DETAILED ACTION
The instant application having Application No. 18/670317 filed on 05/21/2024 is presented for examination by the examiner.
Examiner Notes
Examiner cites particular columns and line numbers in the references as applied to the claims below for the convenience of the applicant. Although the specified citations are representative of the teachings in the art and are applied to the specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested that, in preparing responses, the applicant fully consider the references in entirety as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior art or disclosed by the examiner.
Drawings
The applicant’s drawings submitted are acceptable for examination purposes.
Information Disclosure Statement
As required by M.P.E.P. 609, the applicant’s submissions of the Information Disclosure Statement dated 05/21/2024 is acknowledged by the examiner and the cited references have been considered in the examination of the claims now pending.
Claim Rejections - 35 USC § 103
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.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 13 and 20 are rejected under 35 U.S.C. 103(a) as being unpatentable over US 2009/0132867 to Stefansson et al. (hereafter “Stefansson”) in view of US 2019/0213475 to Erlandson, US 2020/0117513 to Li et al. (hereafter “li”), US 2015/0170053 to Miao, and US 2019/0392305 to Gu et al. (hereafter “Gu”).
As per claim 1, Stefansson discloses a system comprising:
a system transceiver configured to obtain a request to execute a task from a user device (FIG. 3; paragraph 0052: scheduler 310 receiving a job from client 300); and
a system processor communicatively coupled to the system transceiver, wherein the system processor is configured to:
obtain the request from the system transceiver (FIG. 3; paragraph 0052: “scheduler 310 may receive a job 340, and may distribute or divide job 340 into tasks (e.g., tasks 350-1, 350-2, 350-3, and 350-4). Scheduler 310 may send tasks 350-1, 350-2, 350-3, and 350-4 to hardware UE 200 (e.g., to processor 210-1, 210-2, 210-3, and 210-4, respectively) for execution” [Wingdings font/0xE0] scheduler distributing the job to UE 200 including processors 210-1 to 210-4).
Stefansson discloses to execute the task responsive to obtaining the request (FIG. 3; paragraph 0052), however, Stefansson does not explicitly disclose determine a machine learning (ML) model required to be implemented to execute the task;
determine a user device type;
determine a first ML sub-model, associated with the ML model, to be executed on the user device, and a second ML sub-model, associated with the ML model, to be executed on a server, based on the user device type; and
cause the user device to execute the first ML sub-model and the server to execute the second ML sub-model to execute the task.
Erlandson further discloses determine a machine learning (ML) model required to be implemented to execute the task (paragraph 0013, 0020 and 0028: “Based on determining that the particular version of the machine-learning model 114 that has the lowest descriptor value, the computing device 100 can select the particular version of the machine-learning model 114 for performing a task. The task can be a computing task, such as generating a predictive forecast, predicting a result, or analyzing data.”).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Erlandson into Stefansson’s teaching because it would provide for the purpose of selecting the particular version of the machine-learning model 114 to perform the task, as opposed to other versions of the machine-learning model, may result in the task being executed faster (Erlandson, paragraph 0020).
Li further discloses determine a user device type (paragraph 0019: “The MEC device 104 may determine the processing capability and/or one or more characteristics of the user device 102 to determine how the processes of application X are to be split. For example, the MEC device 104 may determine the processing capability of the user device 102 based on an identifier for the user device 102 (e.g., an international mobile equipment identity (IMEI)), a model of the user device 102, a type of the user device 102, and/or the like.”)
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Li into Stefansson’s teaching and Erlandson’s teaching because it would provide for the purpose of enabling a dynamic allocation of processes of an application to be split between a user device and a server device of the cloud environment (e.g., a MEC device of the MEC environment) based on one or more characteristics associated with the user device (Li, paragraph 0010).
Miao further discloses determine a first ML sub-model (paragraph 0046-0047: “selecting a subset of a machine learning model to load into RAM of a client device, according to various example embodiments”), associated with the ML model, to be executed on the user device, based on the user device type (paragraphs 0046-0047: “some parts of a machine learning model may be stored remotely and/or archived. In some implementations, the client device prioritizes various portions of the machine learning model to determine an order in which the various portions are loaded into RAM. Such prioritizing can be based, at least in part, on type or content of applications hosted by the client device, history or patterns of use of the client device, type of client device, and so on.”); and
cause the user device to execute the first ML sub-model (paragraphs 0046-0047).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Miao into Stefansson’s teaching, Erlandson’s teaching, and Li’s teaching because it would provide for the purpose of Personalizing machine learning may be performed locally at a personal computing device, and may include selecting a subset of a machine learning model to load into memory (Miao, paragraph 0002).
Gu further discloses a second ML sub-model, associated with the ML model, to be executed on a server (FIGs. 2 and 4; paragraphs 0026, 0042-0044, 0063 and 0081: “the BackNet subnet model 220 has already been provided by the client side operation and loaded into the privacy enhancing deep learning cloud service server computing device(s).”); and
the server to execute the second ML sub-model to execute the task (FIGs. 2 and 4; paragraphs 0026,0042-0044, 0063 and 0081: “The output shape of a FrontNet subnet model is compatible with the input shape of its corresponding BackNet subnet model. IR is delivered as an output for the FrontNet subnet model and is an input to the subsequent BackNet subnet model which continues the computation to get a result y*.” [Wingdings font/0xE0] the sub-model backnet executing the computation task).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Gu into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, and Miao’s teaching because it would provide for the purpose of Personalizing machine learning may be performed locally at a personal computing device, and may include selecting a subset of a machine learning model to load into memory (Miao, paragraph 0002).
As per claim 13, it is method claim, which recite(s) the same limitations as those of claim 1. Accordingly, claim 13 is rejected for the same reasons as set forth in the rejection of claim 1.
As per claim 20, it is method claim, which recite(s) the same limitations as those of claim 1. Accordingly, claim 20 is rejected for the same reasons as set forth in the rejection of claim 1.
Claims 2 and 14 are rejected under 35 U.S.C. 103(a) as being unpatentable over Stefansson in view of Erlandson, Li, Miao, and Gu, as applied to claim 1, and further in view of US 2023/0342278 to Padmanabha et al. (hereafter “Padmanabha”)
As per claim 2, Stefansson does not explicitly disclose calculate a required computation load to execute the ML model; and
determine the first ML sub-model and the second ML sub-model based on the required computation load.
Padmanabha further discloses calculate a required computation load to execute the ML model (FIGs. 2 and 4; paragraphs 0005, 0053-0056 and 0076: “The load forecaster component 22 outputs the load forecast 28 and an optimizer component 32 may use the load forecast 28 in combination with the machine learning model information 18 to determine one or more split locations 34 in the machine learning model 16 to divide the machine learning model 16 into smaller portions.”); and
determine the first ML sub-model and the second ML sub-model based on the required computation load (FIGs. 2 and 4; paragraphs 0005, 0053-0056 and 0076: “if the input batch size 42 is sixteen and the load forecast 28 for the batch size at the split location 34 is eight, the machine learning model 16 may be divided in half by the optimizer component 32 into two machine learning model portions 36.”).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Padmanabha into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Miao’s teaching, and Gu’s teaching because it would provide for the purpose of receiving a load forecast for a machine learning model to process received requests and generate inferences for the received requests (Padmanabha, paragraph 0006).
As per claim 14, it is method claim, which recite(s) the same limitations as those of claim 2. Accordingly, claim 14 is rejected for the same reasons as set forth in the rejection of claim 2.
Claims 3, 6, 15 and 18 are rejected under 35 U.S.C. 103(a) as being unpatentable over Stefansson in view of Erlandson, Li, Miao, and Gu, as applied to claims 1 and 13, and further in view of US 2017/0277994 to Sharifi et al. (hereafter “Sharifi”)
As per claim 3, Stefansson does not explicitly disclose determine available computing resources of the user device, from a plurality of computing resources, to execute the ML model; and
determine the first ML sub-model and the second ML sub-model based on the available computing resources.
Sharifi further discloses determine available computing resources of the user device, from a plurality of computing resources, to execute the ML model (FIGs. 3A-C; paragraph 0026-0027: comparing the speed/bandwidth of the client device/server with a threshold); and
determine the first ML sub-model and the second ML sub-model based on the available computing resources (FIGs. 3A-C; paragraph 0026-0027: dividing/assigning/partitioning model layers into client device/layer depending on the speed/bandwidth good/bad).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Sharifi into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Miao’s teaching, and Gu’s teaching because it would provide for the purpose of depending on the processing task, one of the client and server computing devices can select one of a plurality of ANNs each defining a distribution of the processing task (Sharifi, paragraph 0016).
As per claim 6, Stefansson does not explicitly disclose determine a network status associated with the user device; and
determine the first ML sub-model and the second ML sub-model based on the network status.
Sharifi further discloses determine a network status associated with the user device (FIGs. 3A-C; paragraph 0026-0027: comparing the speed/bandwidth of the client device/server with a threshold); and
determine the first ML sub-model and the second ML sub-model based on the network status (FIGs. 3A-C; paragraph 0026-0027: dividing/assigning/partitioning model layers into client device/layer depending on the speed/bandwidth good/bad).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Sharifi into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Miao’s teaching, and Gu’s teaching because it would provide for the purpose of depending on the processing task, one of the client and server computing devices can select one of a plurality of ANNs each defining a distribution of the processing task (Sharifi, paragraph 0016).
As per claim 15, it is method claim, which recite(s) the same limitations as those of claim 3. Accordingly, claim 15 is rejected for the same reasons as set forth in the rejection of claim 3.
As per claim 18, it is method claim, which recite(s) the same limitations as those of claim 6. Accordingly, claim 18 is rejected for the same reasons as set forth in the rejection of claim 6.
Claims 4 and 16 are rejected under 35 U.S.C. 103(a) as being unpatentable over Stefansson in view of Erlandson, Li, Miao, and Gu, as applied to claims 1 and 13, and further in view of Sharifi and US 2012/0131591 to Moorthi et al. (hereafter “Moorthi”)
As per claim 4, Stefansson does not explicitly disclose obtain additional inputs to execute the task, wherein the additional inputs comprise one or more of a latency, a cost, an accuracy, or privacy; and
determine the first ML sub-model and the second ML sub-model based on the additional inputs.
Sharifi further discloses determine the first ML sub-model and the second ML sub-model based on the additional inputs (paragraphs 00256-0028).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Sharifi into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Miao’s teaching, and Gu’s teaching because it would provide for the purpose of depending on the processing task, one of the client and server computing devices can select one of a plurality of ANNs each defining a distribution of the processing task (Sharifi, paragraph 0016).
Moorthi further discloses obtain additional inputs to execute the task, wherein the additional inputs comprise one or more of a latency, a cost, an accuracy, or privacy (paragraph 0054)
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Moorthi into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Miao’s teaching, Gu’s teaching, and Sharifi’s teaching because it would provide for the purpose of balancing available compute cycles and costs associated with different compute provider platforms with any translation costs required to separate a compute job into partitions handled by different compute providers (Moorthi, paragraph 0004).
As per claim 16, it is method claim, which recite(s) the same limitations as those of claim 4. Accordingly, claim 16 is rejected for the same reasons as set forth in the rejection of claim 4.
Claims 5 and 17 are rejected under 35 U.S.C. 103(a) as being unpatentable over Stefansson in view of Erlandson, Li, Miao, and Gu, as applied to claims 1 and 13, and further in view US 2022/0109742 to Kumar et al. (hereafter “Kumar”)
As per claim 5, Stefansson does not explicitly disclose determine a battery status of the user device; and
determine the first ML sub-model and the second ML sub-model based on the battery status.
Kumar further discloses determine a battery status of the user device (paragraphs 0025-0026, 0028, 0061 and 0070-0072); and
determine the first ML sub-model and the second ML sub-model based on the battery status (paragraphs 0025-0026, 0028, 0061 and 0070-0072).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Kumar’s teaching into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Miao’s teaching, and Gu’s teaching because it would provide for the purpose of the sum of the estimated computation energy consumption and the transmission energy consumption is above the available battery supply power or above the decided energy consumption threshold (e.g., power usage limit), the example processor circuitry determines that the first portion is too large and will adjust the number of layers in the first portion accordingly (Kumar, paragraph 0028).
As per claim 17, it is method claim, which recite(s) the same limitations as those of claim 5. Accordingly, claim 17 is rejected for the same reasons as set forth in the rejection of claim 5.
Claims 7 and 19 are rejected under 35 U.S.C. 103(a) as being unpatentable over Stefansson in view of Erlandson, Li, Miao, and Gu, as applied to claims 1 and 13, and further in view of US 2021/0390486 to Chu et al. and US 2021/0365806 to Sumanth et al. (hereafter “Sumanth”)
As per claim 7, Stefansson does not explicitly disclose determine that the user device is idle; and
cause the user device to execute the first ML sub-model responsive to determining that the user device is idle.
Chu further discloses determine that the user device is idle (paragraph 0076 and 0133)
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Chu into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Miao’s teaching, and Gu’s teaching because it would provide for the purpose of processing data about a sequence of tasks and rank the tasks based on state information to prioritize the tasks (Chu, paragraph 0005).
Sumanth further discloses cause the user device to execute the first ML sub-model responsive to determining that the user device is idle (paragraphs 0062 and 0127)
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Moorthi into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Miao’s teaching, Gu’s teaching, and Chu’s teaching because it would provide for the purpose of the suggestions can be provided on a user interface for a user to select, thereby increasing efficiency for the user, who would otherwise have to perform additional actions or keystrokes to perform the selection (Sumanth, paragraph 0002).
As per claim 19, it is method claim, which recite(s) the same limitations as those of claim 7. Accordingly, claim 17 is rejected for the same reasons as set forth in the rejection of claim 7.
Claim 8 is rejected under 35 U.S.C. 103(a) as being unpatentable over Stefansson in view of Erlandson, Li, Miao, and Gu, as applied to claim 1, and further in view of 2022/0245459 to Laskaridis et al. (hereafter “Laskaridis”)
As per claim 8, Stefansson does not explicitly disclose cause fetch the first ML sub-model from the server responsive to determining the first ML sub-model;
transmit the first ML sub-model from the server to the user device; and
cause the user device to execute the first ML sub-model, responsive to transmitting the first ML sub-model.
Miao further discloses cause the user device to execute the first ML sub-model, responsive to transmitting the first ML sub-model (paragraphs 0046-0047).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Miao into Stefansson’s teaching, Erlandson’s teaching, and Li’s teaching because it would provide for the purpose of Personalizing machine learning may be performed locally at a personal computing device, and may include selecting a subset of a machine learning model to load into memory (Miao, paragraph 0002).
Laskaridis further discloses wherein the system processor is further configured to:
fetch the first ML sub-model from the server responsive to determining the first ML sub-model (FIGs 4A-B; paragraphs 0039-0042);
transmit the first ML sub-model from the server to the user device (FIGs 4A-B; paragraphs 0039-0042).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Laskaridis into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Mao’s teaching, and Gu’s teaching because it would provide for the purpose of a client device which is being used to perform other tasks at training time (e.g. capturing images, participating in a call, etc.) may still be able to participate in training by extracting a submodel based on the resources available for training at that particular time (Laskaridis, paragraph 0044).
Claims 9-10 are rejected under 35 U.S.C. 103(a) as being unpatentable over Stefansson in view of Erlandson, Li, Miao, and Gu, as applied to claim 1, and further in view of US 2025/0165803 to Huang et al. (hereafter “Huang’)
As per claim 9, Stefansson does not explicitly disclose wherein the system processor is further configured to transmit a first command signal to the user device to execute the first ML sub-model on the user device.
Huang further discloses wherein the system processor is further configured to transmit a first command signal to the user device to execute the first ML sub-model on the user device (paragraph 0080).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Huang into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Mao’s teaching, and Gu’s teaching because it would provide for the purpose of a machine learning model implemented at the user device at which the user request is initiated may be executed based on a command transmitted from the model execution orchestration tool (Huang, paragraph 0080).
As per claim 10, Stefansson does not explicitly disclose wherein the system processor is further configured to transmit a second command signal to the server to execute the second ML sub-model on the server.
Huang further discloses wherein the system processor is further configured to transmit a second command signal to the server to execute the second ML sub-model on the server (paragraph 0080).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Huang into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Mao’s teaching, and Gu’s teaching because it would provide for the purpose of a machine learning model implemented at the user device at which the user request is initiated may be executed based on a command transmitted from the model execution orchestration tool (Huang, paragraph 0080).
Claims 11-12 are rejected under 35 U.S.C. 103(a) as being unpatentable over Stefansson in view of Erlandson, Li, Miao, and Gu, as applied to claim 1, and further in view of Huang, and further in view of US 2023/0395089 to Ekstrand et al. (hereafter “Ekstrand”)
As per claim 11, Stefansson does not explicitly disclose wherein the system processor is further configured to cause the user device to execute the first ML sub-model and the server to execute the second ML sub-model sequentially.
Huang further discloses configured to cause the user device to execute the first ML sub-model and the server to execute the second ML sub-model (paragraph 0080).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Huang into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Mao’s teaching, and Gu’s teaching because it would provide for the purpose of a machine learning model implemented at the user device at which the user request is initiated may be executed based on a command transmitted from the model execution orchestration tool (Huang, paragraph 0080).
Ekstrand further discloses executing the first sub-model and the second sub-model sequentially (paragraph 0088)
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Ekstrand into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Mao’s teaching, Gu’s teaching, and Huang’s teaching because it would provide for the purpose of the model may be configured to reconstruct all the bands in a single step, which eliminates the need for sequential execution of the MLP sub-layers. The model can be also configured to output fewer bands than the full number of channels at the time, which would then require using more than one of the MLP sub-layers operating in a sequence (Ekstrand, paragraph 0088).
As per claim 12, Stefansson does not explicitly disclose wherein the system processor is further configured to cause the user device to execute the first ML sub-model and the server to execute the second ML sub-model simultaneously.
Huang further discloses configured to cause the user device to execute the first ML sub-model and the server to execute the second ML sub-model (paragraph 0080).
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Huang into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Mao’s teaching, and Gu’s teaching because it would provide for the purpose of a machine learning model implemented at the user device at which the user request is initiated may be executed based on a command transmitted from the model execution orchestration tool (Huang, paragraph 0080).
Ekstrand further discloses executing the first sub-model and the second sub-model simultaneously (paragraph 0088)
It would have been obvious to a person having ordinary skill in the art at the time before the effective filling date of the claimed invention to combine a teaching of Ekstrand into Stefansson’s teaching, Erlandson’s teaching, Li’s teaching, Mao’s teaching, Gu’s teaching, and Huang’s teaching because it would provide for the purpose of the model may be configured to reconstruct all the bands in a single step, which eliminates the need for sequential execution of the MLP sub-layers. The model can be also configured to output fewer bands than the full number of channels at the time, which would then require using more than one of the MLP sub-layers operating in a sequence (Ekstrand, paragraph 0088).
Conclusion
The following prior art made of record and not relied upon is cited to establish the level of skill in the applicant’s art and those arts considered reasonably pertinent to applicant’s disclosure. See MPEP 707.05(c).
Prior arts:
US 2022/0391776 to Mody
In cases where the ML model is run on each frame of a 30 frame per second video, the ML model may be executed on a particular core, such as core 402A, 30 times per second. In some cases, multiple ML models may be executed on a single core 402. Other ML models, such as ML models 406B . . . 406N, may be initialized and continue to run on other cores, such as cores 402B, . . . 402N. These ML models 406 may execute concurrently and asynchronously. That is, multiple ML models 406 may run at the same time without synchronization as between the ML models 406.
US 2022/038084 to Lamy
(i) assigning sequentially-ordered layers of a machine learning model to a plurality of compute nodes, each of the layers being assigned to exactly one of the nodes; (ii) dividing training data into micro-batches; (iii) forward-propagating the micro-batches through the model, each node operating in parallel to generate respective activation states for the micro-batches with their assigned layers, and with the activation states being communicated between the nodes according to the layers' sequential ordering
US 2022/0300618 to Ding
the server may split the initial ML model differently for different clients. To illustrate, in implementations in which the initial ML model is a neural network (NN) having multiple layers, the server may split the initial ML model such that a first subset of layers corresponds to the first partial ML model and a second subset of layers corresponds to the third partial ML model.
US 2021/0133555 to Qiu
In asynchronous distributed training strategy of, workers may perform further learning tasks immediately after completing a previous learning task. More specifically, in a cluster with data parallelism, sub-models are learned separately with different shares of workers. One worker may immediately start a learning task for a next sub-model after completing a learning task for a previous sub-model and need not wait for other workers to finish their learning tasks for the previous sub-model.
Any inquiry concerning this communication should be directed to examiner Tuan Dao, whose telephone/fax numbers are (571) 270 3387 and (571) 270 4387, respectively. The examiner can normally be reached on every Monday-Thursday, and the second Friday of the bi-week from 7:30AM to 5:00PM.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Pierre Vital, can be reached at (571) 272 4215.
The fax phone number for the organization where this application or proceeding is assigned is (571) 273 8300.
Any inquiry of a general nature of relating to the status of this application or proceeding should be directed to the TC 2100 Group receptionist whose telephone number is (571) 272 2100.
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) Form at https://www.uspto.gov/patents/uspto-automated- interview-request-air-form.
/TUAN C DAO/ Primary Examiner, Art Unit 2198
Prosecution Insights