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 .
Claims 1-20 are presented for the examination.
§ 101 2. 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 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
As to Claims 1, 2, 3, 11, 13, 14, 15, 20 have been rejected under 35 USC 101 for abstract idea without significantly more. Under Step 2A, Prong 1, the “ estimating an inference unit (IU) processing load to be applied to the processing unit ”, “ estimating an estimated prompt load and an estimated generation load of an AI inferencing workload” recite a mental process since “ estimate” are functions that can be reasonably performed in the human mind with the aid of pen and paper through observation, evaluation, judgment, opinion.
Under Prong 2, the additional element “ receiving an estimated prompt load and an estimated generation load of an AI inferencing workload to be fulfilled by a processing unit of a computing node of the distributed AI inferencing platform, allocating fractional processing capacity of the processing unit for fulfilling the AI inferencing workload based at least in part on the IU processing load” are recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer component, or merely a generic computer or generic computer components to perform the judicial exception, Accordingly, the additional elements do not integrate the recited judicial exception into a practical application, and the claim is therefore directed to the judicial exception. See MPEP 2106.05(f).
Under Step 2B, the additional elements “ receiving an estimated prompt load and an estimated generation load of an AI inferencing workload to be fulfilled by a processing unit of a computing node of the distributed AI inferencing platform” - this generally have been a mental process although the processing unit could be a generic computer component if the spec describes it as actual computer hardware.
“allocating fractional processing capacity of the processing unit for fulfilling the AI inferencing workload based at least in part on the IU processing load” - this is mere instructions to apply the mental process under mpep 2106.05(f), amounts to merely generally linking the use of the judicial exception to a particular technological environment or field or use, and is merely applying the judicial exception, therefore, does not amount to significantly more, hence, cannot provide an inventive concept.
4. The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application. See MPEP 2106.05(d). Thus, the claim is not patent eligible.
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.
Claim(s) 1, 13, 20 are rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) and further in view of Zeng( US 20220201555 A1).
As to claim 1, LEE teaches A method for artificial intelligence (AI) inferencing workload allocation, the method comprising( provides an artificial intelligence chip including a compute node which is capable of computing a computational model which implements various artificial intelligence network algorithms, para[0015]/ The AI accelerator 140 refers to a hardware component which sequentially or simultaneously computes at least one computational model allocated to the compute node 100. The AI accelerator 140 may include at least one processor and a memory shared by the at least one processor , para[0057]/ he plurality of interconnected compute nodes 100 included in the cluster may perform a task together. For example, the task may include learning of input data based on a computational model or inference of a result value, para[0079);
at a computing device of a distributed AI inferencing platform, receiving an estimated prompt load and an estimated generation load of an AI inferencing workload to be fulfilled by a processing unit of a computing node of the distributed AI inferencing platform(each compute node N.sub.a of the cluster determines a contributable computational load of each compute node N.sub.a for the expected computational load determined in step 420, para[0138], ln 1-5/ provides an artificial intelligence chip including a compute node which is capable of computing a computational model which implements various artificial intelligence network algorithms, para[0015]/ The computational model processing method may include a step 410 of receiving input data by a home node N.sub.H, which is one of a plurality of compute nodes, a step S420 of determining, by the home node N.sub.H, an expected computational load for processing a computational model for the input data, a step S430 of sharing (or transmitting), by the compute nodes N.sub.a, a contributable computational load in response to a request of the home node N.sub.H, and a step S440 of selecting, by the home node N.sub.H, a master node N.sub.M to distribute the expected computational load among the plurality of compute nodes based on the received contributable computational load, para[0130]/ Additionally, the computational model processing method may further include a step 530 of performing computation for learning/inference based on the computational model of each of the least one compute node to which the expected computational load was distributed in step 520, para[0149]/ Fig.4/5)
based at least in part on the estimated prompt load and the estimated generation load, estimating an inference unit (IU) processing load to be applied to the processing unit while fulfilling the AI inferencing workload( determine a contributable computational load of the first compute node N.sub.1 within one minute, which is the computational constraint, as 70, in consideration of the current computational load, a waiting computational load, a computing ability, and available resources. In step 440, the home node N.sub.H may determine a master node N.sub.M to distribute the expected computational load among the plurality of compute nodes based on the contributable computational load shared in step 430. In one example, the home node N.sub.H may determine a compute node having the largest contributable computational load among the plurality of compute nodes as the master node N.sub.M, para[0140]/ Fig. 4);
Sivathanu teaches allocating fractional processing capacity of the processing unit for fulfilling the AI inferencing workload based at least in part on the IU processing load( The scale 240 routine is configured to scale up and/or scale down the quantity, quality, and/or type of infrastructure resources being used to execute an AI workload. For instance, if additional infrastructure resources are available, an AI workload may be scaled up to make use of those additional infrastructure resources. Alternatively, if a new AI workload requires some infrastructure resources in use executing a current AI workload, the current AI workload may be scaled down to free up some resources for the new AI workload (e.g., the new AI workload may be associated with a higher priority or tier than the current AI workload), para[0066]/ monitor performance of the set of AI workloads being executed using the infrastructure resources; based on monitoring performance of the set of AI workloads, determine, for each AI workload, a dynamic preemption score indicative of a relative likelihood that the AI workload will be preempted, wherein the dynamic preemption score is based on at least one performance threshold requirement of the priority tier with which the AI workload is associated; based on determining that an AI workload will be preempted, identify an AI workload for preemption based on the dynamic preemption score of the AI workload, wherein the identified AI workload has the dynamic preemption score indicative of a highest likelihood of preemption of all dynamic preemption scores of the set of AI workloads; and preempt the identified AI workload. [0174] wherein the at least one memory and the computer program code is configured to, with the at least one processor, further cause the at least one processor to: identify a set of spare infrastructure resources for distribution to the set of AI workloads; identify a first subset of AI workloads associated with a high priority tier of the set of AI workloads; distribute a first subset of infrastructure resources of the set of spare infrastructure resources to the first subset of AI workloads based on a scale-up priority requirement of the high priority tier; identify a second subset of AI workloads associated with a standard priority tier of the set of AI workloads; [0175] distribute a second subset of infrastructure resources of the set of spare infrastructure resources to the second subset of AI workloads based on a scale-up priority requirement of the standard priority tier, wherein the second subset of infrastructure resources is smaller than the first subset of infrastructure resources , para[0173] to para[0175]).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because this manages the scheduling of AI workloads on infrastructure in a fair and efficient manner presents substantial challenges to data scientists trying to accelerate the algorithmic innovations of AI.
Zeng teaches allocating fractional processing capacity of the processing unit for fulfilling the AI inferencing workload( where θ(t) is defined from two factors: the “computing workload” and the “network capacity.” The computing workload on the edge server is timely consumed by CPU utilization (C(t)) and GPU utilization (G(t)). The first term, [(1−G(t))*f.sub.1+(1−C(t))*f.sub.2]*f.sub.3 determines the computing workload. Network capacity on the edge server refers to timely bandwidth utilization. The second term, (B(t)). (1−B(t))*f.sub.4 determines the Network capacity. When θ(t)<0.5, the inference workload is executed at the edge server 150. When θ(t)≥0.5, the inference workload is executed at the uCPE gateway 140, para[0033]).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because this determines an allocation strategy based on the collected utilization rate data from the gateway and the edge server.
As to claims 13, 20, they are rejected for the same reason as to claim 1 above. In additional, Lee teaches processor( the processor, para[0012], ln 1-2).
Claim(s) 2, 3, 4, 5, 14, 15, 16 are rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) in view of Zeng( US 20220201555 A1) and further in view of Catalano( US 20210334753 A1).
As to claim 2, Catalano teaches wherein the estimated prompt load is estimated based at least in part on an input prompt token quantity of a sample workload associated with a same user as the AI inferencing workload( FIG. 15 illustrates a process for completing a task utilizing the system described hereinabove. In step 1500 the system receives a project scope containing task details from a user or other input source. The artificial intelligence engine 800 determines a suggested set of steps to complete the task in step 1505 and transmits the suggested steps to one or more users in step 1510. In step 1515, the artificial intelligence engine 800 may also determine a timeline for the task completion for one or more users of the system by consulting the scheduling module 810 to determine availability, workload, and historical data related to each user, In step 1520, the project management subsystem 1200 determines if the project is proceeding as anticipated or if indications, communications, or other information has been received which may indicate that the project status and estimated completion differs from an expected state. If it is determined that the project is not proceeding according to the initial prediction, one or more communications may be sent to one or more users alerting them of the change. para[0062], ln 1-24).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because this facilitates workflow optimization throughout an organization and auxiliary parties working on an event, project, or task.
As to claim 3, Catalano teaches the estimated generation load is estimated based at least in part on a quantity of output tokens generated for the sample workload( para[0052], ln 3-27/ para[0052], ln 1-32/ para[0053]) for the same reason as to claim 2 above.
As to claim 4, Catalano teaches the sample workload includes a previous input prompt provided to the distributed AI inferencing platform in a prior inferencing request associated with the same user as the AI inferencing workload( para[0051]) for the same reason as to claim 2 above.
As to claim 5, Catalono teaches estimate an amount of time used to process each input token of a sample input prompt( para[0062], ln 1-24) for the same reason as to claim 2 above. In additional, Brown teaches a current-pass token quantity and an input prompt index are input to a statistical model to estimate an amount of time used to process each input token of a sample input prompt, and wherein the IU processing load is proportional to the amount of time used to process each input token of the sample input prompt( With reference again to FIG. 7, the task optimization analysis component 702 can employ various machine learning and/or statistical task optimization models/algorithms to facilitate determinizing how to schedule tasks and assign resources to the tasks based on the various parameters/variables described above (e.g., associated with the tasks, the patients associated with the tasks, the healthcare workers that perform the tasks, the non-human resources needed for the tasks, and in some implementations, the forecasted task demand and resource availability). These task optimization models can be configured to determine an optimal task scheduling and resource assignment scheme based on various optimization criteria, including but not limited to: meeting fixed constraints associated with the tasks (e.g., regarding timing, location, resource requirements, ordering constraints, grouping constraints, priority constraints, etc), minimizing delays between performance of the healthcare tasks, ensuring all healthcare tasks are delivered in accordance with defined quality and regulatory requirements, maximizing utilization of available resources, minimizing losses, maximizing revenue, meeting patient preferences with respect to when, where and who performs healthcare tasks, and meeting healthcare worker preferences with respect to preferred tasks, timing and location for perfuming the tasks. In various embodiments, the task optimization analysis component 702 can be configured to employ one or more task optimization models (functions, algorithms, etc.) that determines the optimal task scheduling and resource assignment scheme based on a combination of two or more of these optimization criteria. So for example, assume the healthcare system has a pool of different workers with different capabilities, skill levels, salaries, locations, time availability, schedules, worker preferences, etc, para[0131], ln 1-40/ The task optimization analysis component 702 can be configured to evaluate the indexed task data 112, the resources availability data 116 and other relevant parameters provided by the one or more healthcare information sources/systems to determine task scheduling and resource assignment information 126 regarding how to schedule tasks and assign resources to the tasks for an integrated healthcare system in real-time in a manner that optimizes performance of the integrated healthcare system as a whole by coordinating, para[0127]/ Fig. 7 ).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because this facilitates determining the availability information (e.g., using one or more models developed/trained based on historical activity information for the healthcare workers and historical performance of the healthcare tasks under various operating conditions/contexts of the healthcare system).
As to claims 14, 15, 16, they are rejected for the same reasons as to claims 2, 3, 4 above.
Claim(s) 6, 19 are rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) in view of Zeng( US 20220201555 A1) and further in view of Boyapalle(US 20230134096 A1).
As to claim 6, Zeng teaches the fractional processing capacity of the processing unit is allocated as a fractional capacity allocation, para[0033] for the same reason as to claim 1 above and Boyapalle teaches the fractional processing capacity of the processing unit is allocated as a fractional capacity allocation, and wherein a size of the fractional capacity allocation relative to a total processing capacity of the processing unit is proportional to the IU processing load( Moreover, within each different type of cloud service, a cloud servicer provider may also offer different performance tiers. For example, a high-performance tier (e.g., tier 1) cloud service may allocate a first amount of cloud compute power to execute a workload, typically at a first cost. A mid-performance tier (e.g., tier 2) cloud service may allocate a second amount of cloud compute power smaller than the first amount to execute the workload, typically at a second cost smaller than the first cost. A low-performance tier (e.g., tier 3) cloud service may allocate a third amount of cloud compute power smaller than the second amount to execute the workload, typically at a third cost smaller than the second cost, para[0127]).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because this allows different type of cloud service may offer a different level of performance when executing a given workload.
As to claim 19, it is rejected for the same reason as to claim 6 above.
Claim(s) 7, 17 are rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) in view of Zeng( US 20220201555 A1) and further in view of Xu( US 20220300323 A1).
As to claim 7, Xu teaches the fractional processing capacity is allocated as a first fractional capacity allocation, and wherein a second fractional capacity allocation of the processing unit is allocated for concurrently fulfilling a second AI inferencing workload associated with a different user( distributed training is usually used to meet a timeliness requirement of a job with a network transmission requirement, for example, an AI training job. If a distributed training manner is used, different jobs may contend for same hardware resources. Therefore, a scheduler is required to schedule hardware resources for different jobs of a plurality of users, to allocate appropriate nodes (for example, servers) to different jobs for operating tasks included in the jobs. A current scheduler usually allocates, based on a hardware resource requirement of a task, a node having appropriate hardware resources, and ignores a requirement for network performance in the AI training job, para[0005], 1-12/ a smaller cross-node quantity of the candidate node indicates that the another job currently operated by the candidate node interacts with another node for a quite small quantity of times. By reducing the increasing amplitude for the performance score of the candidate node, para[0022], ln 12-18).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because this allows the affinity between parameter node tasks needs to be reduced (that is, the anti-affinity is improved), to enable the parameter node tasks to be placed on a plurality of different nodes to the greatest extent.
As to claim 17, it is rejected for the same reason as to claim 17 above.
Claim(s) 8 is rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) in view of Zeng( US 20220201555 A1) and further in view of Yeh( US 20200197793 A1).
As to claim 8, Yeh teaches AI inferencing workload is fulfilled with an inferencing latency and an inferencing latency variability that is isolated from other AI inferencing workloads concurrently fulfilled by the processing unit( ccordingly, certain processing nodes may be dedicated to performing certain processing tasks involved with the interactive video system, such as visual (e.g., graphics) processing, audio processing, artificial intelligence (AI) calculations, physics calculations, and/or the like, based on their respective capabilities. Further, the switched fabric network may divide these processing tasks among a number of suitable processing nodes. Moreover, the switched fabric network may facilitate remote direct memory access (RDMA). Accordingly, using RDMA, data may be transferred between physically separate processing nodes to perform a processing task with the latency of an internal computer bus, para[0018], ln 21-37).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because this allows processing tasks involved with implementing the interactive system may be distributed among one or more processing nodes, which may be specialized and/or optimized to perform particular processing tasks.
Claim(s) 9 is rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) in view of Zeng( US 20220201555 A1) and further in view of Tarkhanyan( US 20220094690 A1).
As to claim 9, Tarkhanyan teaches outputting an indication of an observed inferencing load used while fulfilling the AI inferencing workload( workload can migrate dynamically from a cloud to an edge, and from the edge to the cloud. An example of this can be shown by artificial intelligence (AI)/machine learning (ML) processing, where both inferencing and training functions can be performed at any execution locality across the edge and cloud, para[0003], ln 1-7/ The trusted group of node clusters for the workload is to include the first node cluster and the second node cluster based, at least in part, on determining that a first compute node in the first node cluster meets a first security requirement of a workload execution policy associated with the workload and on the attestation results indicating that a second compute node in the second node cluster meets a second security requirement of the workload execution policy, para[0186], ln 11-25).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because this enables trusted and coherent execution across dynamic domain boundaries for workload execution.
Claim(s) 10 is rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) in view of Zeng( US 20220201555 A1) and further in view of LEE( US 20220236860 A1).
As to claim 10, Lee teaches AI inferencing workload is fulfilled by an AI model deployed in the distributed AI inferencing platform, and wherein the method further comprises outputting a deployment-level processing load summary indicating a deployment-level processing load used for fulfilling inferencing requests provided to the AI model( he executing module is connected to the loading module and configured to when an execution command is triggered, input the loaded image data set to the loaded AI model to perform an inference calculation, and generate an inference result based on the inference calculation, and detect whether the selected recommended template has a precision. When the selected recommended template has a precision, the executing module directly load the precision, and when the selected recommended template does not have the precision, the executing module calculates the precision corresponding to the inference result, and set the calculated precision as the precision of the selected recommended template. The display module is connected to the loading module and the executing module, and configured to use the loaded dashboard to display the inference result and the precision, which is directly loaded or calculated, on the GUI, para[0005], ln 32-53).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because provides this technical solution to solve the conventional technical problem of insufficient convenience in model selection and operation.
Claim(s) 11 is rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) in view of Zeng( US 20220201555 A1) and further in view of Hermoni( US 10951485 B1)
As to claim 11, Hermoni teaches the AI inferencing workload is associated with a selected AI model, and wherein estimating the IU processing load includes estimating workload-related performance characteristics for the selected AI model( At operation 1510, the overall load of monitoring and processing the log-data associated with the current resolution level may be determined. For example, the load may be determined according to the number of parameters per second that should be monitored, and/or determined according to the number of parameters per second that should be scanned and analyzed by selected AI models. The processing load may be determined, for example, as a combination such as the number of monitored parameters per second multiplied by a number of respective AI models, col 39, ln 29-40).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because provides allows this initially detect network situation at lower data load, and higher resolution (more detailed) stages may provide a higher level of confidence to detect a potential or actual network situation.
Claim(s) 12, 18 are rejected under 35 U.S.C. 103 as being unpatentable over LEE( US 20200250509 A1) in view of SIVATHANU( US 20220318052 A1) in view of Zeng( US 20220201555 A1) and further in view of DARJI( US 20220365821 A1).
As to claim 12, Darji teaches transmitting an indication of the IU processing load for display in a graphical user interface (GUI)( fingerprint analysis of workload characteristics employs an artificial intelligence (‘AI’) and machine learning (‘ML’) model. For example, the AI/ML model may be trained on a training set of historical workload characteristics to identify typical or optimal configurations of a particular database workload type. In these implementations, the AI/ML, model is used to detect the patterns in the workload characteristics received from the storage systems 810, 812 and generate the fingerprint information 822, 828, para[0286], ln 1-10/ includes receiving 1002 a request 1003 to add a new database workload, wherein the request 1003 indicates the particular workload type. In some examples, the database deployment service 840 provides a user interface to users (i.e., database services consumers) that includes mechanisms for adding a new workload to the storage environment 818. A library of database workload types may be presented in the user interface. For example, the user interface may include drop-down menu of various database workload types. In some implementations, when a database workload type is selected, the fingerprint information 622 for that workload type is displayed in the user interface. For example, the user interface may display fingerprint information 622 that includes read performance parameters such as the average read I/O size, read IOPS, read bandwidth as well as write performance parameters such as the average write I/O size, write IOPS, and write bandwidth. The displayed fingerprint information may also include a capacity, number of volumes, data reduction factor, as well as other parameters discussed above, para[0299]/ provide this particular data compliance service, the data compliance service may be presented to a user (e.g., via a GUI) and selected by the user., ), para[0221], ln 1-5).
It would have been obvious to one of the ordinary skill in the art before the effective filling date of claimed invention was made to modify the above teaching to incorporate the above feature because provides allows the user input identifying one or more characteristics of the workload may include information that is used to make inferences that are used to generate the performance profile of the workload.
As to claim 18, it is rejected for the same reason as to claim 12 above.
Conclusion
US 20190379613 A1 teaches execute different workloads (e.g., applications) that may communicate with each other through one or more of the interconnects to achieve a goal, such as to distributive train an artificial intelligence model, to make inferences using the trained artificial intelligence model, to perform cryptographic operations, etc.
US 20220201555 A1 teaches where θ(t) is defined from two factors: the “computing workload” and the “network capacity.” The computing workload on the edge server is timely consumed by CPU utilization (C(t)) and GPU utilization (G(t)). The first term, [(1−G(t))*f.sub.1+(1−C(t))*f.sub.2]*f.sub.3 determines the computing workload. Network capacity on the edge server refers to timely bandwidth utilization. The second term, (B(t)). (1−B(t))*f.sub.4 determines the Network capacity. When θ(t)<0.5, the inference workload is executed at the edge server 150. When θ(t)≥0.5, the inference workload is executed at the uCPE gateway 140.
US 20220318052 A1 teaches system for scheduling the execution of artificial intelligence (AI) workloads, such as training and inferencing workloads, on a diverse pool of infrastructure resources distributed across a variety of regions. A global scheduler receives a set of AI workloads to be executed, wherein each AI workload of the set of AI workloads is associated with a resource ticket value indicative of a share of
US 20240086699 A1 teaches memory footprint or other performance indicators. In some aspects, the end devices may determine the current hardware capabilities based on a physical hardware configuration. Additionally, in some aspects, the current hardware capabilities may be determined based on the current workload, an estimated time to completion, or other performance metrics.
US 20210073034 A1 teaches data centers located across different geographic regions. Users may request that the service provider allocate computing resources in these data centers to support their computing workloads
.
US 20200250509 A1 teaches step of calculating a computational amount of the computational model and a step of calculating a bandwidth of the computational model. Here, the computational amount refers to a computational amount which is required to execute a learning/inference algorithm for the input data using the computational model. The bandwidth refers to the number of intermediate data and parameters generated for each layer of the computational model. In step 420, the home node N.sub.H may determine an expected computational load for processing a computational model for the input data based on the computational amount and the bandwidth of the computational model.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LECHI TRUONG whose telephone number is (571)272-3767. The examiner can normally be reached 10-8 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor Young Kevin can be reached on (571)270-3180. 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.
/LECHI TRUONG/ Primary Examiner, Art Unit 2194
Prosecution Insights