DETAILED ACTION
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 on 04/08/2026 have been fully considered but they are not persuasive.
In reference to the 35 U.S.C. 103 rejections of the previous office action.
APPLICANT REMARKS:
“Applicant disagrees with aspects of the rejection, nevertheless solely in the interest of expediting prosecution, independent claims 1, 9, and 16 have been amended to recite, in part:
"determining, by the computing system, an acceptance rate of state changes of the variables during the stochastic process, the acceptance rate corresponding to how many respective states of the respective variables are changed during the stochastic process over the trials within a monitoring interval".
The cited references fail to teach all limitations of the independent claims as amended. The Office Action maintains that it would have been obvious to substitute De Gloria's "acceptance probability" for Sridharan's "qc(P)" metric to control parallelism, asserting that both metrics provide "insight into the performance of the system" and are thus suitable substitutes. However, this rationale does not address the technical distinctions introduced by the amended claims, nor is it supported by the cited references.” (pg. 8-10)
EXAMINER RESPONSE:
Applicant's arguments filed have been fully considered but they are not persuasive.
First, De Gloria on page 165, teaches “The probability to accept a transition (k, ki) with cost ∆Ckki is given by:”
PNG
media_image1.png
64
256
media_image1.png
Greyscale
Further, based on the claims amendments, examiner would like to point to Sridharan on page 37, which teaches “Passively monitor the execution environment for changes, as the parallelism manager employs Popt parallelism for the application” and page 47 which states “To do this, whenever the AE detects a change in the execution environment, it sets the instantaneous parallelism of the application to one, i.e., P = 1, for a pre-defined time period in all our experiments, monitors its execution and establishes a baseline performance, Perf(1).” And “Further, monitoring at this granularity makes it difficult to capture the Operating System context switching overheads, which typically happens, on an average, at 50ms granularity. With 100ms and bigger time intervals, we observed that the impact of monitoring on the application performance is negligible. Therefore, the AE uses 100ms as the monitoring interval for measuring instantaneous performance of the application.” Further on page 50 Sridharan states “the AE then enters into a passive monitoring mode where it periodically monitors the processor, memory, I/O and energy usage characteristics of the application for changes. To do so, it maintains a usage vector that comprises a variety of metrics, including Instruction fetch rate, last level cache miss rate, CPU utilization, memory bandwidth utilization, instructions retired per second, number of page faults, processor and system energy consumption, among others.”
These lines teach that execution environments are being monitored for executions and changes for a pre-defined time period to establish a baseline performance (acceptance rate). This clearly teaches that the acceptance rate is being established over multiple executions (trials) during a pre-defined monitored interval.
Therefore, the 103 rejections are being maintained.
Status of Claims
The present application is being examined under the amended claims filed on 04/08/2026.
Claims 1, 2, 4-21 are rejected.
Claims 1, 2, 4-21 are pending.
Claims 1, 9 and 17 have been amended.
Claim 3 is cancelled.
Prior Art References
The short names that are used to identify the references of prior art in the analysis that follows are:
De Gloria, A., Faraboschi, P. and Olivieri, M., 1993. Clustered Boltzmann Machines: Massively parallel architectures for constrained optimization problems. Parallel Computing, 19(2), pp.163-175. (Hereafter, “De Gloria”).
Sridharan, S., Gupta, G. and Sohi, G.S., 2014, June. Adaptive, efficient, parallel execution of parallel programs. In Proceedings of the 35th ACM SIGPLAN Conference on Programming Language Design and Implementation (pp. 169-180). (Hereafter, “Sridharan”).
Dabiri, K., 2019. Replica exchange mcmc engine for combinatorial optimization problems. University of Toronto (Canada). (Hereafter, “Dabiri”).
Jedermann, R., Paul, H. and Lang, W., 2017, February. In-network processing by the example of maxima estimation in spatial fields. In SCC 2017; 11th International ITG Conference on Systems, Communications and Coding (pp. 1-6). VDE. (Hereafter, “Jedermann”).
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.
Claims 1, 2, 4-21 are rejected under 35 U.S.C. 103.
Claims 1, 2 and 9-11 are rejected under 35 U.S.C. 103 as being unpatentable over De Gloria in view of Sridharan.
Claims 4-7, 12-15, and 17-21 are rejected under 35 U.S.C. 103 as being unpatentable over De Gloria in view of Sridharan in further view of Dabiri.
Claims 8 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over De Gloria in view of Sridharan in further view of Dabiri in further view of Jedermann.
In reference to claim 1.
De Gloria teaches:
“1. (Currently Amended) A method comprising: obtaining, by a computing system, a state matrix of a system that represents an optimization problem, the state matrix including variables that each represent a characteristic related to the optimization problem;”
(De Gloria 165, “A Boltzmann Machine consists of a number of logical neurons, N, that can be represented by an undirected graph G = (V, E), where {v0, … v(n-1)} denotes the set of vertices corresponding to the logical neurons […] A number is associated with each vertex vi, denoting the state of the corresponding ith logical neuron, i.e. 0 or 1, corresponding to 'off' or 'on', respectively. The state of vertex vi in configuration k is denoted by si(k) and a configuration k of the Boltzmann Machine is uniquely defined by the states of all the individual vertices.”, The “state matrix” is taught by the vertex vector “{v0, … v(n-1)}”)
(De Gloria 166, “We can formalize a combinatorial optimization problem in terms of a Boltzmann Machine by defining a function F : S -> R that assigns each solution S a real value. So, the problem is to find a feasible solution for which F is maximal.”, The “optimization problem” is taught by the formalized “combinatorial optimization problem”.)
“obtaining, by the computing system, weights that correspond to the variables, each respective weight relating to one or more relationships between a respective variable and one or more other variables of the state matrix;”
(De Gloria 165, “An edge (vi, vj) is defined to be activated in a given configuration k if connection weight wij is associated with each edge (vi, vj), determining the strength of the connection between the vertices vi and vj.”, The “weights” are taught by each entry of the weight matrix “wij”.)
“obtaining, by the computing system, a local field matrix that includes local field values, the local field values indicating interactions between the variables as influenced by the respective weights of the respective variables;”
(De Gloria Equation 4, The “local field” is taught by “hi”)
PNG
media_image2.png
66
213
media_image2.png
Greyscale
“performing, by the computing system based on the weights and the local field values, a stochastic process with respect to changing a respective state of one or more of the variables,”
(De Gloria 165, “From a given configuration k, a neighboring configuration ki can be obtained by changing the state of the vertex i, so that:”)
PNG
media_image3.png
111
259
media_image3.png
Greyscale
“the stochastic process including performing trials with respect to one or more of the variables, in which a respective trial determines whether to change a respective state of a respective variable; determining, by the computing system, an acceptance rate of state changes of the variables during the stochastic process, the acceptance rate corresponding to how many respective states of the respective variables are changed during the stochastic process;”
(De Gloria 165, “The probability to accept a transition (k, ki) with cost ∆Ckki is given by:”)
PNG
media_image1.png
64
256
media_image1.png
Greyscale
“and obtaining, by the computing system as a solution to the optimization problem, an updated state matrix that includes one or more respective state changes to one or more of the variables of the state matrix as accepted during the trials,”
(De Gloria, “The whole process can be thought as a sequence of Markov chains and it can be demonstrated that for long enough chains the system converges to a unique equilibrium state.”)
Sridharan teaches:
“… respective variables are changed during the stochastic process over the trials with a monitoring interval”
(Sridharan 37, “Passively monitor the execution environment for changes, as the parallelism manager employs Popt parallelism for the application.”)
(Sridharan 47, “To do this, whenever the AE detects a change in the execution environment, it sets the instantaneous parallelism of the application to one, i.e., P = 1, for a pre-defined time period in all our experiments, monitors its execution and establishes a baseline performance, Perf(1).” And “Further, monitoring at this granularity makes it difficult to capture the Operating System context switching overheads, which typically happens, on an average, at 50ms granularity. With 100ms and bigger time intervals, we observed that the impact of monitoring on the application performance is negligible. Therefore, the AE uses 100ms as the monitoring interval for measuring instantaneous performance of the application”)
(Sridharan 50 “the AE then enters into a passive monitoring mode where it periodically monitors the processor, memory, I/O and energy usage characteristics of the application for changes. To do so, it maintains a usage vector that comprises a variety of metrics, including Instruction fetch rate, last level cache miss rate, CPU utilization, memory bandwidth utilization, instructions retired per second, number of page faults, processor and system energy consumption, among others.”)
“adjusting a degree of parallelism used by the computing system with respect to performing the trials based on the determined acceptance rate;”
(Sridharan 38, “The role of the Parallelism Manager (PM) is to automatically control the execution of parallel computations in the application in order to match the degree of parallelism determined by the AE.”, As described in Sridharan, parallelism is adjusted based on the qc(P) metric, however when taken in combination with De Gloria, the probability to accept a transition can be substituted in for this metric. Refer to the motivation to combine De Gloria and Sridharan.)
(Sridharan 44, “This causes qc(P) to reduce, providing additional speedup from P = 2 to P = 8. Even when qc(P) remains relatively constant, from P = 8 to P = 16, Barneshut continues to speed up. Hence stable or decreasing qc(P) indicates that additional parallelism is likely to improve performance.”)
(Sridharan Figure 4.2(b))
PNG
media_image4.png
262
342
media_image4.png
Greyscale
“the adjusting of the degree of parallelism causing the computing system to one or more of: obtain the solution faster than if the adjusting of the degree of parallelism were not performed; use fewer processing resources than if the adjusting of the degree of parallelism were not performed; or improve its ability to obtain the solution than if the adjusting of the degree of parallelism were not performed.”
(Sridharan 45, “Maximize application throughput”)
(Sridharan 46, “Minimize resource consumption cost”)
Motivation to combine De Gloria, Sridharan.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine De Gloria, Sridharan.
De Gloria discloses a parallel architecture of clustered Boltzmann machines for use in solving optimization problems.
Sridharan discloses a survey of adaptive execution of parallel programs.
One would be motivated to combine these references because, while Sridharan uses the metric “qc(P)” to scale parallelism, the “acceptance probability” of De Gloria would be a suitable substitute for determining the degree of parallelism. This substitution is obvious because Sridharan uses qc(P) to “provide[[s]] insights into how the execution of the application’s parallel region is influenced by the current operating conditions” (Sridharan 43). Similarly, in De Gloria, the acceptance probability provides insight into the performance of the system in solving the optimization problem (De Gloria 165, “The annealing process begins with c = c0 and several consecutive trials of transitions (k, ki) are randomly tossed according to the probability Bkki. As c approaches 0 the accepted transitions are less and less frequent and finally the Boltzmann Machine stabilizes in a steady configuration”. In both sources, the respective metric provides a direct read on the influence of parallelism on the system.
Further, MPEP § 2143(I) EXAMPLES OF RATIONALES sets forth the Supreme Court rationales for obviousness, including:
(B) Simple substitution of one known element for another to obtain predictable results;
(E) "Obvious to try" – choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success;
(G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention.
In reference to claim 2.
Sridharan teaches:
“2. (Previously Presented) The method of claim 1, wherein adjusting the degree of parallelism includes one of: increasing the degree of parallelism as the acceptance rate decreases; and decreasing the degree of parallelism as the acceptance rate increases.”
(Sridharan 37, “(1) Establish the relationship between the application’s instantaneous degree of parallelism, instantaneous performance and the side effects using a scalability model. (2) Using this information, determine the optimum degree of parallelism, Popt, for a given performance objective. (3) Passively monitor the execution environment for changes, as the parallelism manager employs Popt parallelism for the application. (4) Go to step 1 if the operating conditions change”; Sridharan 44, “This causes qc(P) to reduce, providing additional speedup from P = 2 to P = 8. Even when qc(P) remains relatively constant, from P = 8 to P = 16, Barneshut continues to speed up. Hence stable or decreasing qc(P) indicates that additional parallelism is likely to improve performance.”, The AE determines optimal parallelism for each performance objective using steps 1-4.)
(Sridharan Figure 4.2(b))
In reference to claim 4.
Dabiri teaches:
“4. (Original) The method of claim 1, further comprising adjusting, based on the determined acceptance rate, an offset applied to one or more of local field values of the local field matrix while performing the trials.”
(Dabiri 19, “If none of the neurons are accepted as a candidate for an update, offset parameter Eoff is incremented by a fixed value, update block is bypassed, next MCMC iteration starts, and the acceptance probability for that MCMC iteration is calculated by: [see below]”)
PNG
media_image5.png
78
478
media_image5.png
Greyscale
(Dabiri 19, “In the next iteration, candidates are generated by calculating ∆Ei(X) = -(1 - 2xi) hi(X) – Eoff”)
Motivation to combine De Gloria, Sridharan, Dabiri.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine De Gloria, Sridharan, Dabiri.
De Gloria, Sridharan discloses a parallel architecture of clustered Boltzmann machines for use in solving optimization problems.
Dabiri discloses efficient parallel implementations of Boltzmann machines.
One would be motivated to combine these references because Dabiri’s teaching of escaping the local energy minima when solving optimization problems would be an obvious improvement to the system described by the combination of De Gloria and Sridharan(Dabiri 26, “[…] because optimization algorithms employ several techniques during the simulation to speed up the process, or to escape a non-optimal state, such as abrupt alterations in temperature, or the set scheme in where the height of energy barriers are lowered temporarily to escape local minima.”).
Further, MPEP § 2143(I) EXAMPLES OF RATIONALES sets forth the Supreme Court rationales for obviousness, including:
(A) Combining prior art elements according to known methods to yield predictable results;
(C) Use of known technique to improve similar devices (methods, or products) in the same way;
(D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results;
(E) "Obvious to try" – choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success;
In reference to claim 5.
Dabiri teaches:
“5. (Original) The method of claim 4, wherein adjusting the offset includes increasing the offset in response to the acceptance rate being zero.”
(Dabiri 19, “If none of the neurons are accepted as a candidate for an update, offset parameter Eoff is incremented by a fixed value, update block is bypassed, next MCMC iteration starts, and the acceptance probability for that MCMC iteration is calculated by: [EQUATION 2.18]”, “None of the neurons [being] accepted” teaches “the acceptance rate being zero”.)
(Dabiri 19, “The Eoff will keep incrementing while no candidates are found.”)
In reference to claim 6.
Dabiri teaches:
“6. (Original) The method of claim 4, wherein adjusting the offset includes removing the offset in response to at least one change being accepted.”
(Dabiri 19, “If none of the neurons are accepted as a candidate for an update, off set parameter Eoff is incremented by a fixed value”)
(Dabiri 19-20, “As soon as a candidate is observed, Eoff is reset to zero.”)
In reference to claim 7.
Dabiri teaches:
“7. (Original) The method of claim 4, wherein adjusting the offset includes incrementally changing a value of the offset.”
(Dabiri 19, “The Eoff will keep incrementing while no candidates are found.”)
In reference to claim 8.
Dabiri teaches:
“8. (Previously Presented) The method of claim 1, further comprising: [identifying a highest local field] value [of the local field matrix]; and using the [highest local field] value as an offset applied to one or more of local field values of the local field matrix while performing the trials.”
(Dabiri 57, “Simple incremental offset is used as demonstrated in Section 2.6, where Eoff is incremented by a constant value at every iteration that no candidate is found to be flipped. […] The increment rate must be optimized for each problem which is a big disadvantage of this scheme.”)
Jedermann teaches:
“identifying a highest local field value of the local field matrix”
(Jedermann 1, “In this paper we present an alternate algorithm, which searches the maxima of the field by spatial regression of candidate regions instead of applying a diffusion model”)
Motivation to combine De Gloria, Sridharan, Dabiri, Jedermann.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine De Gloria, Sridharan, Dabiri, Jedermann.
De Gloria, Sridharan, Dabiri discloses a parallel architecture of clustered Boltzmann machines for use in solving optimization problems.
Jedermann discloses field extrema finding techniques to the local fields outlined in De Gloria and Dabiri.
One would be motivated to combine these references because the identification of the highest local field value and use of it as an offset provides an automated way to select the offset without needing to manually optimize it as discussed in Dabiri (Dabiri 57, “Simple incremental offset is used as demonstrated in Section 2.6, where Eoff is incremented by a constant value at every iteration that no candidate is found to be flipped. Eoff is reset to zero when a neuron is accepted to be flipped. The increment rate must be optimized for each problem which is a big disadvantage of this scheme.”).
Further, MPEP § 2143(I) EXAMPLES OF RATIONALES sets forth the Supreme Court rationales for obviousness, including:
(A) Combining prior art elements according to known methods to yield predictable results;
(B) Simple substitution of one known element for another to obtain predictable results;
(D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results;
(E) "Obvious to try" – choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success;
(F) Known work in one field of endeavor may prompt variations of it for use in either the same field or a different one based on design incentives or other market forces if the variations are predictable to one of ordinary skill in the art;
In reference to claims 9-16, 20.
Claim 9 is substantially similar to claim 1; 10 and 11 to 2; 12 and 20 to 4; 13 to 5, 14 to 6, 15 to 7, and 16 to 8. These claims are thus rejected using the same art of the similar claims.
In reference to claim 17.
Claims 17 is substantially similar to claim 1 and thus is rejected using the same art. Claim 17 adds the following limitation which is rejected using Dabiri.
“a plurality of replica exchange units, each respective replica exchange unit of the plurality of replica exchange units including:”
(Dabiri 29, “The overview of the implementation is presented in Fig. 3.1. It illustrates that the design is constructed of three main segments. M replicas must perform one MCMC iteration in parallel, and at certain points during the process, the configurations of adjacent replicas are examined to be exchanged.”)
PNG
media_image6.png
427
684
media_image6.png
Greyscale
In reference to claim 18.
Dabiri teaches:
“18. (Previously Presented) The system of claim 17, wherein the operations performed by the controller further include directing performance of a replica exchange process by the plurality of replica exchange units.”
(Dabiri 29, “The overview of the implementation is presented in Fig. 3.1. It illustrates that the design is constructed of three main segments. M replicas must perform one MCMC iteration in parallel, and at certain points during the process, the configurations of adjacent replicas are examined to be exchanged.”)
In reference to claim 19.
Dabiri teaches:
“19. (Original) The system of claim 17, wherein two or more of the replica exchange units operate as a merged replica exchange unit with respect to a same replica of the state matrix.”
(Dabiri 20, “Replica exchange MCMC or parallel tempering is a population-based MCMC method where M copies of the original system (called replicas) run simultaneously, each at a different temperature.”)
In reference to claim 21.
Sridharan teaches:
“operating at different levels of parallelism during solving of the optimization problem”
(Sridharan 38, “The role of the Parallelism Manager (PM) is to automatically control the execution of parallel computations in the application in order to match the degree of parallelism determined by the AE.”)
Dabiri teaches:
“21. (Previously Presented) The method of claim 1, wherein the system includes a plurality of replica processing units (RPUs) [operating at different levels of parallelism during solving of the optimization problem].”
(Dabiri 20, “Replica exchange MCMC or parallel tempering is a population-based MCMC method where M copies of the original system (called replicas) run simultaneously, each at a different temperature.”)
(Dabiri 29, “The overview of the implementation is presented in Fig. 3.1. It illustrates that the design is constructed of three main segments. M replicas must perform one MCMC iteration in parallel, and at certain points during the process, the configurations of adjacent replicas are examined to be exchanged.”)
Conclusion
THIS ACTION IS MADE FINAL. 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 Usmaan Saeed whose telephone number is (571)272-4046. The examiner can normally be reached Monday-Friday 9:00am - 5:30pm.
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 Director, David Wiley can be reached at 571-272-3923. 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.
/USMAAN SAEED/Supervisory Patent Examiner, Art Unit 2146
Prosecution Insights