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 .
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 27-29, 31-40 and 42-46 are rejected under 35 U.S.C. 103 as being unpatentable over Berger et al (US 2022/0030515, hereinafter Berger), in view of Rajkumar et al (US 2023/0004802, hereinafter Rajkumar) and in view of Quan et al (US 2024/0135516, hereinafter Quan).
Regarding claim 27, Berger discloses a method performed by a network node for managing communication in a communication network (base station supporting communications with UEs, Para [0004]), the method comprising: but does not explicitly disclose wherein the level of risk is associated with a level of a degradation of a performance in the communication network. Berger discloses sensing is combined with machine learning algorithm to enhance 5G RAN protocols and enhancement for a base station sleep mode aided by this sensing, Para [0028], sensing combined with ML algorithm to enhance power saving of the RAN, Para [0053]. Rajkumar discloses training machine learning model with low quality information degrades model accuracy and degrades performance, Para [0163], obvious to one of ordinary skill using a low accuracy ML model would reduce performance in the network; and does not disclose obtaining an indication relating to a level of risk in using an artificial intelligence (AI) module for managing a feature (Quan discloses determine ratio of a count of positive results to the count of a second portion as the accuracy rate of the trained machine learning model, Para [0111]), comparing the obtained level of risk with a set level of risk (Quan discloses determine if accuracy rate is greater than a threshold (e.g. 90%), Para [0111]); and based on the comparison, activating or deactivating the AI module for managing the feature (Quan discloses if accuracy is greater than threshold determine the trained machine learning model as the correction model, otherwise initiate new training iterations, Para [0111], using the correction model, Para [0096] ergo it is active). It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to utilize the techniques taught by Quan and Rajkumar in the system of Berger in order to the utilize of machine learning algorithms, where only highly accurate models are used to prevent reduction of performance in the network.
Regarding claims 28 and 39, Berger discloses the method/node of claim 27/37, wherein obtaining comprises calculating the level of risk using a calculation model (obvious to one of ordinary skill to use a calculation model, the model being a “calculation” model is the Applicant being their own lexicographer).
Regarding claims 29 and 40, Berger discloses the method/node of claim 28/39, wherein the calculation model comprises a Markov model (Markov models are known in art and widely used for risk calculation, obvious variation to one of ordinary skill to use Markov model).
Regarding claims 31 and 42, Berger discloses the method/node of claim 28/39, wherein the calculated level of risk is a measure of improvement of the feature relative degradation of the performance (obvious to one of ordinary skill the risk is the improvement relative to degradation of performance).
Regarding claims 32 and 43, Berger discloses the method/node of claim 27/37, wherein the AI module for managing the feature is controlling sleep modes of a radio network node. Berger discloses sensing is combined with machine learning algorithm to enhance 5G RAN protocols and enhancement for a base station sleep mode aided by this sensing, Para [0028], sensing combined with ML algorithm to enhance power saving of the RAN, Para [0053].
Regarding claims 33 and 44, Berger discloses the method/node of claim 27/37, wherein the indication is received from another network node (obvious variation to one of ordinary skill in the art).
Regarding claims 34 and 45, Berger discloses the method/node of claim 27/37, wherein the AI module is deactivated when the obtained level of risk is above the set level of risk, and otherwise the AI module is activated (Quan discloses if accuracy is greater than threshold determine the trained machine learning model as the correct model, otherwise initiate new training iterations, Para [0111], in this case low accuracy means higher risk and not to use the model);
Regarding claim 35, Berger discloses the method of claim 27, wherein the AI module comprises a neural network model, a machine learning model, or a deep learning model (Quan discloses neural network model or deep learning, Para [0071]).
Regarding claims 36 and 46, Berger discloses the method/node of claim 27/46, wherein the obtained indication indicates a retraining of the AI module when differing from an actual indication (Quan discloses initiate new training iterations when accuracy is low, Para [0111]).
Regarding claim 38, Berger discloses the network node of claim 37, wherein the network node comprises a radio network node comprising the AI module. Berger discloses base station in RAN network with machine learning algorithm, Para [0053] having ML module would be obvious to one of ordinary skill.
Regarding claim 37, Berger discloses a network node for managing communication in a communication network, wherein the network node comprises processing circuitry and memory storing instructions for execution by the processing circuitry (base station supporting communications with UEs, Para [0004], processor and memory, Fig. 2), whereby the network node is configured to: but does not explicitly disclose wherein the level of risk is associated with a level of a degradation of a performance in the communication network. Berger discloses sensing is combined with machine learning algorithm to enhance 5G RAN protocols and enhancement for a base station sleep mode aided by this sensing, Para [0028], sensing combined with ML algorithm to enhance power saving of the RAN, Para [0053]. Rajkumar discloses training machine learning model with low quality information degrades model accuracy and degrades performance, Para [0163], obvious to one of ordinary skill using a low accuracy ML model would reduce performance in the network; and does not disclose obtaining an indication relating to a level of risk in using an artificial intelligence (AI) module for managing a feature (Quan discloses determine ratio of a count of positive results to the count of a second portion as the accuracy rate of the trained machine learning model, Para [0111]), comparing the obtained level of risk with a set level of risk (Quan discloses determine if accuracy rate is greater than a threshold (e.g. 90%), Para [0111]); and based on the comparison, activating or deactivating the AI module for managing the feature (Quan discloses if accuracy is greater than threshold determine the trained machine learning model as the correct model, otherwise initiate new training iterations, Para [0111]). It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to utilize the techniques taught by Quan and Rajkumar in the system of Berger in order to the utilize of machine learning algorithms, where only highly accurate models are used to prevent reduction of performance in the network.
Allowable Subject Matter
Claims 30 and 41 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.
Response to Arguments
Applicant's arguments filed 1/29/2026 have been fully considered but they are not persuasive. Applicant argues Berger and Rajkumar fail to disclose any of the steps of the method claim and apparatus claim. Applicant states the combination of the two references supposedly teach a poorly trained ML model degrades model accuracy and degrades performance and the office action concludes it is obvious using a low accuracy ML model would reduce network performance in the network. In response, Applicant doesn’t dispute the teaching or the conclusion of the Examiner and even states “that may be”. Applicant argues Rajkumar is not directed to use of a ML model but instead concerned with the training of the ML model. Applicant argues “using a low-accuracy model is probably not good” is a banal statement and Rajkumar doesn’t disclose activating or deactivating use of a model. In response, Rajkumar was not used to disclose the limitation of “using an artificial intelligence (AI) module” that was the Quan reference. Further the argument is wrong anyway, Rajkumar discloses “using a machine learning model” in Para [0010]. It is also not clear why “using a low-accuracy model is probably not good” is in quotation marks, as that is not a statement from the Examiner but one made up by the Applicant. Applicant appears to indirectly admit using a low accuracy ML model is not good (i.e. risky) by stating it is banal, meaning obvious, conclusion. Last, Rajkumar is not used to disclose activating or deactivating a ML model, as that limitation is disclosed by Quan and the argument is moot. Applicant argues over the step of “obtaining an indication of risk” and the office action points the accuracy rate of a trained ML model of Quan. Applicant explains the accuracy rate of Quan and states if the accuracy rate is not high enough, as determined by some arbitrary threshold (Applicant admits the comparison limitation here), the system does more training or starts training a new ML model. Applicant concludes accuracy rate is not a reasonable indication of a level of risk. In response, Applicant just makes a conclusion statement without actual argument. Accuracy is generally considered the inverse of risk, therefore an indication of risk. If the accuracy of the ML model is high, the risk of using the ML model will be lower. Applicant argues there is no suggestion the accuracy rate threshold is associated with a degradation of performance in a communication network because Quan isn’t concerned with communication network behavior. Applicant admits Berger uses ML model in a communication network but there is no explanation of how Quan’s accuracy testing for an ML model can be used in Berger. In response, Quan wasn’t used to disclose performance in the communication network but Berger was. Quan’s method is a blueprint for “gating” use of AL based on proven reliability and uses performance checks to mitigate risk, that is calculating accuracy of ML model, comparing to a threshold and toggle usage of the ML model accordingly, easily applicable to the ML model in Berger used to enhance power savings. Applicant repeats the same argument that accuracy is not an indication of risk. Applicant makes a conclusory statement, however high accuracy means less risk and low accuracy means higher risk, therefore accuracy is an inverse indication of risk. Applicant already admitted it was a banal conclusion that a poorly trained ML model (i.e. low accuracy) is not a good idea to use (i.e. risky). Using highly accurate ML models, leads to better performance of the ML models. Applicant argues against the dependent claims. Applicant argues the Examiner makes no attempt to find the limitation of “calculating the level of risk using a calculation model”. Applicant states the Examiner takes official notice but requests evidence be provided and the rejections are in error. In response, Applicant has no argument of Examiner’s error but merely asks evidence to be provided. According to MPEP 2144.03: "To adequately traverse a finding based on official notice, an applicant must specifically point out the supposed errors in the examiner’s action, which would include stating why the noticed fact is not considered to be common knowledge or well-known in the art. A mere request by the applicant that the examiner provide documentary evidence in support of an officially-noticed fact is not a proper traversal." Applicant fails to traverse the official notice. Applicant further argues over another dependent claim. Applicant admits the existence of Markov models is not disputed. Applicant argues the office action has not shown or explained how a Markov model might be applied to the prior art. Applicant makes a statement about the Quan reference which is unrelated. In response, Markov models are widely known to be used for risk calculation, by calculating the probability of transitioning between states over time. Markov models are well-established probabilistic tool for modeling stochastic processes, particularly in systems involving state transitions, uncertainty and time-dependent risks and commonly applied in network management and engineering systems. Markov models can simulate state changes due to AI induced errors making risk calculation intuitive and computationally efficient. Applicant already admits the well-known existence of Markov models and therefore the widely known usage in risk calculation.
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KEVIN CUNNINGHAM whose telephone number is (571) 272-1765. The examiner can normally be reached Monday through Thursday 7:30-18:00 (EST).
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Huy Vu can be reached on (571) 272-3155. The fax number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/KEVIN M CUNNINGHAM/Primary Examiner, Art Unit 2461
Prosecution Insights