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 .
This application has been examined. Claims 1-10 are pending.
The Group and/or Art Unit location of your application in the PTO has changed. To aid in correlating any papers for this application, all further correspondence regarding this application should be directed to Group Art Unit 2175.
Specification
The title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed.
Claim Rejections - 35 USC § 112
The following is a quotation of the second paragraph of 35 U.S.C. 112:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1, 2, 4, 7, and 9 are rejected under 35 U.S.C. 112(b) as failing to particularly point out and distinctly claim the subject matter regarded as the invention:
• Claim 1: “maximizing a profit rate and an utilization rate” is indefinite. “A profit rate” lacks antecedent basis and is not defined, and it is unclear whether the claimed plan must actually achieve a maximum (a result) or merely seek to increase these quantities. Additionally, reciting a “machine learning module based on … heuristic algorithms” is internally inconsistent because a heuristic algorithm (e.g., a genetic or particle-swarm algorithm) is not necessarily a machine-learning technique; the metes and bounds of “machine learning module” are therefore unclear.
• Claim 2: the recitation “model parameters, which further comprise: target variables, decision variables, …” appears incomplete; the claim does not distinctly set forth the full set of recited parameters. Clarification/completion is required.
• Claim 4: “the constraints can calculate profit rate, utilization rate, and …” is indefinite - “can calculate” is capability language that does not positively require any step, the recitation appears incomplete, and a “constraint” does not itself “calculate.”
• Claim 7: “the real-time data comprises: electricity prices in any combination of different time zones, different time periods, GPU resources in use, …” appears incomplete; the listed members of the “real-time data” group are not distinctly set forth.
• Claim 9: the objective function (printed as “max=(a-TP+-fUR)”) is indefinite as written; the operators and the relationship of weight coefficients α and β to the total profit (TP) and utilization rate (UR) terms cannot be determined with reasonable certainty from the claim. Correction to clearly recite, e.g., max = (α·TP + β·UR), is required.
Claim Interpretation - 35 U.S.C. 112(f)
The claims are method (process) claims and recite “using a machine learning module” and “storing … in a storage module.” Because the claims do not use “means for” or “step for” language and recite acts rather than means-plus-function or step-plus-function elements, 35 U.S.C. 112(f) is not invoked. The examiner notes, however, that “machine learning module” and “storage module” are recited generically; should the claims be amended into apparatus form reciting these as functional “modules” without sufficient structure, a 112(f) interpretation (and corresponding 112(b)/112(a) analysis of the disclosed structure) may apply.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-10 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1 (statutory category): Claims 1-10 are directed to a method and therefore fall within the “process” category of 35 U.S.C. 101.
2A, Prong 1: the claims recite a judicial exception.
With respect to independent claim 1, and carried into dependent claims 2-10, the claims recite the following limitations that fall within the enumerated groupings of abstract ideas (see MPEP 2106.04(a)):
• “according to objective function and constraints … calculate a computing power scheduling plan that maximizing a profit rate and an utilization rate of the computing power platform” (a mathematical concept - optimizing an objective function subject to constraints; and a certain method of organizing human activity - a fundamental economic practice of maximizing profit and resource utilization in allocating a commercial resource).
• “receiving computing power demand information from at least one time zone and storing the demand information” (collecting and storing information - a step that, apart from generic computer recitation, can be performed in the human mind or by a person with pen and paper).
• “the objective function is: max = (α·TP + β·UR) (claim 9); target variables include total profit and GPU utilization rate (claim 3); constraints to calculate profit rate, utilization rate, and scheduling (claim 4)” (mathematical concepts - mathematical relationships, formular, and calculations).
• “the constraints include UR = ΣΣ A(i,t) / (Σ G(i) × 24) and TP = ΣΣ (P(i,t) × A(i,t)) − ΣΣ (E(i,t) active × A(i,t) + E(i,t) idle × I(i,t) + C(i,t) deprec × G(i)), with A(i,t) and I(i,t) defined by further equations (claim 10)” (mathematical concepts - mathematical formular and calculations that define the utilization rate, total profit, and resource-allocation quantities).
• “predicts future GPU resource requirements by analyzing trends of historical data and real-time data (claim 6); extracts and summarizes a behavior pattern of a user or a system according to the historical data (claim 8)” (a mental process - observation, evaluation, and judgment that can be performed in the human mind, and a mathematical/analytical concept).
Under the broadest reasonable interpretation, these limitations cover a fundamental economic practice (deciding how to allocate a commercial computing resource so as to maximize profit and utilization), a mathematical optimization, and evaluations that can be performed mentally. The mere recitation of generic computer components (below) does not take the limitations out of the mental-process, mathematical-concept, or organizing-human-activity groupings. The claims therefore recite an abstract idea.
2A, Prong 2: This judicial exception is not integrated into a practical application.
The claims recite the following additional elements: a “computing power platform,” a “machine learning module based on at least one of heuristic algorithms, machine learning algorithms, or deep learning algorithms,” a “storage module,” “GPU resources,” and the step of “allocating GPU resources … according to the computing power scheduling plan.” These additional elements, individually and in combination, do not integrate the abstract idea into a practical application:
• The computing power platform, machine learning module, storage module, and GPUs are recited at a high level of generality as generic computing components, and the machine learning/heuristic/deep-learning module is invoked merely as a tool to perform the recited optimization - i.e., “apply it” on a generic computer (MPEP 2106.05(f)).
• The claims do not recite an improvement to the functioning of a computer or to any other technology or technical field; they use generic computing to implement a scheduling/optimization scheme, and the asserted advantages (higher profit and utilization) are improvements to the abstract idea/business goal itself rather than a technical improvement to the computer or to GPU hardware (MPEP 2106.05(a) - no improvement to the computer or to a technology is claimed).
• “Receiving … and storing” the demand information is insignificant pre-solution data gathering/storage, and “allocating GPU resources … according to the … plan” is insignificant post-solution activity that merely applies the result and limits the abstract idea to a field of use (MPEP 2106.05(g) and 2106.05(h)).
Accordingly, the judicial exception is not integrated into a practical application, and the claims are directed to the abstract idea.
2B: The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Considered individually and as an ordered combination, the additional elements do not amount to significantly more than the abstract idea. The computing power platform, storage module, generic machine-learning/heuristic/deep-learning module, and GPUs perform only their well-understood, routine, and conventional functions - receiving data, storing data in memory, performing computations/optimization, and allocating computing resources according to a computed result. Receiving or transmitting data over a network and storing and retrieving information in memory are recognized as well-understood, routine, and conventional computer functions (MPEP 2106.05(d)(II)(i)). Using a generic machine-learning model to predict demand and compute an allocation is likewise the conventional use of such tools and does not supply an inventive concept. The ordered combination adds nothing beyond the sum of the parts.
Claims 1-9 are therefore patent-ineligible under 35 U.S.C. 101.
Dependent claims 2-10 do not cure the deficiency: claims 2-4, 9 and 10 further define the mathematical model, objective function, and constraint equations (further abstract); claims 5-8 further recite the same generic machine-learning module performing prediction/analysis and dynamic adjustment (further abstract and/or insignificant extra-solution activity). None recites additional elements that integrate the exception into a practical application or provide an inventive concept.
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 t which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-5, 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al., “Sustainable AIGC Workload Scheduling of Geo-Distributed Data Centers: A Multi-Agent Reinforcement Learning Approach” Apr. 2023) (“Zhang”).
In order to expedite and avoid piecemeal prosecution, the following rejection is made to the extent that the claims are understood, by considering those elements which are understood and interpreting their function in a manner which is consistent with the recited goals of the claims, and then applying the best available art.
The examiner relies on the entire teachings of Zhang reference; the applicant should carefully consider the entire teachings of the above-mentioned references to better understand the examiner’s position.
In regard to claim 1, Zhang discloses a global cross-time zone computing power scheduling method, executed by a computing power platform (as shown in Fig. 1, which is reproduced below for ease of reference and convenience, Hoddie discloses a method/system executed by a geo-distributed cloud computing platform that schedules GPU (ML training) workloads across data centers in different time zones (Abstract, ; § I-III), comprising:
step 1, receiving computing power demand information from at least one time zone and storing the demand information in a storage module (in Zhang, receives/aggregates per-region workload (GPU) demand across multiple time zones as input to the scheduler (Abstract; ; § I & II).
step 2, according to objective function and constraints, using a machine learning module based on at least one of heuristic algorithms, machine learning algorithms, or deep learning algorithms, to calculate a computing power scheduling plan (in Zhang, uses multi-agent reinforcement learning (a machine learning/deep learning module) to compute a scheduling plan that optimizes computing-capacity utilization and minimizes electricity cost subject to capacity/demand constraints (Abstract; § I-III). and
step 3, allocating GPU resources in at least one time zone according to the computing power scheduling plan (in Zhang, allocates/migrates GPU resources among the data centers (time zones) per the computed schedule (Abstract; § I-III). But Zhang does not expressly disclose an objective expressed as maximizing a “profit rate”. Regarding the express “profit rate”/“total profit” objective, Zhang already minimizes electricity cost while maximizing utilization. Because profit equals revenue minus cost, and maximizing utilization of a paid computing resource increases revenue while minimizing electricity cost reduces cost, formulating Zhang’s objective as a maximization of profit (and expressing a multi-objective trade-off as a weighted sum, max = (α·TP + β·UR)) would have been an obvious refinement that maximizing provider profit is the ordinary commercial objective of a computing-power platform, and a weighted-sum scalarization of competing objectives is a routine optimization technique that yields predictable results.
In regard to claim 2, Zhang discloses wherein the objective function and constraints comprise: model parameters, which further comprise: target variables, decision variables, parameters, and response variables (in Zhang: the RL/optimization formulation inherently includes model parameters - an objective, decision variables (allocations), and parameters - (§ I-III)).
In regard to claim 3, Zhang discloses wherein the target variables include: total profit and GPU utilization rate within a specified period of time (in Zhang, GPU utilization is an explicit objective.) Even though Zhang does not expressly disclose total profit and GPU utilization rate within a specified period of time, however by adding “total profit” as a target variable (revenue − electricity cost) is the obvious commercial objective.
In regard to claim 4, Zhang discloses wherein the constraints can calculate profit rate, utilization rate, and scheduling constraint (in Zhang, scheduling is subject to capacity/demand constraints and a cost/utilization objective). Even though Zhang does not expressly disclose profit-rate and utilization-rate constraints, however by expressing profit-rate and utilization-rate constraints is a routine formulation choice.
In regard to claim 5, Zhang discloses wherein the machine learning module based on one of the heuristic algorithms, machine learning algorithms, or deep learning algorithms; dynamically adjusts the computing power scheduling plan according to the real-time computing power demand information, the available GPU resources, and the electricity price (in Zhang, the RL scheduler dynamically adjusts allocation in response to real-time demand, available (surplus) capacity, and time-varying electricity prices (Abstract; § I-III).
In regard to claim 9, even though Zhang does not expressly disclose wherein the objective function is: max=(a-TP+-fUR), where a and P are weight coefficients, however expressing the multi-objective trade-off as a weighted sum max = (α·TP + β·UR) is a routine scalarization of Zhang’s competing cost/utilization objectives.
PNG
media_image1.png
345
946
media_image1.png
Greyscale
(in Zhang, optimizes utilization and electricity (active/idle power) cost across regions and time, migrating workloads (transfers) subject to per-region capacity and demand (Abstract; § I). Zhang does not expressly recite: the particular equations for UR, TP (including the depreciation cost term C(i,t) deprec), A(i,t), and I(i,t). However, expressing utilization as GPUs-used/GPUs-available, and profit as revenue (price × GPUs used) minus active-power, idle-power, and depreciation costs that with allocated GPUs A(i,t) = min(demand, capacity available after transfers) and idle GPUs I(i,t) = capacity − allocated - is a routine mathematical/accounting formulation of Zhang’s cost/utilization objective.
Claims 6-8 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al., “Sustainable AIGC Workload Scheduling of Geo-Distributed Data Centers: A Multi-Agent Reinforcement Learning Approach” Apr. 2023) (“Zhang”) in view of Baughman et al. (“Baughman”) (US 11,082,301).
The examiner relies on the entire teachings of Zhang and Baughman references; the applicant should carefully consider the entire teachings of the above-mentioned references to better understand the examiner’s position.
In regard to claim 6, Zhang discloses the claimed subject matter as discussed above rejection except the teaching of wherein the machine learning module based on one of the heuristic algorithms, machine learning algorithms, or deep learning algorithms predicts future GPU resource requirements by analyzing trends of historical data and real-time data and adjusts GPU resource allocation in advance. In the same field of endeavor, Baughman discloses forecasting future requirements from historical + real-time data and adjusting allocation in advance (as shown in Fig. 5, 7, which is reproduced below for ease of reference and convenience, Baughman discloses a model forecasting future demand from a historical timeline and current data, provisioning resources ahead of time. See col. 2:55-3:36; col. 12:10-67; col. 13:24-20:35; (claim 1)).
PNG
media_image2.png
812
702
media_image2.png
Greyscale
PNG
media_image3.png
619
553
media_image3.png
Greyscale
Zhang and Baughman are analogous art directed to allocating computing resources based on aggregated/predicted demand. Zhang teaches the cross-time-zone, geo-distributed GPU scheduling and the electricity-cost/utilization optimization; Baughman teaches predicting future resource demand from historical and current (real-time) data and provisioning resources in advance. It would have been obvious to one of ordinary skill before the effective filing date to incorporate Baughman’s predictive, ahead-of-time provisioning into Zhang’s cross-time-zone scheduler, because doing so combines known elements to yield the predictable benefit of reducing scheduling delays from sudden demand and improving utilization (KSR rationale (A); MPEP 2143), and applies a known technique (ML demand forecasting and pre-provisioning) to a known system (a geo-distributed GPU scheduler) ready for improvement (KSR rationale (C)/(D)). A reasonable expectation of success existed because both references apply machine-learning models to resource-demand data to drive allocation.
In regard to claim 7, Zhang discloses wherein the real-time data comprises: electricity prices in any combination of different time zones, different time periods, GPU resources in use, and idling GPU resources (in Zhang: real-time inputs include electricity prices across regions/time zones and time periods, and GPU usage/surplus (idle) capacity (Abstract; § I-III).
In regard to claim 8, Baughman discloses wherein the machine learning module based on one of the heuristic algorithms, machine learning algorithms, or deep learning algorithms extracts and summarizes a behavior pattern of a user or a system according to the historical data to adjust the GPU resource allocation in advance (in Baughman, deriving predicted bursts/spikes from a historical timeline of a previous event (a behavior pattern) to provision ahead of time. See col. 2:55-3:36; col. 12:10-67; (claim 1)). It would have been obvious to one of ordinary skill before the effective filing date to incorporate Baughman’s predictive, ahead-of-time provisioning into Zhang’s cross-time-zone scheduler, because doing so combines known elements to yield the predictable benefit of reducing scheduling delays from sudden demand and improving utilization (KSR rationale (A); MPEP 2143), and applies a known technique (ML demand forecasting and pre-provisioning) to a known system (a geo-distributed GPU scheduler) ready for improvement (KSR rationale (C)/(D)). A reasonable expectation of success existed because both references apply machine-learning models to resource-demand data to drive allocation.
Examiner's note:
Examiner has cited particular paragraphs, columns and line numbers in the references applied to the claims above for the convenience of the Applicant. Although the specified citations are representative of the teachings of the art and are applied to specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested from the Applicant in preparing responses, to fully consider the references in entirety as potentially teaching all or part of the claimed invention, as well as the context of the passages as taught by the prior art or disclosed by the Examiner.
Conclusion
All claims are rejected.
The prior arts made of record and not relied upon are considered pertinent to applicant's disclosure.
Zhang/Niyato et al. (DRL geo-distributed DC) dislose a game-theoretic deep reinforcement learning minimizing energy cost/carbon for AI workloads across geographically distributed data centers (time-of-use pricing).
ACM Computing Surveys, “Deep Learning Workload Scheduling in GPU Datacenters: A Survey” (2024, DOI 10.1145/3638757) disclose a Surveys GPU-datacenter schedulers that predict future utilization from history (e.g., Helios CES) and use heuristic cost minimization (e.g., ANDREAS MINLP) as evidences conventionality.
Rong, Qin & An, “Competitive Cloud Pricing for Long-Term Revenue Maximization,” JCST 34(3) (2019) disclose a profit/revenue-maximization pricing/allocation for cloud providers as evidences the conventional profit objective
Hu et al., “Efficient Resources Provisioning Based on Load Forecasting in Cloud” (2014) disclose a SVR-based multi-step load forecasting + provisioning to maximize utilization while meeting SLAs.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to examiner Raymond Phan, whose telephone number is (571) 272-3630. The examiner can normally be reached on Monday-Friday from 6:30AM- 3:00PM. The Group Fax No. (571) 273-8300.
Communications via Internet e-mail regarding this application, other than those under 35 U.S.C. 132 or which otherwise require a signature, may be used by the applicant and should be addressed to [raymond.phan@uspto.gov].
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, Andrew Jung can be reached at (571) 270-3779. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
All Internet e-mail communications will be made of record in the application file. PTO employees do not engage in Internet communications where there exists a possibility that sensitive information could be identified or exchanged unless the record includes a properly signed express waiver of the confidentiality requirements of 35 U.S.C. 122. This is more clearly set forth in the Interim Internet Usage Policy published in the Official Gazette of the Patent and Trademark on February 25, 1997 at 1195 OG 89.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see hop://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
Any inquiry of a general nature or relating to the status of this application should be directed to the TC 2100 central telephone number is (571) 272-2100.
/RAYMOND N PHAN/
Primary Examiner, Art Unit 2175
Prosecution Insights