DETAILED ACTION
This Office Action is in response to the remarks entered on 07/02/2026. Claims 1, 9, 17 are amended. Claims 18-20 are cancelled. Claims 21-23 are newly added. Claims 1-17, 21-23 are pending.
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
The Applicant’s arguments regarding the rejection of above claims have been fully considered.
In reference to Applicant’s arguments about:
35 USC 101 Rejections.
Examiner’s response:
Rejections are withdrawn in view of Applicant’s arguments and claim amendments.
In reference to Applicant’s arguments about:
35 USC 103 Rejections.
Examiner’s response:
Applicant's amendments and new claims added necessitated the new grounds of rejection presented in this Office action below; therefore, Applicant’s arguments have been fully considered but are moot in view of new grounds of rejections.
Examiner respectfully would like to remind applicant about the claim limitations which are deemed to contain allowable subject matter (claims 7 and 15). Examiner would like to suggest applicant to include these limitations into independent claims in order to withdraw the prior art rejections and expedite prosecution.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied 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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 9 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al (US 2021/0126737- hereinafter Zhang) in view of Valipour et al (US Pub. No. 2023/0139718- hereinafter Valipour) and further in view of Beauchesne et al (US Patent 11,509,674- hereinafter Beauchesne).
Referring to Claim 1, Zhang teaches a computer-implemented method comprising:
receiving, by one or more processors, at an edge device running a local instance of a machine learning (ML) model, a set of inference data comprising a plurality of datapoints, wherein the local instance of the ML model is a deployed version of the ML model running in a cloud environment, and wherein the ML model was trained in the cloud environment and then deployed to the edge device (see Zhang at Fig. 13 and [0251]: “When the associated terminal 120 and AI server jointly train the AI model, the terminal 120 may first download an universal version or basic version of AI model suitable for the current specific base station 110 from the AI server. Also, the terminal 120 may continuously accumulate corresponding terminal data during use”. Therefore, the terminal is interpreted as the edge device, the terminal data is interpreted as inference data, and since a model is downloaded initially from an AI server, this is interpreted as the ML model trained in the cloud and deployed to the edge device);
running, by the one or more processors, the plurality of datapoints through one or more filters to determine a probability for each datapoint of whether a respective datapoint should be sent back to the cloud environment and used for retraining the ML model (see Zhang at [0251]: “Also, the terminal 120 may continuously accumulate corresponding terminal data during use; after the accumulated data exceeds a certain threshold, the terminal may re-train the AI model based on the downloaded AI model and the collected local data; a new AI model obtained by training may be uploaded to the AI server”. Therefore, since the updates are uploaded to the server after meeting a threshold, this is interpreted as “filtering” the datapoints to determine a probability to be sent back according to the threshold); and
determining, by the one or more processors, for each datapoint, whether the probability for the respective datapoint meets a send back threshold that is required to be met before the respective datapoint is sent back to the cloud environment (see Zhang at [0251]: “Also, the terminal 120 may continuously accumulate corresponding terminal data during use; after the accumulated data exceeds a certain threshold, the terminal may re-train the AI model based on the downloaded AI model and the collected local data; a new AI model obtained by training may be uploaded to the AI server”. Therefore, since the updates are uploaded to the server after meeting a threshold, this is interpreted as running the datapoints and determining a probability to be sent back according to the threshold).
However, Zhang is silent specifically regarding the determination of a probability for each datapoint of whether a respective datapoint should be sent back to the cloud environment and used for retraining the ML model, and wherein datapoints that do not meet the send back threshold are discarded at the edge device.
Valipour teaches, in an analogous system, determine a probability for each datapoint of whether a respective datapoint should be sent back to the cloud environment and used for retraining the ML model (see Valipour at [0135]: “At the end of each iteration, probabilities 521-522 are compared to detect whether or not data drift occurred. If probability 522 exceeds probability 521 by at least a threshold difference, then recent population 542 has diverged from old population 541 and data drift is detected, in which case ML model 530 should be retrained”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zhang with the above teachings of Valipour by having a model deployed in an edge device, running it, and determining if an update is needed, as taught by Zhang, wherein the update needed is based on a probability, as taught by Valipour. The modification would have been obvious because one of ordinary skill in the art would be motivated to retrain a ML model only if it is needed based on a probability in order to maximize accuracy and minimize cost (as suggested by Valipour at [0040]: “For example, retraining may take hours or days. Thus, there is a natural tension between frequent retraining to maximize accuracy and infrequent retraining to minimize cost”. Further at [0135]: “If probability 522 exceeds probability 521 by at least a threshold difference, then recent population 542 has diverged from old population 541 and data drift is detected, in which case ML model 530 should be retrained. Otherwise, if a maximum count of iterations occurred, then data drift has not occurred and retraining ML model 530 is unneeded”).
Beauchesne teaches, in an analogous system, wherein datapoints that do not meet the send back threshold are discarded at the edge device (see Beauchesne at Col. 15: lines 25-57: “the process may generate each synthetic datapoint as a candidate, and then check the candidate datapoint against an acceptance criterion to determine whether the candidate datapoint should be retained (e.g. added to a dataset for machine learning applications). The acceptance criterion may be based on the density function determined for the observed dataset. For example, the acceptance criterion may accept a candidate datapoint if the datapoint is within an acceptable range of probability values indicated by the density function. Datapoint 854 may be discarded because it is outside such an acceptable range, which means that the datapoint is too dissimilar from the observed datapoints. Depending on the embodiment, generated synthetic datapoints may be discarded for a variety of reasons. In some embodiments, a candidate datapoint may be discarded if its probability of observation based on the density function is too high. In some embodiments, a candidate datapoint may be discarded if it is too close to other synthetic datapoints in the feature space. In some embodiments, a candidate datapoint may be discard if there are already a sufficient number of synthetic datapoints generated in a similarity group of the candidate point. In some embodiments, candidate datapoints may be discarded in a random fashion”. Therefore, Beauchesne’s determination of retaining or discarding each data points based on certain criteria, in combination with Zhang’s use of data for retraining the model and Valipour probability measures would have been obvious to be combined as a person having ordinary skill in the art in order to determine whether to use specific datapoints for training a model based on a probability for retraining, and discarding the datapoints that don’t meet the specific criteria).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang and Valipour with the above teachings of Beauchesne by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score and drift, as taught by Zhang and Valipour, and also considering individual datapoints to use them for retraining and discarding the ones that don’t meet the criteria, as taught by Beauchesne. The modification would have been obvious because one of ordinary skill in the art would be motivated to evaluate each individual datapoint for retaining it or discarding it based on an acceptance criterion (as suggested by Beauchesne at [Col. 15]: “check the candidate datapoint against an acceptance criterion to determine whether the candidate datapoint should be retained (e.g. added to a dataset for machine learning applications). The acceptance criterion may be based on the density function determined for the observed dataset. For example, the acceptance criterion may accept a candidate datapoint if the datapoint is within an acceptable range of probability values indicated by the density function. Datapoint 854 may be discarded because it is outside such an acceptable range”).
Referring to independent Claim 9 and Claim 17, they are rejected on the same basis as independent claim 1, mutatis mutandis, since they are analogous claims.
Claims 2, 4, 10, 12 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Valipour, Beauchesne and further in view of Dodwell et al (US Pub. No. 2021/0174223 - hereinafter Dodwell).
Referring to Claim 2, the combination of Zhang, Valipour and Beauchesne teaches the computer-implemented method of claim 1, wherein running the plurality of datapoints through the one or more filters comprises:
running, by the one or more processors, the plurality of datapoints through a drift detection model that outputs a drift score for each datapoint on how different the respective datapoint is from other datapoints used to train the ML model (see Valipour at [0135]: “If probability 522 exceeds probability 521 by at least a threshold difference, then recent population 542 has diverged from old population 541 and data drift is detected, in which case ML model 530 should be retrained”. Further, Beauchesne teaches the analysis for each datapoint as explained at Claim 1);
running, by the one or more processors, the plurality of datapoints through a criteria filter that outputs a criteria score for how well each datapoint satisfies a preset set of criteria (see Zhang at [0251]: “Also, the terminal 120 may continuously accumulate corresponding terminal data during use; after the accumulated data exceeds a certain threshold, the terminal may re-train the AI model based on the downloaded AI model and the collected local data”. Therefore, this metric using a threshold for updating training is interpreted as the criteria score. Furthermore see Valipour at [0024]: “A target machine learning model may be retrained with recent data when a comparison of the first fitness score to the second fitness score indicates data drift”. Therefore, this score also used to measure drift and therefore retraining is also interpreted as a criteria score); and
running, by the one or more processors, the bias score, the drift score, and the criteria score for each datapoint through a send back ML model that outputs a final score for each datapoint of the probability of whether the respective datapoint should be sent back to the cloud environment for retraining the ML model (see Zhang at [0251]: “Also, the terminal 120 may continuously accumulate corresponding terminal data during use; after the accumulated data exceeds a certain threshold, the terminal may re-train the AI model based on the downloaded AI model and the collected local data”. Therefore, this metric using a threshold for updating training is interpreted as the criteria score. Furthermore see Valipour at [0024]: “A target machine learning model may be retrained with recent data when a comparison of the first fitness score to the second fitness score indicates data drift”. Also at [0135]: “At the end of each iteration, probabilities 521-522 are compared to detect whether or not data drift occurred. If probability 522 exceeds probability 521 by at least a threshold difference, then recent population 542 has diverged from old population 541 and data drift is detected, in which case ML model 530 should be retrained”). Therefore, this score also used to measure drift and therefore retraining is also interpreted as a criteria score, the drift is interpreted as the drift score).
However, the combination fails to teach:
running, by the one or more processors, the plurality of datapoints through a bias detection model that outputs a bias score for each datapoint on how likely the respective datapoint is to have been misclassified by the local instance of the ML model; and
running, by the one or more processors, the bias score, for each datapoint through a send back ML model that outputs a final score for each datapoint of the probability of whether the respective datapoint should be sent back to the cloud environment for retraining the ML model.
Dodwell teaches, in an analogous system,
running, by the one or more processors, the plurality of datapoints through a bias detection model that outputs a bias score for each datapoint on how likely the respective datapoint is to have been misclassified by the local instance of the ML model (see Dodwell at [0027]: “The bias mitigation server 202 then computes a potential bias score either in numeric form or in another form, such as labeling including “risky,” “moderately risky,” or “safe.” After such labeling, the machine learning model would be retrained (again outside the fast path) to reduce the future risk of input that was labeled as “risky.”); and
running, by the one or more processors, the bias score, for each datapoint through a send back ML model that outputs a final score for each datapoint of the probability of whether the respective datapoint should be sent back to the cloud environment for retraining the ML model (see Dodwell at [0027]: “The bias mitigation server 202 then computes a potential bias score either in numeric form or in another form, such as labeling including “risky,” “moderately risky,” or “safe.” After such labeling, the machine learning model would be retrained (again outside the fast path) to reduce the future risk of input that was labeled as “risky.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour and Beauchesne with the above teachings of Dodwell by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score and drift, as taught Zhang, Valipour and Beauchesne, and also considering bias for retraining purposes, as taught by Dodwell. The modification would have been obvious because one of ordinary skill in the art would be motivated to retrain a ML model based on a combined score of criteria, drift and bias in order to maximize accuracy and minimize cost (as suggested by Valipour at [0040]: “For example, retraining may take hours or days. Thus, there is a natural tension between frequent retraining to maximize accuracy and infrequent retraining to minimize cost”) and to minimize both the risk of negative impact and the potential of lost revenue by mitigating biases (as suggested by Dodwell at [0027]: “The process factors the cost of a label to minimize both the risk of negative impact and the potential of lost revenue”).
Referring to Claim 4, the combination of Zhang, Valipour, Beauchesne and Dodwell teaches the computer-implemented method of claim 2, wherein running the plurality of datapoints through the bias detection model, running the plurality of datapoints through the drift detection model, running the plurality of datapoints through the criteria filter, running the bias score, the drift score, and the criteria score for each datapoint through the send back ML model, and determining whether the final score for a respective datapoint meets the send back threshold are completed on the edge device (see Zhang at [0251]: “Also, the terminal 120 may continuously accumulate corresponding terminal data during use; after the accumulated data exceeds a certain threshold, the terminal may re-train the AI model based on the downloaded AI model and the collected local data; a new AI model obtained by training may be uploaded to the AI server”. Therefore, since the updates are uploaded to the server after meeting a threshold, this is interpreted as the determination of the score to send back to the server being completed at the edge device. Furthermore, as explained at Claim 2, Zhang teaches the criteria, Valipour teaches the drift score, and Dodwell teaches the bias score).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour and Beauchesne with the above teachings of Dodwell by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score and drift, as taught by Zhang, Valipour and Beauchesne, and also considering bias for retraining purposes, as taught by Dodwell. The modification would have been obvious because one of ordinary skill in the art would be motivated to retrain a ML model based on a combined score of criteria, drift and bias in order to maximize accuracy and minimize cost (as suggested by Valipour at [0040]: “For example, retraining may take hours or days. Thus, there is a natural tension between frequent retraining to maximize accuracy and infrequent retraining to minimize cost”) and to minimize both the risk of negative impact and the potential of lost revenue by mitigating biases (as suggested by Dodwell at [0027]: “The process factors the cost of a label to minimize both the risk of negative impact and the potential of lost revenue”).
Referring to dependent Claim 10, it is rejected on the same basis as dependent claim 2, mutatis mutandis, since they are analogous claims.
Referring to dependent Claim 12, it is rejected on the same basis as dependent claim 4, mutatis mutandis, since they are analogous claims.
Claims 3, 6, 11, 14 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Valipour, Beauchesne, Dodwell, and further in view of Watson (US Patent No. 11,868,852- hereinafter Watson).
Referring to Claim 3, the combination of Zhang, Valipour, Beauchesne, Dodwell teaches the computer-implemented method of claim 2, however, fails to teach wherein running the plurality of datapoints through the one or more filters further comprises:
running, by the one or more processors, at the edge device, the plurality of datapoints through a user flag filter that outputs a flag score based on whether the respective datapoint was flagged as an incorrect inference by a user.
Watson teaches, in an analogous system, running, by the one or more processors, at the edge device, the plurality of datapoints through a user flag filter that outputs a flag score based on whether the respective datapoint was flagged as an incorrect inference by a user (see Watson at Col. 14: lines 1-6: “Since the data objects were annotated with entity-determined risk scores, the estimated risk scores from the regressor can be compared against the entity-determined risk scores to determine which data objects had risk estimates that differed by more than an allowable amount from the entity-determined scores”. See also lines 11-12: “The or other approaches can be used to flag the data objects as having incorrect risk score estimates”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour, Beauchesne, Dodwell with the above teachings of Watson by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score, drift and bias for retraining purposes, as taught by Zhang, Valipour, Beauchesne, Dodwell, and also considering flagging outputs that are incorrect for retraining purposes, as taught by Watson. The modification would have been obvious because one of ordinary skill in the art would be motivated to improve the accuracy of the model (as suggested by Watson at Col. 11: 3-8: “This information can be surfaced to a user, but can also be utilized by a system or service to automatically retrain the model in order to improve the accuracy of the model. Any relevant documents can then be reprocessed in order to reduce any inaccuracies in the relevant risk scores”).
Referring to Claim 6, the combination of Zhang, Valipour, Beauchesne, Dodwell and Watson teaches the computer-implemented method of claim 3, wherein the send back ML model is a logistic regression ML model trained to take the flag score, the bias score, the drift score, and the criteria score for each datapoint as input features and outputs the probability as the final score between zero (0) and one (1) that the respective datapoint should be sent back to the cloud environment (see Watson at Col. 14: lines 1-6: “Since the data objects were annotated with entity-determined risk scores, the estimated risk scores from the regressor can be compared against the entity-determined risk scores to determine which data objects had risk estimates that differed by more than an allowable amount from the entity-determined scores”. Further, Col. 2: 21-27 recites “In examples herein the risk score goes from 1-10 on a linear scale, but various other scores and scales can be used as well within the scope of the various embodiments. This information can then be used to train and test the random forests for purposes of estimating risk scores based at least in part upon these or other learned features”. Therefore, this regressor taught by Watson is interpreted as the logistic regression ML model).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour, Beauchesne, Dodwell with the above teachings of Watson by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score, drift and bias for retraining purposes, as taught by Zhang, Valipour, Beauchesne, Dodwell, and also considering flagging outputs that are incorrect for retraining purposes, as taught by Watson. The modification would have been obvious because one of ordinary skill in the art would be motivated to improve the accuracy of the model (as suggested by Watson at Col. 11: 3-8: “This information can be surfaced to a user, but can also be utilized by a system or service to automatically retrain the model in order to improve the accuracy of the model. Any relevant documents can then be reprocessed in order to reduce any inaccuracies in the relevant risk scores”).
The combination of Zhang, Valipour, Beauchesne, Dodwell and Watson fails to explicitly teach the final score between zero (0) and one (1). However this difference is only found in the nonfunctional descriptive material and is not functionally involved in the steps recited. A limitation on a claim can broadly be thought of as its ability to make a meaningful contribution to the definition of the invention in a claim. In other words, language that is not functionally interrelated with the useful acts, structure, or properties of the claimed invention will not serve as a limitation. Thus, this descriptive material will not distinguish the claimed invention from the prior art in terms of patentability, see In re Gulack, 703 F.2d 1381, 1385, 21 7 USPQ 401, 404 (Fed. Cir. 1983); In re Lowry, 32 F.3d 1579, 32 USPQ2d 1031 (Fed. Cir. 1994). In the present case, the steps recited in these claims would be performed the same regardless of the score being between 0 and 1, or 1 to 10, as the scale is what matters in order to conclude that a score closest to the far left limit is a low score, and a score close to the far right is a high score, as Watson even suggests that a score goes from 1-10 on a linear scale, but various other scores and scales can be used as well within the scope of the various embodiments.
Nonfunctional descriptive material cannot render nonobvious an invention that would have otherwise been obvious. In re Ngai, 367 F.3d 1336, 1339, 70USPQ2d 1862, 1864 (Fed. Cir. 2004). Cf. In re Gulack, 703 F.2d 1381, 1385, 21 7 USPQ 401, 404 (Fed. Cir. 1983) (when descriptive material is not functionally related to the substrate, the descriptive material will not distinguish the invention from the prior art in terms of patentability).
A limitation on a claim can broadly be thought of as its ability to make a meaningful contribution to the definition of the invention in a claim. In other words, language that is not functionally interrelated with the useful acts, structure, or properties of the claimed invention will not serve as a limitation. Simply stated, in the instant claims, these limitations are non-functional descriptive material which is not functionally involved in the functionality of the claimed invention. In the instant claims, the score scale will be used analogously, regardless on being either from 0 to 1, or 1 to 10, as the scaling is what matters. Thus, this descriptive material will not distinguish the claimed invention in terms of patentability and cannot render nonobvious an invention that would have otherwise been obvious, see In re Gulack, 703 F.2d 1381, 1385, 217 USPQ 401, 404 (Fed. Cir. 1983); In re Lowry, 32 F3.d 1579, 32 USPQ2d 1031 (Fed. Cir. 1994)).
Therefore, it would have been obvious to a person of ordinary skill in the art at the time the invention was made to recite that the final score goes between 0 and 1, since these limitations do not functionally relate to the steps in the method claimed and no specific relevance or criticality is ascribed to this limitation as claimed.
Referring to dependent Claim 11, it is rejected on the same basis as dependent claim 3, mutatis mutandis, since they are analogous claims.
Referring to dependent Claim 14, it is rejected on the same basis as dependent claim 6, mutatis mutandis, since they are analogous claims.
Claims 5 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Valipour, Beauchesne, Dodwell, and further in view of Sibley et al (US Pub. No. 2022/0183208- hereinafter Sibley).
Referring to Claim 5, the combination of Zhang, Valipour, Beauchesne, Dodwell teaches the computer-implemented method of claim 2, however, fails to teach wherein running the plurality of datapoints through the bias detection model, running the plurality of datapoints through the drift detection model, running the plurality of datapoints through the criteria filter, running the bias score, the drift score, and the criteria score for each datapoint through the send back ML model, and determining whether the final score for a respective datapoint meets the send back threshold are completed on an edge server that communicates with the edge device.
Sibley teaches, in an analogous system, running the plurality of datapoints through the bias detection model, running the plurality of datapoints through the drift detection model, running the plurality of datapoints through the criteria filter, running the bias score, the drift score, and the criteria score for each datapoint through the send back ML model, and determining whether the final score for a respective datapoint meets the send back threshold are completed on an edge server that communicates with the edge device (see at [0443]: “In some embodiments, the system further includes an edge server positioned between the offsite computing resources and the onsite computing resources, wherein the edge serve is configured for: (1) facilitating a communication from the offsite computing resources to the onsite platform, (2) facilitating a communication from the onsite platform to the offsite computing resources, or (3) offload computing tasks from the onsite platform in coordination with the onsite platform. In some embodiments, the edge server offloads ML computing tasks from the onsite platform such that the onsite platform is limited to performing computer vision analysis of the result of activating the treatment mechanism”. Therefore, this edge server (interpreted as the claimed edge server) performing tasks to offload the onsite platform (interpreted as the edge device) is analogous to the claimed completion of all the determination steps as claimed. Furthermore, as explained in Claim 2, the combination of Zhang, Valipour and Dodwell teaches everything except being done at the edge server).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour, Beauchesne, Dodwell with the above teachings of Sibley by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score, bias and drift, as taught by Zhang, Valipour, Beauchesne, Dodwell, wherein all these determinations are done at a n edge server, as taught by Sibley. The modification would have been obvious because one of ordinary skill in the art would be motivated to offload computing tasks from the edge devices (as suggested by Sibley at [0443]: “In some embodiments, the system further includes an edge server positioned between the offsite computing resources and the onsite computing resources, wherein the edge serve is configured for: (1) facilitating a communication from the offsite computing resources to the onsite platform, (2) facilitating a communication from the onsite platform to the offsite computing resources, or (3) offload computing tasks from the onsite platform in coordination with the onsite platform. In some embodiments, the edge server offloads ML computing tasks from the onsite platform such that the onsite platform is limited to performing computer vision analysis of the result of activating the treatment mechanism”).
Referring to dependent Claim 13, it is rejected on the same basis as dependent claim 5, mutatis mutandis, since they are analogous claims.
Claims 8 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Valipour, Beauchesne, and further in view of Yan et al (US Pub. No. 2022/0200858- hereinafter Yan).
Referring to Claim 8, the combination of Zhang, Valipour, Beauchesne teaches the computer-implemented method of claim 1, however, fails to teach wherein the send back threshold dynamically adjusts based on network conditions between the edge device and the cloud environment.
Yan teaches, in an analogous system, wherein the send back threshold dynamically adjusts based on network conditions between the edge device and the cloud environment (see Yan at [0055]: “In an embodiment, when running, the network device may divide a running time of the network device into a plurality of periods, and obtain network statistical data respectively corresponding to the plurality of periods. Then the network device may determine and configure the ECN high threshold, the ECN low threshold, and the ECN mark probability based on network statistical data in different periods, so that the ECN high threshold, the ECN low threshold, and the ECN mark probability are dynamically adapted to a current network transmission characteristic (network traffic model) of the network device, thereby ensuring network transmission performance”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour, Beauchesne with the above teachings of Yan by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score and drift, as taught by Zhang, Valipour, Beauchesne, and sending the update based on network conditions, as taught by Yan. The modification would have been obvious because one of ordinary skill in the art would be motivated to ensure network transmission performance (as suggested by Yan at [0055]: “Then the network device may determine and configure the ECN high threshold, the ECN low threshold, and the ECN mark probability based on network statistical data in different periods, so that the ECN high threshold, the ECN low threshold, and the ECN mark probability are dynamically adapted to a current network transmission characteristic (network traffic model) of the network device, thereby ensuring network transmission performance”).
Referring to dependent Claim 16, it is rejected on the same basis as dependent claim 8, mutatis mutandis, since they are analogous claims.
Claims 21 and 23 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Valipour, Beauchesne, and further in view of Raviv (US Pub. No. 2022/0144392- hereinafter Raviv).
Referring to Claim 21, the combination of Zhang, Valipour, Beauchesne teaches the computer-implemented method of claim 1, however, fails to teach wherein the probability is used to generate a priority ranking for individual datapoints prior to transmission to the cloud environment.
Raviv teaches, in an analogous system, wherein the probability is used to generate a priority ranking for individual datapoints prior to transmission to the cloud environment (see Raviv at [0004]: “transmitting at least some of the data using the at least one remote data communication link, wherein data are transmitted according to priority determined for the data, thereby facilitating transmission of relevant data for the purpose of training one or more machine learning algorithms providing output based on these data”, and [0100-0101]: “According to some embodiments, transmission of data according to their priority can include transmitting data in a (temporal) order which reflects priority, e.g. first data with higher priority, and then data with lower priority. Therefore, a priority queue can be created, wherein the data with higher priority are located at the top of the priority queue, and transmitted first, whereas the data with lower priority are located at the bottom of the priority queue, and transmitted last”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour, Beauchesne with the above teachings of Raviv by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score and drift, as taught by Zhang, Valipour, Beauchesne, and determining each individual datapoint to be useful or not to update the model based on a priority ranking, as taught by Raviv. The modification would have been obvious because one of ordinary skill in the art would be motivated to facilitate transmission of relevant data for the purpose of training one or more machine learning algorithms according to its priority (as suggested by Raviv at [0004]: “wherein data are transmitted according to priority determined for the data, thereby facilitating transmission of relevant data for the purpose of training one or more machine learning algorithms”).
Referring to Claim 23, the combination of Zhang, Valipour, Beauchesne teaches the computer-implemented method of claim 1, however, fails to teach wherein datapoints meeting the send back threshold are queued for transmission to the cloud environment for retraining of the ML model.
Raviv teaches, in an analogous system, wherein datapoints meeting the send back threshold are queued for transmission to the cloud environment for retraining of the ML model (see Raviv at [0004]: “transmitting at least some of the data using the at least one remote data communication link, wherein data are transmitted according to priority determined for the data, thereby facilitating transmission of relevant data for the purpose of training one or more machine learning algorithms providing output based on these data”, and [0100-0101]: “According to some embodiments, transmission of data according to their priority can include transmitting data in a (temporal) order which reflects priority, e.g. first data with higher priority, and then data with lower priority. Therefore, a priority queue can be created, wherein the data with higher priority are located at the top of the priority queue, and transmitted first, whereas the data with lower priority are located at the bottom of the priority queue, and transmitted last”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour, Beauchesne with the above teachings of Raviv by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score and drift, as taught by Zhang, Valipour, Beauchesne, and determining each individual datapoint to be useful or not to update the model based on a priority ranking, as taught by Raviv. The modification would have been obvious because one of ordinary skill in the art would be motivated to facilitate transmission of relevant data for the purpose of training one or more machine learning algorithms according to its priority (as suggested by Raviv at [0004]: “wherein data are transmitted according to priority determined for the data, thereby facilitating transmission of relevant data for the purpose of training one or more machine learning algorithms”).
Claim 22 is rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Valipour, Beauchesne, and further in view of Asaba (US Pub. No. 2022/0021604 - hereinafter Asaba).
Referring to Claim 22, the combination of Zhang, Valipour, Beauchesne teaches the computer-implemented method of claim 1, however, fails to teach wherein the send back threshold is adjusted based on measured network latency between the edge device and the cloud environment.
Asaba teaches, in an analogous system, wherein the send back threshold is adjusted based on measured network latency between the edge device and the cloud environment (see Asaba at [0087]: “The bandwidth information 520 is information indicating a network bandwidth of each information processing device 300 (also referred to as each controller 330). The bandwidth information 520 is used as a threshold in adjusting the amount of imaging data transmitted from each edge device 200 to each transmission destination”. Further, at [0111]: “For example, as communication performance, the master controller acquires information on latency and bandwidth between each edge device 200 and each controller 330. For example, the edge device 200 measures latency between the own device and each controller 330 by using a Ping command, and notifies the master controller of the measured latency”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Zhang, Valipour, Beauchesne with the above teachings of Asaba by having a model deployed in an edge device, running it, and determining if an update is needed based on a criteria score and drift, as taught by Zhang, Valipour, Beauchesne, wherein one of the criteria to transmit the updates is based on latency, as taught by Asaba. The modification would have been obvious because one of ordinary skill in the art would be motivated to adjust the amount of datapoint to be transmitted based on latency (as suggested by Asaba at [0087]: “adjusting the amount of imaging data transmitted from each edge device 200 to each transmission destination”).
Allowable Subject Matter
Claims 7 and 15 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
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 LUIS A SITIRICHE whose telephone number is (571)270-1316. The examiner can normally be reached M-F 9am-6pm.
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, David Yi can be reached at (571) 270-7519. 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.
/LUIS A SITIRICHE/ Primary Examiner, Art Unit 2126
Prosecution Insights