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 filed 2 June 2026 have been fully considered but they are not persuasive.
Applicant contends that “Subramanian’s reward/penalty system is directed to improving the AI model’s learning over time, not to real-time remediation of resource insufficiency during workload execution;” therefore Subramanian does not teach the recited “automatically allocate one or more additional computing resources responsive to determining that the one or more computing resources are insufficient during execution of the one or more workloads.” Remarks at 9-10.
The Examiner respectfully disagrees. Subramanian method is applied to “a repeat performance of a workload whose prior execution failed an SLA requirement.” Subramanian at col. 7:1-3. Further, Subramanian teaches that telemetry data is collected for workload that is being performed – col. 4:48-50, “[f]or a workload request that is being performing or has completed, one or more of the following can be collected: telemetry data” and col. 10:45-46, “[w]hile the workload is running and potentially after its completion, telemetry data is collected.” Subramanian determines recommendations for resource reconfiguration based on the telemetry data and resource reconfigurations can be applied – col. 12:12-17, “[n]eural network 622 can receive inputs related to workload performance such as telemetry data, performance data, boundedness, accumulated reward, configured resources and provide a variety of outputs that can be candidate suggestions for suggested resource configuration” and col. 3:53-55, “[p]od manager 114 can accept or reject the resource allocation suggestion from accelerator 116.”
The Examiner finds that Subramanians system that collects telemetry data from currently executing workloads, determined resource reconfigurations to achieve SLA requirements, and accepting the resource reconfigurations teaches or at least suggests to a person having ordinary skill the recited “automatically allocate one or more additional computing resources responsive to determining that the one or more computing resources are insufficient during execution of the one or more workloads.”
Applicant contends that Da Silva does not teach “a confidence level indicating a probability that the predicted one or more computing resources will satisfy needs of the one or more workloads,” because “Da Silva’s confidence value indicates whether a generic prediction … is correct.” Remarks at 11-12.
The Examiner respectfully disagrees. Examiner notes MPEP § 2145, “One cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references.” Here, Subramanian teaches a prediction of resource configuration for a workload (col. 5:43-45, “AI model 222 runs its inference model and produces a predicted best or recommended resource configuration for that workload”). Da Silva is combined with Subramanian’s prediction to establish that the prediction comprises a confidence level indicating a probability (Da Silva, ¶ 32, “A confidence value is a value that indicates a confidence of a prediction. For example, the confidence value may accompany a prediction (e.g., processing workload prediction, processor type prediction, etc.) and may indicate a likelihood that the prediction is correct”).
The Examiner finds that a person having ordinary skill would have found the recited “a confidence level indicating a probability that the predicted one or more computing resources will satisfy needs of the one or more workloads” obvious in view of Subramanian’s prediction of a resource configuration for a workload combined with Da Silva’s teaching of a confidence level indicating a probability that a predicting computing resource will satisfy needs of a workload.
Regarding arguments directed to claim 6, they a moot in view of the new grounds of rejection provided below.
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 (i.e., changing from AIA to pre-AIA ) 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.
The factual inquiries 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.
Claim(s) 1-3, 5, 7-14, 16, 18-22, 23-26, and, 28-30 is/are rejected under 35 U.S.C. 103 as being unpatentable over Subramanian (US 11,507,430) and further in view of Abdulaal (US 2022/0138019) and further in view of Da Silva (US 2022/0156639).
Regarding claim 1, Subramanian teaches: A processor, comprising: one or more circuits to use one or more neural networks that receive as inputs both respective characteristics of one or more workloads (col. 5:35-37, “Accelerator 220 provides to AI model 222 the workload parameters from pod manager 210”) and one or more current system conditions (col. 5:37-43, “and also information related to the workload from workload table 224. Workload table 224 keeps track of previously performed workloads and their characteristics such as one or more of: boundedness (e.g., utilization of one or more of: processor, memory, network, storage, or cache), applied resource allocations, telemetry data, or workload performance characteristic(s)”) to predict one or more computing resources to perform one or more workloads (col. 5:43-45, “AI model 222 runs its inference model and produces a predicted best or recommended resource configuration for that workload”) based at least on analyzing, concurrently, the respective characteristics of the one or more workloads and one or more current system conditions (col. 4:65-67 and col. 5:1-3, “the AI model can consider any of measured telemetry data, performance indicators, boundedness, utilized compute resources, or evaluation or monitoring of the application performance (including the application's own evaluation of its performance)”) and to cause allocation of the one or more computing resources for execution of the one or more workloads (col. 5:49-50, “Pod manager 210 can choose to accept or reject the recommended resource configuration”); and
automatically allocate one or more additional computing resources responsive to determining that the one or more computing resources are insufficient during execution of the one or more workloads (col. 4:48-50, “[f]or a workload request that is being performing or has completed, one or more of the following can be collected: telemetry data;” col. 10:45-46, “[w]hile the workload is running and potentially after its completion, telemetry data is collected” col. 12:12-17, “[n]eural network 622 can receive inputs related to workload performance such as telemetry data, performance data, boundedness, accumulated reward, configured resources and provide a variety of outputs that can be candidate suggestions for suggested resource configuration;” and col. 3:53-55, “[p]od manager 114 can accept or reject the resource allocation suggestion from accelerator 116”).
Subramanian does not teach as clearly as Abdulaal teaches: predict one or more computing resources to perform one or more workloads (¶ 2, “generating performance predictions associated with the compliant hardware configurations using the workload features, a portion of the hardware specification information associated with the compliant hardware configurations, and a second machine learning model; generating a recommendation using the performance predictions, and the recommendation specifies a hardware configuration of the compliant hardware configurations”).
It would have been obvious to a person having ordinary skill in the art, at the effective filing date of the invention, to have applied the known technique of predict one or more computing resources to perform one or more workloads, as taught by Abdulaal, in the same way to the inputs, as taught by Subramanian. Both inventions are in the field of forecasting neural network or machine learning computing needs, and combining them would have predictably resulted in solving the problem of “poor resource allocation and utilization on the hardware platforms,” as indicated by Abdulaal (¶ 1).
Subramanian and Abdulaal do not teach, however, Da Silva teaches: determine a confidence level indicating a probability that the predicted one or more computing resources will satisfy needs of the one or more workloads (¶ 32, “A confidence value is a value that indicates a confidence of a prediction. For example, the confidence value may accompany a prediction (e.g., processing workload prediction, processor type prediction, etc.) and may indicate a likelihood that the prediction is correct”).
It would have been obvious to a person having ordinary skill in the art, at the effective filing date of the invention, to have applied the known technique of determine a confidence level indicating a probability that the predicted one or more computing resources will satisfy needs of the one or more workloads, as taught by Da Silva, in the same way to predicted one or more computing resources, as taught by Subramanian and Abdulaal. Both inventions are in the field of predicting resource usage of machine learning/neural networks, and combining them would have predictably resulted in “reducing loading delay for a project or projects,” as indicated by Da Silva (¶ 36).
Regarding claim 2, Abdulaal teaches: The processor of claim 1, wherein the one or more neural networks are further to recommend one or more nodes of a data center to perform the one or more workloads (¶ 25, “The recommendation may include a data node identifier and one or more hardware component identifiers”), wherein the one or more nodes comprise the one or more computing resources (¶ 3, “a data node of the data nodes includes a processor and memory”).
Regarding claim 3, Subramanian teaches: The processor of claim 2, further comprising: the one or more circuits to receive selection of the recommended one or more nodes of the data center (col. 3:53-55, “Pod manager 114 can accept or reject the resource allocation suggestion from accelerator 116”) and to allocate the recommended one or more nodes of the data center for the one or more workloads (claim 5, “causing at least a portion of the computing resources to perform the workload using the recommendation of the computing resource allocation”).
Regarding claim 5, Abdulaal teaches: The processor of claim 1, further comprising: the one or more circuits to process a first input associated with the one or more workloads (¶ 4, “The method includes obtaining, by the recommendation engine, a workload; generating workload features associated with the workload”) and a second input associated with available computing resources of a data center(¶ 4, “determining compliant hardware configurations of the data cluster using the workload features”) to predict the one or more computing resources to perform the one or more workloads from the available computing resources of the data center (¶ 4, “generating performance predictions associated with the compliant hardware configurations using the workload features”).
Regarding claim 7, Subramanian teaches: The processor of claim 5, wherein the first input comprises information on at least two of data to be processed, a type of operation to perform on the data, a target time to complete the operation, or a type of model to be used for the operation (col. 4:65-67 and col. 5:1-3, “the AI model can consider any of measured telemetry data, performance indicators, boundedness, utilized compute resources, or evaluation or monitoring of the application performance (including the application's own evaluation of its performance)”).
Regarding claim 8, Abdulaal teaches: The processor of claim 5, wherein the second input comprises at least one of compute resources availability, network resources availability, storage resources availability, or memory resources availability (¶ 65, “The component(s) characteristics (218) may specify performance information of the associated component. The performance information may include, for example, clock speed, memory type, memory size, utilization, number of CPU cores, cache types, utilization, memory clock speed, maximum power limit, and other and/or additional performance information associated with the components without departing from the invention”).
Regarding claim 9, Subramanian teaches: The processor of claim 1, wherein the one or more circuits are further to generate a job profile comprising one or more workload details of the one or more workloads (col. 5:38-43, “Workload table 224 keeps track of previously performed workloads and their characteristics such as one or more of: boundedness (e.g., utilization of one or more of: processor, memory, network, storage, or cache), applied resource allocations, telemetry data, or workload performance characteristic(s)”).
Regarding claim 10, Subramanian teaches: The processor of claim 1, further comprising: the one or more circuits to monitor execution of the one or more workloads on the one or more computing resources (col. 10:45-47, “While the workload is running and potentially after its completion, telemetry data is collected on one or a variety of nodes or monitoring point or points”), determine whether the one or more computing resources were optimal for execution of the one or more workloads (claim 1, “determining at least one performance indicator associated with performance of the workload based at least, in part, on the received telemetry data”), and update training of the one or more neural networks (col. 2:17-20, “The AI model does not have to be trained as it will be configured to continuously learn on-the-go using, for example, reinforcement learning that develops based on rewards or penalties from resources it has suggested for use”).
Regarding claim 11, Da Silva teaches: The processor of claim 1, further comprising: the one or more circuits to output the confidence level (¶ 32, “the first machine learning model may produce the confidence value”).
Claims 12-14, 16, 18-22 recite commensurate subject matter as claims 1-3, 5, and 7-11. Therefore, they are rejected for the same reasons.
Regarding claim 23, Abdulaal teaches: The data center of claim 12, wherein the one or more computing resources are components of a plurality of nodes of the data center that have heterogeneous hardware (¶ 65, “The component(s) characteristics (218) may specify performance information of the associated component. The performance information may include, for example, clock speed, memory type, memory size, utilization, number of CPU cores, cache types, utilization, memory clock speed, maximum power limit, and other and/or additional performance information associated with the components without departing from the invention”), and wherein a first node of the plurality of nodes varies from a second node of the plurality of nodes in terms of at least one of performance parameters of compute resources, performance parameters of memory resources, performance parameters of network resources, or performance parameters of storage resources (claim 6, “the hardware specification information specifies components and component characteristics associated with the hardware of the data nodes of the data cluster” and ¶ 65, “The performance information may include, for example, clock speed, memory type, memory size, utilization, number of CPU cores, cache types, utilization, memory clock speed, maximum power limit, and other and/or additional performance information associated with the components without departing from the invention”).
Claims 24-26, 28-30 recite commensurate subject matter as claims 1, 2, 5-8, and 10. Therefore, they are rejected for the same reasons.
Claim(s) 6, 17, and 27 is/are rejected under 35 U.S.C. 103 as being unpatentable over Subramanian, Abdulaal, and Da Silva, as applied above, and further in view of Tee (US 2022/0188677).
Regarding claim 6, Subramanian, Abdulaal, and Da Silva do not teach; however, Tee teaches: the first input comprises an industry associated with the one or more workloads (¶ 91, “S250 may function to set or identify a failure region based on subscriber preferences, subscriber profiles, and/or any suitable attribute associated with a subscriber to the tuning service including industry-specific (e.g., global or macro attributes) attributes”) and a type of operation associated with the one or more workloads (¶ 93, “setting a failure region may first include normalizing dimensions of hyperparameters of a mode”), wherein the type of operation comprises at least one of a transfer learning operation, a hyper-parameter tuning operation (¶ 93, “prior to setting a failure region, S250 may function to normalize the dimensions of a first hyperparameter in a set of hyperparameters to the dimensions of the second hyperparameter of a set”), a data preparation operation, a data serving pipeline operation, or an experiment visualization operation.
It would have been obvious to a person having ordinary skill in the art, at the effective filing date of the invention, to have applied the known technique of the first input comprises an industry associated with the one or more workloads and a type of operation associated with the one or more workloads, wherein the type of operation comprises at least one of a transfer learning operation, a hyper-parameter tuning operation, a data preparation operation, a data serving pipeline operation, or an experiment visualization operation, as taught by Tee, in the same way to predicted one or more computing resources, as taught by Subramanian, Abdulaal, and Da Silva. Both inventions are in the field of predicting resource usage of machine learning/neural networks, and combining them would have predictably resulted in “a tuning service that tunes hyperparameters of computer-based models in the computer optimization and machine learning fields,” as indicated by Tee (¶ 2).
Claims 17 and 27 recite commensurate subject matter as claim 6. Therefore, they are rejected for the same reasons.
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 nonprovisional extension fee (37 CFR 1.17(a)) 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JACOB D DASCOMB whose telephone number is (571)272-9993. The examiner can normally be reached M-F 9:00-5:00.
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, 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.
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.
/JACOB D DASCOMB/ Primary Examiner, Art Unit 2198
Prosecution Insights