DETAILED ACTION
This action is responsive to Remarks and Claim amendments filed on March 09, 2026.
Claims 29, 31, 35-36, 39, 41, 44-45 and 48 have been amended. Claim 40 has been canceled. Claim 49 has been newly added.
Claims 29-39 and 41-49 are pending and are presented to examination.
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Examiner Notes
Examiner cites particular columns, paragraphs, figures and line numbers in the references as applied to the claims below for the convenience of the applicant. Although the specified citations are representative of the teachings in the art and are applied to the specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested that, in preparing responses, the applicant fully consider the references in their entirety as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior art or disclosed by the examiner.
Response to Amendments
The objection to the specification (Abstract of the Disclosure) is withdrawn in view of applicant’s amendments.
The objection of claims 29, 36, 39 and 45 is withdrawn in view of applicant’s amendments.
The rejection of claims 31 and 41 under 35 U.S.C. 112(b) is withdrawn in view of applicant’s amendments.
The rejection of claims 29-48 under 35 U.S.C. 101 is withdrawn in view of applicant’s amendments.
Response to Arguments
Applicant has argued that Bernat along with the remaining art of record, does not teach the newly added limitations of independent claims 29, 39 and 48 (Remarks, pages. 12-13). Applicants' arguments have been fully considered and are persuasive. Therefore, the rejection is withdrawn. However, upon further consideration, a new ground of rejection is made as set forth in details below. See Lupesko et al. (US Pat. No. 11,423,283), art being made of record as applied herein.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries 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.
Claims 29-36, 39, 41-45 and 48-49 are rejected under 35 U.S.C. 103 as being unpatentable over Guim Bernat et al. (US Pub. No. 2019/0044831 – hereinafter Bernat – previously presented) in view of Lupesko et al. (US Pat. No. 11,423,283 – hereinafter Lupesko). With respect to claim 29 (currently amended), Bernat teaches a device comprising: a transceiver (See figures 5-6, mesh transceiver 662, wireless network transceiver 666 (and related text) and paragraphs [0029], [0086], [0088]-[0089], “Communications in the cellular network 160, for instance, may be enhanced by systems that offload data, extend communications to more remote devices, or both. The LPWA network 162 may include systems that perform non-Internet protocol (IP) to IP interconnections, addressing, and routing. Further, each of the IoT devices 104 may include the appropriate transceiver for wide area communications with that device. Further, each IoT device 104 may include other transceivers for communications using additional protocols and frequencies. This is discussed further with respect to the communication environment and hardware of an IoT processing device depicted in FIGS. 5 and 6.”) and a processor (See at least figure 6 processor 652 on IoT processing device 650 and paragraphs [0019], [0037], “Often, IoT devices are limited in memory, size, or functionality, allowing larger numbers to be deployed for a similar cost to smaller numbers of larger devices. However, an IoT device may be a smart phone, laptop, tablet, or PC, or other larger device. Further, an IoT device may be a virtual device, such as an application on a smart phone or other computing device. IoT devices may include IoT gateways, used to couple IoT devices to other IoT devices and to cloud applications, for data storage, process control, and the like.”. Furthermore, see paragraphs [0080]-[0081]) configured to:
send, via the transceiver, a model identifier identifying a model and an indication of a first resource constraint associated with the device (See abstract and paragraph [0102]-[0103], [0114], “Various systems and methods for implementing a service-level agreement (SLA) apparatus receive a request from a requester via a network interface of the gateway, the request comprising an inference model identifier that identifies a handler of the request, and a response time indicator.”. See paragraph [0038], “the edge clients send inference requests to the edge network platform (e.g., a platform that is implemented at or near the edge devices, as part of a fog or edge-based topology). In the examples discussed herein, the edge network platform may provide interfaces, applications, or services in the manner of a cloud or similar network-distributed platform, through the use of coordinated edge device functionality through an AI training and inference switch and gateway. It allows each client to specify a Model ID and an optional requirement such as deadline, performance or cost. The gateway contains components that decide which inference model on which assets reachable from the gateway are to be used to satisfy each such request. In turn, each platform in the cloud is equipped with elements that provide necessary information to the gateway for it to make the best-informed decision.”. See figure 3 model ID 310 and paragraphs [0039]-[0040], [0046], [0055], [0060]-[0061], [0066], “FIG. 3 is a block diagram that illustrates one example implementation of an architecture used for the system. An acceleration platform 300 may comprise extensions that include a SLA and service logic element 305 enables service discovery and SLA discovery of models available within the acceleration platforms 300. Using this capability, a solution provider can set or modify the catalogue of available models within each platform. This may be achieved, for example, by using a management and registration interfaces component 355, discussed below. In this example, an AI training to interface switch and gateway 350 may send an incoming inference request, in accordance with the availability of various models, to be received by the SLA and service logic element 305 via a network interface 368. The platform 300 may also comprise a model ID-to-provider performance table (or map) 310, which may provide model ID details (that is, the functions or capabilities available) from each acceleration asset in the platform 300. This table or map 310 may be internal to the platform 300, encapsulated behind the SLA-and-service logic element 305.”. Examiner notes: model identifier/id and constraints/resources). Bernat is silent to disclose, however in an analogous art, Lupesko teaches: receive, via the transceiver in response to the model identifier and the indication of the first resource constraint, a first model configuration of the identified model, wherein the first model configuration is configured to operate within the first resource constraint associated with the device (See abstract, figures 1-10 (and related text) and column 2 lines 45-57, “In some embodiments, an adaptive model engine 109 executes on the edge device to analyze a ML model and adjusts how a model should be executed. For example, the adaptive model engine 109 uses one or more of the following to decide how to adapt a ML model should run, or constituent parts thereof: 1) edge device characteristics including, for example, FLOPS, GPU RAM, CPU RAM, CPU speed, power, network capabilities (wired, wireless, and their types), memory, etc.; 2) model weights; 3) model data type usage; 4) model layers; 5) model operators; 6) web services provider characteristics; 7) available models; 8) available model profiles; and/or 9) desired objectives of ML execution (such as throughput, power usage, and accuracy).”. Furthermore, see column 3 lines 3-16, column 4 lines 24-38, column 5 lines 1-33. Examiner notes: edge devices 101, model 107 and adaptive model based on resources).
process first data using the first model configuration (See figures 1-10 (and related text) and column 5 lines 54-64, “A determination of whether the model being (or to be executed) meets desired performance characteristics given the current device characteristics is made at 805. For example, will the model be able to execute within a desired latency, power envelope, accuracy level, or memory footprint? Or, if the model has yet to be used, will it fit within the available memory? In some embodiments, the desired performance characteristics are provided by a user. If the model will meet desired performance characteristics, then no changes to the model or its execution are made at 807. If the model does not meet desired performance characteristics, in some embodiments a model variant is dynamically generated at 809.”. Examiner notes: processing data). based on processing the first data, send, via the transceiver, a request for a model update and an indication of the second resource constraint associated with the device (See abstract, figures 1-10 and column 4 lines 3-8, “FIG. 4 illustrates embodiments of an adaptive model engine 109. As noted above, this engine executes on an edge device. This engine may perform one of several functions including selecting a model variant 108, updating a model 107 using one or more profile(s) 431, or enabling a model (or portion thereof) on a web services provider”. See column 5 line 61 – column 6 line 9, “If the model will meet desired performance characteristics, then no changes to the model or its execution are made at 807. If the model does not meet desired performance characteristics, in some embodiments a model variant is dynamically generated at 809. For example, some embodiments, a profile of a model variant that will meet the characteristics is selected (one that meets the desired performance characteristics) and used to generate a model variant from the evaluated (or original) model. The model variant may quantize (e.g., change one or more of weights or change data types), change the number of layers, change operator(s) (change to a different operator or fuse operators), unroll the neural network, etc. In some embodiments, a call to an adaptive model service is made which, in turn, provides a profile to use.”. Furthermore, see column 6 lines 23-30. Examiner notes: model adaptation based on resources). receive, via the transceiver in response to the request for the model update and the indication of the second resource constraint, a second model configuration, wherein the second model configuration is configured to operate within the second resource constraint associated with the device and process second data using the second model configuration (See abstract, figures 1-10 and column 4 lines 3-8, “FIG. 4 illustrates embodiments of an adaptive model engine 109. As noted above, this engine executes on an edge device. This engine may perform one of several functions including selecting a model variant 108, updating a model 107 using one or more profile(s) 431, or enabling a model (or portion thereof) on a web services provider”. See column 5 line 61 – column 6 line 9, “If the model will meet desired performance characteristics, then no changes to the model or its execution are made at 807. If the model does not meet desired performance characteristics, in some embodiments a model variant is dynamically generated at 809. For example, some embodiments, a profile of a model variant that will meet the characteristics is selected (one that meets the desired performance characteristics) and used to generate a model variant from the evaluated (or original) model. The model variant may quantize (e.g., change one or more of weights or change data types), change the number of layers, change operator(s) (change to a different operator or fuse operators), unroll the neural network, etc. In some embodiments, a call to an adaptive model service is made which, in turn, provides a profile to use.”. Furthermore, see column 6 lines 23-30. Examiner notes: model adaptation based on resource).
Therefore, it would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Bernat’s teaching with Lupesko’s teaching, as Lupesko would improve data processing and more specifically will provide mechanisms for dynamic machine learning model selection (see paragraph [0001]). With respect to claim 30 (previously presented), Bernat teaches wherein the device comprises a smartphone (see paragraphs [0019], [0037], [0053] “Often, IoT devices are limited in memory, size, or functionality, allowing larger numbers to be deployed for a similar cost to smaller numbers of larger devices. However, an IoT device may be a smart phone, laptop, tablet, or PC, or other larger device. Further, an IoT device may be a virtual device, such as an application on a smart phone or other computing device. IoT devices may include IoT gateways, used to couple IoT devices to other IoT devices and to cloud applications, for data storage, process control, and the like.”).
With respect to claim 31 (currently amended), Bernat teaches wherein the second model configuration comprises the second model (see figure 3 (and related text) and paragraphs [0042]-[0043], [0050], [0060]. Examiner notes: second complete model managing resources). With respect to claim 32 (previously presented), Bernat is silent to disclose, however in an analogous art, Lupesko teaches wherein the second model configuration comprises at least one parameter update to the first model configuration (See column 5 lines 34-43, “FIG. 8 illustrates embodiments of a method performed by an adaptive model engine (such as adaptive model engine 109) of an edge device. At 800, the adaptive model engine is called in some embodiments. For example, in some embodiments, when operating parameters of the edge device change, or what is acceptable performance changes, the adaptive model engine is called by software monitoring usage of the model. In other embodiments, a user calls the adaptive model engine. In still other embodiments, the adaptive model engine is always running.”. See column 7 lines 43-50, “One or more parameters of the received model are changed at 1007. Changing one or more parameters allows for testing how a model reacts to different variables that may be used in the generation of a model variant. Examples of changed parameters including quantizing (such as going from 32-bit floating point to 8-bit integer), layer pruning, changing or fusing operators, and/or unrolling a neural network.”. Examiner notes: sharing operating parameters to adapt proper model). Therefore, it would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Bernat’s teaching with Lupesko’s teaching, as Lupesko would improve data processing and more specifically will provide mechanisms for dynamic machine learning model selection (see paragraph [0001]).
With respect to claim 33 (previously presented), Bernat is silent to disclose, however in an analogous art, Lupesko teaches wherein the second model configuration is configured to operate within the first resource constraint and within the second resource constraint (See abstract, “techniques for model adaptation are described. For example, a method of receiving a call to provide either a model variant or a model variant profile of a deep learning model, the call including desired performance of the deep learning model, a deep learning model identifier, and current edge device characteristics; comparing the received current edge device characteristics to available model variants and profiles based on the desired performance of the deep learning model to generate or select a model variant or profile, the available model variants and profiles determined by the model identifier; and sending the generated or selected model variant or profile to the edge device to use in inference is detailed.” See column 2 line 37-44, “As detailed, a ML model 107 (or a variant model 108) may also be executed on an edge device 101 using execution resources 103. Exemplary edge devices include, but are not limited to: cameras, mobile devices, audio equipment, etc. However, these devices 101 are typically less powerful than the host(s) provided by the web services provider 111 and running the same model with identical parameters may not be practical.”. Furthermore, see column 2 line 58 - column 3 line 2, “In some embodiments, the edge device 101 stores a variant of the model 107 (model variant 108, or model′ 108) or a profile to be used to change the model 107. For example, model variant 108 may quantize the model 107 (for example, use different weights and/or different data types), or remove one or more layers (or nodes of layers), or change one or more operators of the layers of the model 107. In some embodiments, these changes are stored in a profile which is consulted during execution to alter the model 107 on the fly. In some embodiments, the model 107 and model variant 108 (model′) are a part of one or more containers to be executed by the execution resources 103.”. Examiner notes: model 107 and model variant 108). Therefore, it would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Bernat’s teaching with Lupesko’s teaching, as Lupesko would improve data processing and more specifically will provide mechanisms for dynamic machine learning model selection (see paragraph [0001]).
With respect to claim 34 (previously presented), Bernat is silent to disclose, however in an analogous art, Lupesko teaches wherein at least one of the first resource constraint or the second resource constraint comprises at least one of a limit on computational resources associated with the device, or an accuracy constraint (See column 2 lines 45-57 and figures 1-10 (and related text), “In some embodiments, an adaptive model engine 109 executes on the edge device to analyze a ML model and adjusts how a model should be executed. For example, the adaptive model engine 109 uses one or more of the following to decide how to adapt a ML model should run, or constituent parts thereof: 1) edge device characteristics including, for example, FLOPS, GPU RAM, CPU RAM, CPU speed, power, network capabilities (wired, wireless, and their types), memory, etc.; 2) model weights; 3) model data type usage; 4) model layers; 5) model operators; 6) web services provider characteristics; 7) available models; 8) available model profiles; and/or 9) desired objectives of ML execution (such as throughput, power usage, and accuracy).”). Therefore, it would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Bernat’s teaching with Lupesko’s teaching, as Lupesko would improve data processing and more specifically will provide mechanisms for dynamic machine learning model selection (see paragraph [0001]). With respect to claim 35 (currently amended), Bernat is silent to disclose, however in an analogous art, Lupesko teaches wherein at least one of the first resource constraint or the second resource constraint comprises a resource availability constraint at the device (See column 2 lines 45-57 and figures 1-10 (and related text), “In some embodiments, an adaptive model engine 109 executes on the edge device to analyze a ML model and adjusts how a model should be executed. For example, the adaptive model engine 109 uses one or more of the following to decide how to adapt a ML model should run, or constituent parts thereof: 1) edge device characteristics including, for example, FLOPS, GPU RAM, CPU RAM, CPU speed, power, network capabilities (wired, wireless, and their types), memory, etc.; 2) model weights; 3) model data type usage; 4) model layers; 5) model operators; 6) web services provider characteristics; 7) available models; 8) available model profiles; and/or 9) desired objectives of ML execution (such as throughput, power usage, and accuracy).”). Therefore, it would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Bernat’s teaching with Lupesko’s teaching, as Lupesko would improve data processing and more specifically will provide mechanisms for dynamic machine learning model selection (see paragraph [0001]). With respect to claim 36 (currently amended), Bernat is silent to disclose, however in an analogous art, Lupesko teaches wherein the second resource constraint is based on the device moving to an edge computing node close to the device (See figures 5-6 (and related text) and column 4 lines 39-48, “FIG. 5 illustrates an embodiment of model profile. As such, the profile 501 includes fields for a layer identifier, a node identifier (per layer), an input identifier (per node), a weight used by the node for an input, what operator(s) are applied by a layer, the data type used in the layer (such as which floating point (FP) type (e.g., 256, 128 64, 32, or 16-bit), integer, and Boolean), energy usage (overall and by layer), execution time/latency (overall and by layer), memory usage (overall and by layer), and accuracy. In some embodiments, the network is unrolled.”, node ID and weights). Therefore, it would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify Bernat’s teaching with Lupesko’s teaching, as Lupesko would improve data processing and more specifically will provide mechanisms for dynamic machine learning model selection (see paragraph [0001]). With respect to claims 39, 41-45, the claims are directed to a method that corresponds to the device recited in claims 29, 31-34 and 36, respectively (see the rejection of claims 29-34 and 36 above). With respect to claims 48-49, the claims are directed to a medium that corresponds to the device recited in claims 29 and 35, respectively (see the rejection of claims 29 and 35 above; wherein Bernat also teaches such medium in paragraph [0097]).
Claims 37-38 and 46-47 are rejected under 35 U.S.C. 103 as being unpatentable over Guim Bernat et al. (US Pub. No. 2019/0044831) in view of Lupesko et al. (US Pat. No. 11,423,283) and further in view of Tsiry Mayet et al. (“SKIPW: Resource Adaptable RNN with Strict Upper Computational Limit” – hereinafter Mayet – IDS 03/09/2023 – previously presented). With respect to claim 37 (previously presented), Bernat in view of Lupesko is silent to disclose, however in an analogous art, Mayet teaches wherein at least one of the first model configuration or the second model configuration is configured to process at least one of the first data or the second data based on a windowing function (See figures 1-2, 7, windowing function). Therefore, it would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify the combination of Bernat and Lupesko, by utilizing a windowing function as suggested by Mayet, as Mayet would provide a method to allow recurrent neural networks to trade off accuracy for computational cost during the analysis of a sequence (see abstract). With respect to claim 38 (previously presented), Bernat in view of Lupesko is silent to disclose, however in an analogous art, Mayet teaches wherein the first model configuration and the second model configuration comprise a neural network (see at least the abstract and “Introduction” section, which discloses RNN for models). Therefore, it would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to modify the combination of Bernat and Lupesko, by using neural network model as suggested by Mayet, as Mayet would provide a flexible recurrent neural network to dynamically adapt computational cost (see Introduction section, third paragraph). With respect to claims 46-47, the claims are directed to a method that corresponds to the device recited in claims 37-38, respectively (see the rejection of claims 37-38 above).
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 ANIBAL RIVERACRUZ whose telephone number is (571)270-1200. The examiner can normally be reached Monday-Friday 9:30 AM-6:00 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, Hyung S Sough can be reached at 5712726799. 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.
/ANIBAL RIVERACRUZ/Primary Examiner, Art Unit 2192
Prosecution Insights