, first para, “UE reports its interim training results (gradients for the DNN) to the FL server...if conditions not changed...same configuration for multiple iterations...” (reports updated gradients based on configuration)). PEZESHKI also discloses ([0104], “According to aspects of the present disclosure, a base station may configure the UE with the above-mentioned parameters for a particular federated learning process. The UE can then assess the amount of time for computing the weight or gradient updates with the knowledge of these parameters, and report the (approximate) turnaround time. For this option, as long as the above noted parameters are fixed for a given federated learning process, the UE refrains from sending an updated report. When the parameters are reconfigured, the UE sends an updated report” (reporting gradient in turnaround time)).
In re claim 5, the combination discloses the UE of claim 1, wherein PEZESHKI discloses wherein the at least one processor is further configured to cause the UE to receive, from the network node, configuration information that indicates respective reporting resources associated with multiple federated learning models, wherein the federated learning model that is based at least in part on the local training data distribution associated with the UE is a first federated learning model of the multiple federated learning models, and wherein, to cause the UE to transmit the local gradient information, the at least one processor is configured to cause the UE to: transmit the local gradient information in the respective reporting resource associated with the first federated learning model ([0076], “New architectures and training paradigms may further boost the performance of deep learning. Rectified linear units may reduce a training issue known as vanishing gradients. New training techniques may reduce over-fitting and thus enable larger models to achieve better generalization” (discloses resources associated with multiple federated learning models. See also “In re claim 4”)). OPPO also discloses (Fig. x.1-3, page 3, first para, lines 1-6, “training configurations are sent to the training devices at the beginning of each iteration ... the same group of training devices can participate the training with the same configuration for multiple iterations”. Also see under “Training loss” lines 1-15, “...learning local training data in the duration of iteration target loss can be minimized...resource allocation for the training update reporting”).
In re claim 6, the combination discloses the UE of claim 5, wherein OPPO discloses wherein the at least one processor is further configured to cause the UE to: transmit, to the network node, an indication of an updated local training data distribution associated with the UE (Figure x.1-3, Training loss, lines 11-12, “local training updates can be correctly reported to the cloud server...training configurations are needed before the iteration...”); and transmit other local gradient information in the respective reporting resource associated with a second federated learning model of the multiple federated learning models based at least in part on the updated local training data distribution associated with the UE (Figure x.1-3, Training loss, lines 1-15, “how well the trained DNN model fits the training data...local training updates can be correctly reported to the cloud server...”. Page 2, “the UE reports the interim training results (e.g. gradients for the DNN) to the cloud server via 5G UL channels. The server aggregates the gradients from the UEs and updates the global model...” (Global model distribution)). PEZESHKI also discloses ([0031], “Standard machine learning approaches centralize training data on one machine, or in a data center. A federated learning model supports collaborative learning of a shared prediction model among user equipment (UEs) and a base station (or centralized server). Federated learning is a process where a group of UEs receives a machine learning model from a base station and work together to train the model. More specifically, each UE trains the model locally, and sends back either updated neural network model weights or gradient updates from, for example, a locally performed stochastic gradient descent process. The base station receives the updates from all of the UEs in the group and aggregates them, for example by averaging them, to obtain updated global weights of the neural network. The base station sends the updated model to the UEs, and the process repeats, round after round, until a desired performance level from the global model is obtained” (based on updated neural network model weights or gradient updates, a second updated federated learning model is generated by the base station and this process repeats until a desired performance level from the global model is obtained)).
In re claim 8, the combination discloses the UE of claim 1, wherein PEZESHKI discloses wherein the configuration information indicates a scheduling condition ([0004], “For example, data distributions for local datasets collected by different UEs participating in federated learning can be highly skewed based at least in part on different signal-to-interference-plus-noise ratios (SINRs), environment conditions, scheduling scenarios, and/or operating characteristics, among other examples” (scheduling condition)).
In re claim 9, the combination discloses the UE of claim 1, wherein OPPO discloses wherein the at least one processor is further configured to cause the UE to receive configuration information that configures reporting of the local training data distribution associated with the UE, wherein, to cause the UE to transmit the indication of the local training data distribution, the at least one processor is configured to cause the UE to: transmit the indication of the local training data distribution based at least in part on the configuration information (Page 2, last para, “...receive training configuration information from the FL server...”. Page 3, first para, “...UE reports its interim training results based on the configuration...”. Also see page 3, training loss, lines 10-15, “...training configurations are needed before a training is performed ...and training updates reporting...”).
In re claim 10, the combination discloses the UE of claim 1, wherein PEZESHKI discloses wherein the indication of the local training data distribution indicates one or more statistical properties associated with the local training data distribution ([0093], “an average is computed by the gradient averaging block 612”. [0094], “Based on the average gradient gn , the updated model parameters are transmitted from the base station 610 to the UEs 620”).
In re claim 11, the combination discloses the UE of claim 1, wherein OPPO discloses wherein the indication of the local training data distribution indicate a P-value associated with the local training data distribution (page 4, first para, “...QOS for minimizing device group latency...for training update reporting needs to be guaranteed”).
In re claim 12, the combination discloses the UE of claim 1, wherein OPPO discloses wherein the indication of the local training data distribution indicates a respective input training data distribution for each of multiple inputs in a local dataset associated with the UE (page 3, first para, “multiple iterations” and “training loss” (multiple input in training data associated with the UE)).
In re claim 13, the combination discloses the UE of claim 1, wherein OPPO discloses wherein the indication of the local training data distribution indicates a plurality of Gaussian components of a Gaussian mixture distribution, and wherein, for each Gaussian component, of the plurality of Gaussian components, the indication includes: a respective mean, a respective covariance matrix, and a respective mixing probability (Fig. x.1-3, page 2, line 16, “The most agreeable Federated Learning algorithm so far is based on the iterative model averaging” (a person skilled in the art would use Gaussian distribution such as a mean, covariance matrix, and a mixing probability to incorporate the issues related to skewing of the local data due to local conditions for improved accuracy)).
In re claim 14, the combination discloses the UE of claim 13, wherein OPPO discloses wherein the at least one processor is further configured to cause the UE to: receive, from the network node, configuration information that indicates the plurality of Gaussian components (Fig. x.1-3, page 3, first para, lines 1-6, “training configurations are sent to the training devices at the beginning of each iteration”. See also “In re claim 13”).
In re claim 15, the combination discloses the UE of claim 13, wherein OPPO discloses wherein the at least one processor is further configured to cause the UE to: estimate the respective mean, the respective covariance matrix, and the respective mixing probability for each Gaussian component of the plurality of Gaussian components based at least in part on the local training data distribution (Fig. x.1-3, page 2, line 16, “The most agreeable Federated Learning algorithm so far is based on the iterative model averaging” (a person skilled in the art would use statistical calculations on the training data based on the model. See also “In re claim 13”).
In re claim 16, the combination discloses the UE of claim 1, wherein OPPO discloses wherein the indication of the local training data distribution indicates the local training data distribution as a mixture distribution including one or more components associated with a base distribution (a person skilled in the art can appreciate this as a minor design detail in lines with claim 13. See “In re claim 13”).
In re claim 17, the combination discloses the UE of claim 16, wherein OPPO discloses wherein the indication of the local training data distribution indicates, for each component of the one or more components of the mixture distribution: one or more parameters associated with the base distribution, and a mixing probability (a person skilled in the art can appreciate this as a minor design detail in lines with claim 13. See “In re claim 13” and “In re claim 16”. All features are covered in claims 13 and 16).
In re claim 18, the combination discloses the UE of claim 16, wherein OPPO discloses wherein the base distribution includes at least one of a uniform distribution, an exponential distribution, a Gaussian distribution, or an inverse Gaussian distribution (Fig. x.1-3, page 2, line 16, “The most agreeable Federated Learning algorithm so far is based on the iterative model averaging” (a person skilled in the art would use a statistical distribution in the base distribution for the federated learning model)).
In re claim 19, the combination discloses the UE of claim 16, wherein OPPO discloses wherein the at least one processor is further configured to cause the UE to: receive, from the network node, configuration information that indicates at least one of the base distribution or a maximum quantity of the one or more components of the mixture distribution (Fig. x.1-3, page 3, first para, lines 1-6, “training configurations are sent to the training devices at the beginning of each iteration”. See “In re claim 16”).
In re claim 20, the combination discloses the UE of claim 19, wherein Tong discloses wherein the at least one processor is further configured to cause the UE to: transmit, to the network node, capability information that indicates a capability of the UE for reporting the local training data distribution as the mixture distribution, wherein the capability information indicates at least one of a capability for the base distribution or a capability for the maximum quantity of the one or more components of the mixture distribution ([0014], “The method according to the first broad aspect of the present disclosure may include receiving, from a node, information including a report related to AI/ML capability of the node. The method according to the first broad aspect of the present disclosure may further include configuring the node based on the received information, wherein configuring the node includes configuring a node type of the node, the configured node type being one of a plurality of node types. For example, the plurality of node types could include: Type 1 indicative of a node configured to collect a plurality of AI/ML models and aggregate the collected AI/ML models for obtaining a first type AI/ML model, or Type 2 indicative of a node configured to obtain a second type AI/ML model with a set of training data without aggregation operation”. [0039], “the aggregation information to be used by the first node includes aggregation capability of the second node”. [0201], “The AI/ML model transfer may be a heterogenous AI/ML model transfer in which the AI/ML models transfer between different nodes in the topology may involve AI/ML models having different neural network structures. The AI/ML model transfer may include delivery of a full AI/ML model or a partial AI/ML model. The configured topology may include at least one of a connection between at least one aggregation node and zero or more basic nodes or a connection between at least two aggregation nodes”).
In re claims 21 and 30, PEZESHKI discloses a method of wireless communication performed by a network node, the network node ([0126], “A method of wireless communication by a base station”. [0051], “The base station 110”) comprising: at least one memory (Fig. 2: 242); and at least one processor (Fig. 2: 240) communicatively coupled with the at least one memory ([0010], “In other aspects of the present disclosure, an apparatus for wireless communications at a base station includes a processor and memory coupled with the processor. Instructions stored in the memory are operable, when executed by the processor, to cause the apparatus to transmit a machine learning model to a number of user equipment (UEs)”), the at least one processor configured to cause the network node to:
receive an indication of a local training data distribution associated with a user equipment (UE);
transmit configuration information that indicates a radio frequency configuration condition associated with the federated learning model;
assign the UE to a group of UEs associated with a federated learning model (Fig. 9:906, [0085], “group of user equipment (UEs) 620 (e.g., 620a, 620b, 620c)”. [0085], “a global federated learning model 630”). [0096], “For example, a base station may group the UEs for different federated learning rounds according to machine learning capability. If slower UEs are grouped with faster UEs, the slower UEs will be a bottleneck for the training procedure, adversely impacting a convergence time of the federated learning process. Thus, slower UEs may be grouped with other slower UEs, while fast UEs are grouped with other fast UEs. Moreover, different UEs can be paired together for different rounds of the federated learning training process” (grouping the UEs by the base station). [0108], “Based on the received machine learning capability reports, the base station 610 groups the UEs 620 at time t4 and schedules the UEs 620 in accordance with the groupings at time t5...The updates are computed locally at each UE 620, prior to transmission, and will be aggregated at the base station 610 for each round of federated learning”), of multiple federated learning models ([0076], “New architectures and training paradigms may further boost the performance of deep learning. Rectified linear units may reduce a training issue known as vanishing gradients. New training techniques may reduce over-fitting and thus enable larger models to achieve better generalization” (of the multiple federated learning models)), based at least in part on the local training data distribution associated with the UE ([0082], “In contrast, federated learning is a process where a group of UEs receives a machine learning model from a base station and work together to train the model. More specifically, each UE trains the model locally, and sends back either updated neural network model weights or gradient updates from, for example, a locally performed stochastic gradient descent process. The base station receives the updates from all of the UEs in the group and aggregates them, for example by averaging them, to obtain updated global weights of the neural network. The base station sends the updated model to the UEs, and the process repeats, round after round, until a desired performance level from the global model is obtained”); and
receive local gradient information associated with the UE for the federated learning model that is associated with the group of UEs ([0032], “In each round of a federated learning process, a group of UEs sends back weights or gradient updates within a given time interval after they receive the model from the base station. If a UE misses the deadline for sending updates, the weights or gradients will become stale, and the base station will not incorporate the update in the weight or gradient aggregation for that local training round of the federated learning process”. [0046], “For brevity, only one base station 110a is shown as including the ML capability reporting module 138. The ML capability group module 138 may transmit a machine learning model to multiple user equipment (UEs). The ML capability group module 138 may also receive, from each of the UEs, a machine learning processing capability report. The ML capability group module 138 may further group the UEs in accordance with the machine learning processing capability reports, for receiving gradient updates to the machine learning model”), wherein the local gradient information is received based at least in part on the radio frequency configuration condition being satisfied.
PEZESHKI does not explicitly disclose receive an indication of a local training data distribution associated with a user equipment (UE).
OPPO discloses receive an indication of a local training data distribution associated with a user equipment (UE) (Fig. x.1-1, page 2, lines 11-14, “In distributed learning mode, each computing node trains its own DNN model locally with local data, which preserves private information locally. To obtain the global DNN model by sharing local training improvement, the network will communicate with each other to exchange the local model update”. Page 3, first para, lines 1-6, “when finished the local training, a UE reports its interim training results to the FL server”. Page 3, training loss, lines 1-7, “For federated Learning, only when the valuable local training data can be fully learned in the duration of the iteration and the local training updates can be correctly reported to the cloud server within the target duration, the training loss can be minimized” (transmit to a network node local training data associated with the UE)).
It would have been obvious to one of ordinary skill in the art, at the time the invention was filed, to combine the teachings of PEZESHKI with OPPO to provide a method and apparatus for a machine learning iterative process for federated learning wherein each associated group of UEs train the model locally using stochastic gradient descent process, and sends back gradient updates to the base station which aggregates them and sends updated model to the UEs in a repeat iteration process to achieve the desired performance. The advantage of doing so is to capture the skewed local training data based on various conditions in the NR such as different signal-to-interference-plus-noise ratios (SINRs), environment conditions and scheduling scenarios to improve accuracy of federated learning algorithms trained on neural network architectures.
PEZESHKI and OPPO do not explicitly disclose transmit configuration information that indicates a radio frequency configuration condition associated with the federated learning model; wherein the local gradient information is received based at least in part on the radio frequency configuration condition being satisfied.
TANIMURA discloses transmit configuration information that indicates a radio frequency configuration condition associated with the federated learning model; wherein the local gradient information is received based at least in part on the radio frequency configuration condition being satisfied (Fig. 12, Page 5, lines 49-51, “In this way, the federated learning (FL) client 200 transmits the data property characteristic information of the above data and the extracted AI model Ext (AI_model_weight_i) to the FL server 100. The FL server 100 performs federated learning using the data received...” (here weights represent the local gradients). Page 5, lines 23-34, “The data property characteristic information is information that represents the characteristic properties of the data...Thereafter, the transmission/reception unit 207 transmits the data property characteristic information of the data to the FL server 100”. Page 6, lines 45-47, “The transmission/reception unit 207 of each federated learning (FL) client 200 transmits the data property characteristic information calculated by the data property characteristic information calculation unit 203 to the FL server 100 (S804)”. Page 10, lines 26-30, “...If the AI model feature learning unit 1201 determines that the common AI model after federated learning by the FL server 100 does not provide performance above a certain level, it learns the hyperparameters including the data extension method and strength distributed to the FL client 200 so that the common AI model satisfies a certain level of performance. The certain level of performance means, for example, satisfying the conditions specified for each FL client 200” (satisfying a certain level of performance is interpreted as condition being satisfied). Page 11, lines 17-19, “The FL server 100 performs federated learning using the data property characteristic information of the above data for each FL client 200 received from the FL client 200 and the extracted AI models”. Page 13, lines 8-11; 25-29, “In this way, each of the multiple client terminals outputs an input/output screen (interface screen 1500, FIG. 15) having an input area (G06) for inputting the specified parameters and a display area (G08) for outputting the determination result of whether or not the specified relationship between the data property characteristic information is satisfied...In response to this, in each of the above embodiments, the statistical properties are modified by applying individual data augmentation processing adjusted for each base to the local data held by each base, and the characteristics of the local data for each base are aligned. As a result, even if the data distribution of raw data for each base is highly heterogeneous, federated learning is possible while maintaining the performance of the joint AI model” (discloses federated learning utilizing data characteristics and transmitting level of performance based on certain conditions being satisfied. Here the conditions may include radio characteristics such as interference and signal quality)).
It would have been obvious to one of ordinary skill in the art, at the time the invention was filed, to combine the teachings of PEZESHKI and OPPO with TANIMURA to provide a method and apparatus for a machine learning iterative process for federated learning wherein each associated group of UEs train the model locally using stochastic gradient descent process, and sends back gradient updates to the base station which aggregates them and sends updated model to the UEs in a repeat iteration process to achieve the desired performance. The advantage of doing so is to capture the skewed local training data based on various conditions in the NR such as different signal-to-interference-plus-noise ratios (SINRs), environment conditions and scheduling scenarios to improve accuracy of federated learning algorithms trained on neural network architectures.
PEZESHKI, OPPO with TANIMURA do not explicitly disclose that the conditions maybe radio frequency conditions.
Tong discloses that the conditions maybe radio frequency conditions (Fig. 1, [0127], “An air interface generally includes a number of components and associated parameters that collectively specify how a transmission is to be sent and/or received over a wireless communications link between two or more communicating devices...The wireless communications link may support a link between a radio access network and user equipment (e.g. a “Uu” link), and/or the wireless communications link may support a link between device and device, such as between two user equipment’s (e.g. a “sidelink”)...” (radio conditions for an air interface). [0005], “However, communications in wireless communications systems, including communications associated with AI training at multiple nodes, typically occur over non-ideal channels. For example, non-ideal conditions such as electromagnetic interference, signal degradation, phase delays, fading, and other non-idealities may attenuate and/or distort a communication signal or may otherwise interfere with or degrade the communications capabilities of the system” (radio parameters for the model). [0157], “The network device 452 is part of a network (e.g. a radio access network 120)”. [0150], “The carrier, the BWP, or the occupied bandwidth may be signaled by a network device (e.g. base station) dynamically...or be determined by the UE as a function of other parameters that are known by the UE” (radio conditions of the UE). [0133], “As another example, a unified air interface may be self-contained in a frequency domain, and a frequency domain self-contained design may support more flexible radio access network (RAN) slicing through channel resource sharing between different services in both frequency and time”. [0165], “A gradient of the loss function is calculated with respect to the parameters of the DNN, and the calculated gradient is used (e.g., using a gradient descent algorithm) to update the parameters with the goal of minimizing the loss function” (discloses radio conditions for a UE and such conditions are taken into account by the federated learning model to transmit performance gradients). [0167], “Backpropagation involves computing a gradient of the loss function with respect to the parameters to be optimized, and a gradient algorithm (e.g., gradient descent) is used to update the parameters to reduce the loss function...After a training condition is satisfied (e.g., the loss function has converged, or a predefined number of training iterations have been performed), the neural network is considered to be trained. The trained neural network may be deployed (or executed) to generate inferred output data from input data”. [0152], “Future generations of networks may also have access to more accurate and/or new information (compared to previous networks) that may form the basis of inputs to AI models, e.g.: the physical speed/velocity at which a device is moving, a link budget of the device, the channel conditions of the device (radio frequency conditions), one or more device capabilities and/or a service type that is to be supported, sensing information, and/or positioning information, etc.”).
It would have been obvious to one of ordinary skill in the art, at the time the invention was filed, to combine the teachings of PEZESHKI, OPPO, TANIMURA with Tong to provide a method and apparatus for a machine learning iterative process for federated learning wherein each associated group of UEs train the model locally using stochastic gradient descent process, and sends back gradient updates to the base station which aggregates them and sends updated model to the UEs in a repeat iteration process to achieve the desired performance. The advantage of doing so is to capture the skewed local training data based on various conditions in the NR such as different signal-to-interference-plus-noise ratios (SINRs), environment conditions and scheduling scenarios to improve accuracy of federated learning algorithms trained on neural network architectures.
In re claim 22, the combination discloses the network node of claim 21, wherein PEZESHKI discloses wherein the at least one processor is further configured to cause the network node to: transmit, to the UE, an indication of a federated learning model index associated with the federated learning model ([0085], “The model parameters w(n) represent biases and weights of the global federated learning model 630, g(n) represents the gradient estimates, where n is a federated learning round index”).
In re claim 23, the combination discloses the network node of claim 21, wherein OPPO discloses wherein the at least one processor is further configured to cause the network node, to transmit to the UE, an indication of a reporting resource associated with the federated learning model (page 2, last para-page 3, first para, “The candidate training devices report their computation resource available for the training task to the Federated Learning (FL) server...FL server will send the training configurations at the beginning of each iteration to the selected training devices together with global model for training...”), wherein to cause the network node to receive the local gradient information, the at least one processor is configured to cause the network node to: receive the local gradient information in the reporting resource associated with the federated learning model (page 3, first para, “UE reports its interim training results (gradients for the DNN) to the FL server...if conditions not changed...same configuration for multiple iterations...” (reports updated gradients based on configuration)). PEZESHKI also discloses ([0104], “According to aspects of the present disclosure, a base station may configure the UE with the above-mentioned parameters for a particular federated learning process. The UE can then assess the amount of time for computing the weight or gradient updates with the knowledge of these parameters, and report the (approximate) turnaround time. For this option, as long as the above noted parameters are fixed for a given federated learning process, the UE refrains from sending an updated report. When the parameters are reconfigured, the UE sends an updated report” (reporting gradient in turnaround time)).
In re claim 24, the combination discloses the network node of claim 21, wherein PEZESHKI discloses wherein the at least one processor is further configured to cause the network node to transmit, to the UE, configuration information that indicates respective reporting resources associated with multiple federated learning models, wherein the federated learning model that is based at least in part on the local training data distribution associated with the UE is a first federated learning model of the multiple federated learning models, and wherein, to cause the network node to receive the local gradient information, the at least one processor is configured to cause the network node to: receive the local gradient information in the respective reporting resource associated with the first federated learning model ([0076], “New architectures and training paradigms may further boost the performance of deep learning. Rectified linear units may reduce a training issue known as vanishing gradients. New training techniques may reduce over-fitting and thus enable larger models to achieve better generalization” (discloses resources associated with multiple federated learning models. See also “In re claim 4”)). OPPO also discloses (Fig. x.1-3, page 3, first para, lines 1-6, “training configurations are sent to the training devices at the beginning of each iteration ... the same group of training devices can participate the training with the same configuration for multiple iterations”. Also see under “Training loss” lines 1-15, “...learning local training data in the duration of iteration target loss can be minimized...resource allocation for the training update reporting”).
In re claim 25, the combination discloses the network node of claim 21, wherein OPPO discloses wherein the at least one processor is further configured to cause the network node to transmit configuration information that configures reporting of the local training data distribution associated with the UE, wherein, to cause the network node to receive the indication of the local training data distribution, the at least one processor is configured to cause the network node to: receive the indication of the local training data distribution based at least in part on the configuration information (Page 2, last para, “...receive training configuration information from the FL server...”. Page 3, first para, “...UE reports its interim training results based on the configuration...”. Also see page 3, training loss, lines 10-15, “...training configurations are needed before a training is performed ...and training updates reporting...”).
In re claim 26, the combination discloses the network node of claim 21, wherein PEZESHKI discloses wherein the indication of the local training data distribution indicates one or more statistical properties associated with the local training data distribution ([0093], “an average is computed by the gradient averaging block 612”. [0094], “Based on the average gradient gn , the updated model parameters are transmitted from the base station 610 to the UEs 620”).
In re claim 27, the combination discloses the network node of claim 21, wherein OPPO discloses wherein the indication of the local training data distribution indicates a plurality of Gaussian components of a Gaussian mixture distribution, and wherein, for each Gaussian component, of the plurality of Gaussian components, the indication includes: a respective mean, a respective covariance matrix, and a respective mixing probability (Fig. x.1-3, page 2, line 16, “The most agreeable Federated Learning algorithm so far is based on the iterative model averaging” (a person skilled in the art would use Gaussian distribution such as a mean, covariance matrix, and a mixing probability to incorporate the issues related to skewing of the local data due to local conditions for improved accuracy)).
In re claim 28, the combination discloses the network node of claim 21, wherein OPPO discloses wherein the indication of the local training data distribution indicates the local training data distribution as a mixture distribution including one or more components associated with a base distribution (a person skilled in the art can appreciate this as a minor design detail in lines with claim 13. See “In re claim 13”).
In re claim 31, the combination discloses the network node of claim 21, wherein PEZESHKI discloses wherein the configuration information indicates a scheduling condition ([0004], “For example, data distributions for local datasets collected by different UEs participating in federated learning can be highly skewed based at least in part on different signal-to-interference-plus-noise ratios (SINRs), environment conditions, scheduling scenarios, and/or operating characteristics, among other examples” (scheduling condition)).
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Contact
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SWATI JAIN whose telephone number is (571)270-0699. The examiner can normally be reached Mon - Fri (830 am - 530 pm).
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Pan Yuwen can be reached on 5712727855. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/SWATI JAIN/Examiner, Art Unit 2649
/YUWEN PAN/Supervisory Patent Examiner, Art Unit 2649