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 1-10, 13-22 are pending.
Amendments of 03/06/2026 are entered.
Response to Arguments
Arguments presented in the Remarks 03/06/2026 have been fully considered but they are not persuasive.
In response to Reason 1/1A/1B, i.e. optimization vs. diagnosis distinction; input/ouput; simulation to training data.
Applicant’s sole reliance on dictionary definition of “optimization” and “diagnose” is unpersuasive. In the context of Liao’s Self-Organization Networks and DRL network management, optimization intrinsically encompasses diagnosis, that is to say it means detecting performance weaknesses, potential suboptimal parameters, interference, various options via KPIs/metrics and perform corrective actions. Liao explicitly discloses such a system that learns from network measurement, keeps tracks of costs, and performance to perform optimization actions, decisions (outputs) with system performance indicators/KPIs, as well as analysis of parameters to avoid negative outcomes (interference, suboptimal performance) in at least ¶0061-0062, --17, 0088, 0097). The optimization process generates performance metrics that serve as feedback/rewards for iterative model improvement, which requires analysis of feedback and identifies issues and solution, thus solution to improvement, thus perfectly fall within BRI of the term diagnosis.
Furthermore the claim’s diagnosis model is by name only. It lacks any specific structures to distinguish it. Applicant’s attempt to separate “improvement when nothing is broken” from “root cause of failures” rings hollow as the claim provides no such specificity, nor does the act of diagnosis requires a failure event necessarily. Liao’s system framework handles both proactive and reactive correction based on degraded states (i.e. antenna tilts example, feedbacks loop from ¶0060-0066).
Applicant further argues claim 1 requires “injecting the action” into simulated network to generate output of the simulation” including network performance metrics, which are then transformed into training data for diagnosis. This allegedly associates particular actions with degraded performance metrics to predict root cause, thus requires specific inputs to the diagnosis model.
The argument is not persuasive. Claim 1 recites no specific input to the diagnosis model nor does it require the model to predict root cause. Again, the model is “diagnosis” in name only and these requirements mandated by Applicant are no where to be seen in the claim except from applicant’s own conjecture and perhaps from the Specification. Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993).
In response to reason 2, Applicant argues Liao operates only at part/sub network level while the claim allegedly require a model for the target network. This is not a meaningful distinction by at least the fact that, again, the claim does not offer any scope and structures for the claimed system, plus there is no evidence the model as trained specifically target every parts of said system. Liao in at least 0089 broadly defines system, parts, or network element to include base stations, sub-networks, partial networks, or a plurality of cells – which under BRI encompasses a target network as claimed.
In response to reason 3, Liao’s disclosure (for example, ¶094-0095) are limited as Applicant asserts. They are directed to a complex system-level simulators that mimic real network scenarios for training DRL optimization models, including cases without a pre-trained model. This directly supports selecting parameters, executing simulations, generating performance metrics, and using output for model training/fine-tuning. The “part” language does not exclude application to a target network under BRI, and the simulation purpose of Liao aligns with the claimed process. Applicant’s overly narrow reading of Liao ignores the full disclosure.
In response to reason 4, i.e. charge of out-of-context use of ¶0065-0066 and related paragraphs. The argument is not persuasive as they are mere general allegation without specifically pointing a particular limitation being mapped.
Furthermore, the Office Action’s citations of the above paragraphs are proper and contextual. These describe learning with feedback measurements in broader DRL/SON framework, illustrating iterative simulation of actions (i.e. antenna tilt adjustment, etc.) performance metrics as feedback, and training updates, which are exactly the kind of system as claimed. They are not isolated but exemplary of the simulation-to-training in Liao. The mapping is thus appropriate and not taken out of context.
In response to arguments of claims 4, 17, and 21:
The arguments are not persuasive.
Liao discloses determining a similarity measure between the training data deprived from the model and real world data as part of the similarity analysis process (¶0135-0141). This similarity analysis determines whether the model is sufficient to deploy, which directly reads on the claimed step of determining a similarity measure between training data and real world data. See also ¶0118-0122.
In contrary to Applicant’s narrow reading, Liao’s similarity analysis is not limited to parts of the network/system in a way that excludes comparison of simulation data and real world data. The reference explicitly use similarity between sources and target network parts/environments, including measurements and performance data, to decide model deployment, adaptation etc. KPIs serve as the basis for rewards in the DRL training loop, and the measured similarity level facilitates reward and next action/decision. In other words, Liao’s teaching are not limited to network-part similarity. They encompass comparison involving simulation/training data and real-world target data for corrective decisions (see at least ¶0089-0095).
Claim Rejections - 35 USC § 102
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 (i.e., changing from AIA to pre-AIA ) 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 the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-10, 13-22 is/are rejected under 35 U.S.C. 102(a)(2) as being unpatentable over Liao et al. (US 2021/0219384).
As to claim 1:
Liao discloses:
A method of a training manager for generating diagnosis models for mobile networks (Abstract, ¶0078, 0097, ‘agent’ read as the training manager, which perform training of DRL model(s)) for mobile communication network) the method comprising:
selecting, automatically by the training manager, a set of parameters for an action to be simulated in a simulated network where the simulated network replicates a target network for a diagnosis model;
(¶0097, 0099 “ agent selects and executes actions according to a greedy policy based on Q. It is possible to use histories of an arbitrary length as inputs, or a fixed length representation of histories can be employed. In the present example, a model is used which is based on a DRL network to learn successful policies directly from e.g. sensory inputs for the purpose of network optimization.” ‘Action’, ‘Policy’, ‘history’, and ‘input’s as chosen by the agent are represented by parameters/values in computing environment context. “mapping between the network environment measurements as inputs and the optimization actions/decisions as outputs” – measurements are taken and output optimization actions, which show parameters are taken to produce learned change. ¶0094, the system is a simulated network that mimic a real world scenario processed by a model)
executing a simulation of an operation of the simulated network based on the set of parameters of the action to generate an output of the simulation including a set of network performance metrics;
(¶0094, training/operation by the network is to mimic a real world scenario (i.e. simulation. ¶0097, “a DRL network to learn successful policies directly from e.g. sensory inputs for the purpose of network optimization. Specifically, the employed DRL model learns the mapping between the network environment measurements as inputs and the optimization actions/decisions as outputs”, i.e. using inputs/chosen actions/policies to produce output. ¶0065-0066, which shows examples of using sets of real world measurements to simulate operations to output modifications for improving parameters of antennas for performance optimization. ¶0132, 0118-0121, outputs to be compared with existing system include various measurement results of network condition that characterize network optimization result)
transforming the output of the simulation into training data for the diagnosis model; training the diagnosis model with the training data;
(See ¶0066, “ implement a tilt searching algorithm that modifies a network's antenna tilts iteratively based on feedback measurements so as to optimize jointly the uplink and downlink performance of coverage and capacity” – i.e. iterative training process, meaning outputs/results of each iteration (each simulation) are feedbacks to improve for the next iteration, namely used as training data. See ¶0122-0124, which discusses simulator-derived input/output/rewards are fed into the training loss and update process, i.e. transforming outputs into training signal/data. ¶0094, “real measurements in the new network part are used for fine-tuning”. See further at least 0095, 0097, 0104, for discussion on training/retraining the model using training data from the previous iteration’s outputs )
and outputting the diagnosis model for the target network, in response to the diagnosis model meeting a designated quality threshold.
(¶0141, ¶0134, 0135, outputting the model for use when the model is deemed as “sufficient”, based on a similarity analysis which deems the similarity level reaches a” minimum level” to an existing system)
As to claim 13 and 14:
Liao discloses:
A non-transitory machine-readable medium comprising computer program code which when executed by a computer carries out a set of operations for a training manager for generating diagnosis models for mobile networks, (¶0208, CRM) AND
a system of one or more electronic devices, comprising: the non-transitory machine-readable storage medium having stored therein the training manager; and a processor coupled to the non-transitory machine-readable storage medium, (See ¶0205-0208, a system having processor/memory storing CRM) the processor to execute the training manager,
the training manager to select, automatically, a set of parameters for an action to be simulated in a simulated network where the simulated network replicates a target network for a diagnosis model,
(¶0097, 0099 “ agent selects and executes actions according to a greedy policy based on Q. It is possible to use histories of an arbitrary length as inputs, or a fixed length representation of histories can be employed. In the present example, a model is used which is based on a DRL network to learn successful policies directly from e.g. sensory inputs for the purpose of network optimization.” ‘Action’, ‘Policy’, ‘history’, and ‘input’s as chosen by the agent are represented by parameters/values in computing environment context. “mapping between the network environment measurements as inputs and the optimization actions/decisions as outputs” – measurements are taken and output optimization actions, which show parameters are taken to produce learned change. ¶0094, the system is a simulated network that mimic a real world scenario processed by a model)
execute a simulation of an operation of the simulated network based on the set of parameters of the action to generate an output of the simulation including a set of network performance metrics,
(¶0094, training/operation by the network is to mimic a real world scenario (i.e. simulation. ¶0097, “a DRL network to learn successful policies directly from e.g. sensory inputs for the purpose of network optimization. Specifically, the employed DRL model learns the mapping between the network environment measurements as inputs and the optimization actions/decisions as outputs”, i.e. using inputs/chosen actions/policies to produce output. ¶0065-0066, which shows examples of using sets of real world measurements to simulate operations to output modifications for improving parameters of antennas for performance optimization. ¶0132, 0118-0121, outputs to be compared with existing system include various measurement results of network condition that characterize network optimization result)
transform the output of the simulation into training data for a diagnosis model, train the diagnosis model with the training data,
(See ¶0066, “ implement a tilt searching algorithm that modifies a network's antenna tilts iteratively based on feedback measurements so as to optimize jointly the uplink and downlink performance of coverage and capacity” – i.e. iterative training process, meaning outputs/results of each iteration (each simulation) are feedbacks to improve for the next iteration, namely used as training data. See ¶0122-0124, which discusses simulator-derived input/output/rewards are fed into the training loss and update process, i.e. transforming outputs into training signal/data. ¶0094, “real measurements in the new network part are used for fine-tuning”. See further at least 0095, 0097, 0104, for discussion on training/retraining the model using training data from the previous iteration’s outputs )
and output the diagnosis model for the target network, in response to the diagnosis model meeting a designated quality threshold.
(¶0141, ¶0134, 0135, outputting the model for use when the model is deemed as “sufficient”, based on a similarity analysis which deems the similarity level reaches a” minimum level” to an existing system)
As to claims 2, 15, and 19:
Liao discloses all limitations of claim 1/13/14, further comprising: classifying a condition of the simulated network after executing the simulation based on simulated measurements from the simulation. (¶0097, 0099, 0100, 0121, system performance results, i.e. outputs, are mapped to rewards, which indicates an act of classifying performance evaluation of a simulation, i.e. positive or negative, as well as the degrees of rewards)
As to claims 3, 16, 20:
Liao discloses all limitations of claim 2/15/19, wherein the classifying further comprises: determining a state of a configuration parameter for an entity of interest in the set of parameters of the action. (See ¶0122, determining θ, which is the parameter set characterizing the DRL network, i.e. entity of interest, (e.g., weight matrices and bias vectors). The DRL network can be trained e.g. by adjusting the parameter θ.)
As to claims 4, 17, 21:
Liao discloses all limitations of claim 1/13/14, further comprising: determining a similarity measure between the training data and real world data (¶0141, ¶0134, 0135, outputting the model for use when the model is deemed as “sufficient”, based on a similarity analysis which deems the similarity level reaches a” minimum level” to an existing system’s performance (i.e. real world data));
and calculating a reward for selecting a next action to be simulated based on an evaluation of a performance of the diagnosis model and the similarity measure. (¶0118-0122, reward(s) is/are determined, and are used as basis to determine the next iteration of adjustment of the model under training. 0141, ¶0134, 0135, similarity measures are used to judge whether the model is ready for deployment or for further tuning)
As to claim 5:
Liao discloses all limitations of claim 4, wherein a teacher model of the training manager selects the next action to be simulated based on the reward. (See ¶0099, 0088, based on the “reward”, training agent select action for optimization that maximize reward for next iteration of update)
As to claim 6:
Liao discloses all limitations of claim 4, wherein the similarity measure is determined as a density of real measurement neighbors of a simulated measurement from the simulation. (See ¶0132-0133, “a similarity between two systems is defined on the basis of network properties, such as location, geographical features, mobility patterns, and data demand statistics. Basically, there are two types of network properties: single data point (can be multi-dimensional) such as location or size of the network, and a statistical measure (e.g., histogram approximating probability density functions) such as mobility pattern or data demand distribution. The similarity between corresponding single data points (together composing e.g. a high dimensional vector) can be computed, for example, by suitable distance measures”)
As to claim 7:
Liao discloses all limitations of claim 1, wherein the set of parameters includes a selection of a network component to change, a configuration parameter to change, degree of change to the configuration parameter, and a traffic profile to test. (¶0066, 0062, “new base station”, i.e. network component , also antenna tilt, emission power and adjustment per iteration, ¶0038, “a subset of parameters and hyperparameters defining low and medium layers of the pre-trained network optimization model may be selected”)
As to claim 8:
Liao discloses all limitations of claim 1, wherein transforming the output into the training data further comprises: organizing deployment settings of the simulated network into a first matrix indicating whether a location includes a base station, a second matrix indicating traffic level in the location, and a third matrix indicating a clutter class for the location. (See ¶0103—0108, training data include forming a multi-dimensional matrix having multiple sub-matrixes, wherein “a matrix expressing the user activity/demand map is generated where the intensity at each pixel is the measure of the user activities”, i.e. traffic level. “matrix can also be extended to a multi-channel matrix to capture more network environment information, such as signal strength radio map”, which can indicate whether a location include a base station, i.e. a general area with strongest signal strength indicate presence of a base station, whereas a place without any signal strength or weaker values indicate the lack or further distance from a base station. A matrix is generated where each location is associated with represent data rate demand or quality of service class identifiers, i.e. clutter class of the location)
As to claim 9:
Liao discloses all limitations of claim 1, further comprising: applying the diagnosis model to the target network. (See ¶0143-0146, 0178, the trained model is provisioned to the real world system and deployed)
As to claims 10, 18, 22:
Liao discloses all limitations of claim 1/13/14, wherein a teacher model of the training manager is updated to select actions based on associated rewards and can be utilized for subsequent training of additional diagnosis models. (See ¶0099, 0088, based on the “reward”, training agent select action for optimization that maximize reward for next iteration of update. ¶0089, training a plurality of models)
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US 2016/0205697 - The strengths of alternative self-organizing-network (SON) techniques can be leveraged by deriving a compromise result from alternative results generated by the respective SON techniques. In particular, the compromise result may be derived from the alternative results based on reputations assigned to alternative SON techniques used to generate the respective results. The compromise result may be calculated based on weighted averages of the alternative results (e.g., solutions, diagnoses, predicted values, etc.), or on weighted averages of parameters specified by the alternative results (e.g., parameter adjustments, underlying causes, KPI values, etc.). In such an embodiment, the weights applied to the alternative results may be based on the reputations of the corresponding SON techniques used to generate the respective alternative results..
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 QUAN M HUA whose telephone number is (571)270-7232. The examiner can normally be reached 10:30-6:30.
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, Anthony Addy can be reached at 571-272-7795. 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.
/QUAN M HUA/Primary Examiner, Art Unit 2645
Prosecution Insights