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 .
This Office Action is in response to the Applicant Arguments/REMARKS correspondence filed on 03/27/2026.
Claims 1-20 are pending and rejected.
Response to Arguments
First, Applicant’s arguments, see Applicant Arguments/REMARKS, filed 03/27/2026, with respect to the rejection(s) of claims 1 under 35 USC 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of further search and inquiry.
Second, Applicant's arguments filed Applicant Arguments/REMARKS have been fully considered but they are not persuasive in regards to claims 1-12.
Applicant’s argument is not persuasive because the amendment merely adds that assigning values to the new tree child nodes “is performed without a random playout phase,” but Kartal is not limited to conventional MCTS random rollout techniques. Applicant’s argument attacks Kartal as though the Examiner were bodily incorporating the conventional random playout phase into Ouyang. However, the proper inquiry is not whether Kartal’s conventional MCTS embodiment is copied wholesale, but whether Kartal teaches or suggests MCTS/search-tree framework while avoiding or reducing random rollout evaluation.
Kartal teaches this concept. Kartal first describes standard MCTS as proceeding through “selection expansion, rollout, and backpropagation,” and states that standard MCTS uses random actions, i.e. rollouts, to estimate state-action values. (Kartal [0079]). Kartal then explains that MCTS can be combined with neural-network guidance, including AlphaGo-type approaches where “a policy network narrows down move selection” and “a value network helps with leaf evaluation,” thereby “reducing the number of costly rollouts.” (Kartal [0066], [0068]). More importantly, Kartal expressly teaches that AlphaGo Zero “still employed the skeleton of MCTS algorithm,” but “did not perform rollouts to evaluate lead states” and instead “only relied on the value network,” which provided a “significant gain in time-efficiency” because rollouts can be computationally demanding. (Kartal [0069]). Thus, Kartal teaches or at least suggests the concept of using an MCTS-based search structure without a random playout phase.
Applicant’s own Specification confirms that the added phrase is directed to only one distinction from standard MCTS. The Specification explains that standard MCTS includes selection, expansion, playout/rollout, and backpropagation, and that the conventional playout phase involves randomly sampling moves until a terminal state is reaches. (Applicant’s Spec [0102]-[0103]). The Specification then states that, in the CCO/network-cell optimization context, there is no true terminal state because any intermediate node may be a valid network configuration, and therefore randomly sampling new Core-cell configurations and submitting them to the network simulator would waste expensive computational resources. This reasoning led Applicant to eliminate the known MCTS playout phase. (Applicant’s Spec [0117]).
This is consistent with the interview summary. The interview summary indicated that Applicant could overcome the rejection by amending the claims to track the Specification’s specific departures from conventional MCTS. The summary identified elimination of random playout phase in Applicant’s spec [0117], but also separately suggested that Applicant may want to further amend the claims to reflect the predict-then-simulate two-stage evaluation described in Applicant’s spec [0124]-[0127], and {0135]-[0137]. Thos latter portions described using an MLM to predict KPI values for new child-node configurations, calculating reward-function values from the predicted KPIs, ranking/selecting promising child nodes, submitting only selected child nodes to the network simulator, calculating reward values from simulated KPIs, and then backpropagating the simulated reward information. (Applicant’s spec [0124]-[0127], [0135]-[0137]). Because Applicant only amended the claim to say “without a random playout phase,” the claim should not be read as requiring all of those additional predict-then-simulate details unless they are expressly recited elsewhere in the claim.
Accordingly, under the broadest reasonable interpretation consistent with the claim language and the specification, “without a random playout phase” excludes the conventional MCTS step of randomly rolling out from a new child node to a terminal state, but it does not exclude all MCTS-based or MCTS-inspired search-tree techniques. Nor does it require importing unclaimed features from the Specification, such as the complete MLM-based KPI prediction, ranking, selective simulator evaluation, and retraining framework. Kartal directly undermines Applicant’s position because Kartal itself recognizes non-conventional MCTS/DRL approaches in which neural-network guidance or value-network evaluation reduces or eliminates costly rollouts (Kartal [0066], [0068]-[0069]).
Therefore, the amendment does not overcome the rejection because “without a random playout phase” only excludes a conventional random rollout implementation, while Kartal expressly teaches MCTS-based approaches in which rollouts are reduced or omitted and nueral-network/value guidance is used instead. The interview summary indicated that Applicant could more clearly distinguish by amending to the Specification’s specific predict-then-simulate framework, but Applicant did not add those additional limitations. Accordingly, under BRI, the claim remains broad enough to read on a MCTS/search-tree process in which child-node values are assigned without random playout, as taught or suggested by Kartal’s MCTS/value-network teaches particularly Kartal [0066], [0068]-[0069], and [0079].
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 (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 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 1-12 are rejected under 35 U.S.C. 103 as being unpatentable over Ouyang et al (US20190239101A1) in view of Kartal et al (US20200143206A1) in further view of Mishra et al (US20180338250A1).
Regarding claim 1, Ouyang teaches a method, implemented by a data processing system, of adjusting modifiable parameters of network cells of a self-organizing cellular mobile communications network ([0076]-[0079], [0086]-[0092], cellular networks (eNb/cells) and a network reconfiguration system that optimizes and adjusts operating parameters automatically (SON behavior)), comprising:
retrieving a current configuration of network cells currently deployed on field, the current configuration of network cells including, for at least some of the network cells, modifiable parameters ([0075]-[0078], [0096]-[0099], describes collecting and storing current network data, KPIs, and operating parameters of deployed network nodes, which necessarily reflects the current configuration);
starting from the retrieved current configuration of the network cells, exploring different configurations of the network cells, each different configuration of the network cells differing from the retrieved current configuration of the network cells and from other different configurations of the network cells by a change in the value of at least one of the modifiable parameters of at least one network cell ([0087]-[0092], reconfiguration manager determines new values for network parameters (antenna, power, bandwidth, routing, etc.) i.e. explores alternative configurations differing by parameters values);
evaluating the explored different configurations of the network cells, wherein:
said exploring different configurations of the network cells comprises iterating the following steps ([0096]-[0102], KPIs and analytics are used to evaluate candidate configurations and determine their effectiveness):
predicting, by means of a machine learning model, values of network performance indicators corresponding to the different configurations of network cells that correspond to each of the new tree child nodes ([0095]-[0100], [0106]-[0112], analytics engine employs machine learning/deep learning models to predict network behavior and performance indicators (KPIs) based on configuration parameters), and
However, Ouyang fails to teach but Kartal teaches—
building a search tree in which each node of the search tree corresponds to a respective different configuration of the network cells ([0003]-[0006], explicit disclosure of Monte-Carlo tree search where nodes represent states reachable by actions, maps to each node corresponds to a configuration), said building a search tree comprising:
selecting a path of nodes of the search tree from a search tree root node to a search tree leaf node, wherein the nodes in the path are selected by means of a node selection policy ([0004]-[0006] Selection phase uses a node-selection policy (UCB/UCT) to traverse from root to leaf);
expanding the search tree by identifying if at least one new tree child node in respect of the search tree leaf node is available and, if available, assigning to each new tree child node a respective child node value, wherein said assigning values to the new tree child nodes is performed without a random playout phase and comprises ([0005]-[0006], [0066], [0068]-[0069], and [0079] expansion step explicitly adds a new child node corresponding to an unexplored action/state; recognizes non-conventional MCTS/DRL approaches in which neural-network guidance or value-network evaluation reduces or eliminates costly rollouts),
calculating values of a reward function in respect of the predicted values of the network performance indicators ([0006]-[0008] MCTS framework evaluates predicted outcomes via a reward/value associated with nodes, which directly corresponds to a reward function);
selecting, among the new tree child nodes, at least one new tree child node according to a ranking of the calculated values of the new tree child nodes ([0004]-[0006], Node selection is based on ranking according to UCB/UCT values balancing exploration and exploitation);
subjecting to simulation by a network simulator the selected at least one new tree child node for obtaining simulated values of the network performance indicators corresponding to the different configuration of network cells that corresponds to the selected at least one new tree child node ([0006]-[0008], Rollout/simulation phase explicitly simulates future outcomes to evaluate actions, inherently relying on a simulator);
calculating values of the reward function in respect of the simulated values of the network performance indicators ([0006]-[0008], after simulation, a score/reward is computed based on the simulated outcome);
updating a respective node information of the selected at least one new tree child node and of the search tree nodes back along the path from the selected at least one new tree child node to the search tree root node ([0006]-[0008], Explicit back-propagation of reward values from the simulated node back to the root node);
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches a Monte-Carlo Tree Search framework for exploring alternative system configurations by building a search tree, selection nodes using a defined node selection policy, expanding child nodes, performing simulation-based evaluations, calculating reward values, and back-propagating the reward values along the search path. A skilled artisan would have recognized that the search-tree-based planning and reward-driven evaluation of Kartal provides a well-known and efficient mechanism for systematically exploring and ranking alternative parameter configurations generated by the machine-learning based network optimization of Ouyang, thereby improving convergence speed and decision quality.
But Kartal fails to teach—
based on the calculated values of the reward function, assessing a goodness of the different configuration of network cells corresponding to the selected at least one new tree child node compared to the configuration of network cells corresponding to the search tree root node, and
when a suitable goodness is assessed for the al least one new tree child node, terminating said iterating and automatically deploying on field a new configuration of network cells by modifying one or more of the modifiable parameters of network cells.
However, Mishra teaches—
based on the calculated values of the reward function, assessing a goodness of the different configuration of network cells corresponding to the selected at least one new tree child node compared to the configuration of network cells corresponding to the search tree root node ([0034]-[0038], coordinating server determines that a configuration change is beneficial and sends instructions to update configuration parameters of base stations in the live network automatically), and
when a suitable goodness is assessed for the al least one new tree child node, terminating said iterating and automatically deploying on field a new configuration of network cells by modifying one or more of the modifiable parameters of network cells ([0034]-[0038], coordinating server determines that a configuration change is beneficial and sends instructions to update configuration parameters of base stations in the live network automatically).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 2, Ouyang fails to teach the method wherein said node selection policy is based on an Upper Confidence Bound—UCB—criterion.
However, Kartal teaches the method wherein said node selection policy is based on an Upper Confidence Bound—UCB—criterion ([0004]-[0006], explicitly discloses Monte-Carlo Tree Search using a UCB/UCT selection policy to select nodes during the selection phase, directly matching a node selection policy based on UCB).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 3, Ouyang fails to teach the method wherein updating a respective node information comprises updating a node average value and a node visit count, said node average value being calculated by averaging the calculated values of the reward function with the previously calculated reward function values in respect of the node.
However, Kartal teaches the method wherein updating a respective node information comprises updating a node average value and a node visit count, said node average value being calculated by averaging the calculated values of the reward function with the previously calculated reward function values in respect of the node ([0006]-[0008], describes back-propagation in MCTS where, for each visited node, the reward value is accumulated/average and the visit count is incremented, exactly corresponding to updating node average value and node visit count).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 4, Ouyang teaches the method wherein said network performance indicators comprise at least one of the following indicators:
number of overtarget cells that are experiencing a traffic load higher than a predetermined threshold traffic load ([0096]-[0102], [0106]-[0112], explicitly discloses KPIs including traffic load, cell throughput, and user throughput, used by the analytics engine to evaluate network performance, covering the listed indicators);
total cell traffic that exceeds a predetermined overtarget total traffic threshold;
throughput guaranteed by the cells ([0096]-[0102], [0106]-[0112], explicitly discloses KPIs including traffic load, cell throughput, and user throughput, used by the analytics engine to evaluate network performance, covering the listed indicators);
throughput offered to users of the cellular mobile communications network ([0096]-[0102], [0106]-[0112], explicitly discloses KPIs including traffic load, cell throughput, and user throughput, used by the analytics engine to evaluate network performance, covering the listed indicators).
Regarding claim 5, Ouyang teaches the method wherein said reward function is a linear combination of two or more of said network performance indicators ([0098]-[0102], teaches computing composite KPI scores derived from multiple performance indicators to rank or evaluate configurations, which inherently corresponds to a linear (weighted) combination of multiple KPIs forming a reward/value metric).
Regarding claim 6, Ouyang and Kartal fail to teach the method wherein said network simulator is a simulator of the type used in a planning phase of mobile communications network, configured to simulate propagations of radio signals taking into account a description of the geographic area covered by the network cells.
However, Mishra teaches the method wherein said network simulator is a simulator of the type used in a planning phase of mobile communications network, configured to simulate propagations of radio signals taking into account a description of the geographic area covered by the network cells ([0026]-[0031], [0036]-[0038], discloses network planning/optimization tools that evaluate configurations with awareness of geographic areas and radio conditions, consistent with planning-phase network simulators used to simulate radio propagation over geography).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 7, Ouyang and Kartal fail the method wherein said exploring and evaluating different configurations of the network cells are performed off-line in background, while the mobile communications network continues operating according to the current configuration of network cells.
However, Mishra teaches the method wherein said exploring and evaluating different configurations of the network cells are performed off-line in background, while the mobile communications network continues operating according to the current configuration of network cells ([0036]-[0039], [0042]-[0044], explicitly discloses that optimization and evaluation can be performed offline/background at a coordinating serving while the live network continues, prior to deployment of updated parameters).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 8, Ouyang fails to teach the method wherein said exploring different configurations of the network cells comprises, at each iteration or at least every predetermined number of iterations:
building the search tree by selecting as a new search tree root node one of the new tree child nodes identified in previous iterations and which, based on the calculated values of the reward function in respect of the simulated values of the network performance indicators, has an assessed goodness higher than a goodness of the previous search tree root node selected in previous iterations.
However, Kartal teaches the method wherein said exploring different configurations of the network cells comprises, at each iteration or at least every predetermined number of iterations:
building the search tree by selecting as a new search tree root node one of the new tree child nodes identified in previous iterations and which, based on the calculated values of the reward function in respect of the simulated values of the network performance indicators, has an assessed goodness higher than a goodness of the previous search tree root node selected in previous iterations ([0004]-[0008], MCTS inherently updates the root policy over iterations by favoring nodes with higher reward values, effectively re-centering the search on better-performing child nodes across iterations, corresponding to selecting a new root with higher assessed goodness).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 9, Ouyang teaches the method wherein said modifiable parameters of network cells comprise one or more of: cell transmission power, electrical tilt of the cell antenna(s), azimuth of the cell antenna(s), parameters affecting a radiation diagram and a power spatial distribution, and parameters controlling radiation patterns for active antennas ([0087]-[0092], explicitly lists modifiable radio/network parameters such as transmit power, antenna parameters and radiation characteristics, directly matching the claimed list).
Regarding claim 10, Ouyang and Kartal fail to teach the method comprising identifying a geographic area of interest covered by the mobile communications network, wherein identifying the geographic area of interest comprises classifying network cells covering the geographic area of interest in a first class of cells having modifiable parameters that are modifiable by the self-organizing network, and a second class of cells in the neighborhood of the cells of the first class and whose parameters are not modifiable.
However, Mishra teaches the method comprising identifying a geographic area of interest covered by the mobile communications network, wherein identifying the geographic area of interest comprises classifying network cells covering the geographic area of interest in a first class of cells having modifiable parameters that are modifiable by the self-organizing network, and a second class of cells in the neighborhood of the cells of the first class and whose parameters are not modifiable ([0030]-[0035], discloses identifying target cells for optimization and distinguishing them from neighboring cells whose parameters are constrained or not modified, corresponding to classification into first and second classes within a geographic area).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 11, Ouyang teaches the method wherein said machine learning model comprises a regressor having a loss function selected among: mean squared error, mean absolute error, Huber loss, quantile loss ([0106]-[0112], discloses use of machine-learning/deep learning models trained using loss functions to minimize prediction error for KPIs, such regressors inherently use standard loss functions (MSE/MAE) supporting the claimed regressor-based ML model).
Regarding claim 12, Ouyang teaches a data processing system configured for automatically adjusting modifiable parameters of network cells of a self-organizing cellular mobile communications network ([0076]-[0079], [0086]-[0092], cellular networks (eNb/cells) and a network reconfiguration system that optimizes and adjusts operating parameters automatically (SON behavior)), the system comprising:
processing circuitry configured to perform the method of claim 1 ([0006]-[0008] MCTS framework evaluates predicted outcomes via a reward/value associated with nodes, which directly corresponds to a reward function).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Claim 13 are rejected under 35 U.S.C. 103 as being unpatentable over Ouyang et al (US20190239101A1) in view of Kartal et al (US20200143206A1) in further view of Mishra et al (US20180338250A1), in further view of Magnuson et al “Monte Carlo Tree Search and its Applications” (2015).
Regarding claim 13, Ouyang, Kartal and Mishra fail to teach but Magunson teaches but wherein building the search tree comprises, when a value of the reward function of a new child node added to the search tree is higher than a value of the search tree root node: halting the search, discarding the search tree, and starting a new search tree with the new child node at its root (pg. 2 Section 2.1 The Tree Structure, “starting a new search tree…”, “discarding the search tree”, Section 2.2. pg. 3 MCTS split in selection, expansion, backpropagation etc., pg. 6 Section 4.5 Tree Pruning—“halting the search”, “discarding the search tree” and “starting a new search tree…”; teaches MCTS tree pruning and re-rooting based on the selected/highest-valued child-node; specifically teaches that each node has an associated value from simulations and that MCTS chooses the node with the greatest estimated value; further teaches that once MCTS decides which move to make, the chosen child node becomes the new root node and its siblings are discarded; in tree-pruning implementation—explains that after a predefined number of iterations, the algorithm chooses the node with the greatest potential value from the child nodes of the previously chosen action, the chosen node effectively becomes the new root node, and all the siblings of that node along with their subtrees are discarded; teaches the claimed concept of stopping the current search, pruning/discarding less promising portions of the existing tree, and continuing the search with the selected higher-valued child node as the root).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. Lastly, Magunson teaches MCTS tree pruning and re-rooting based on the selected/highest-valued child-node. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Claims 14-20 are rejected under 35 U.S.C. 103 as being unpatentable over Ouyang et al (US20190239101A1) in view of Kartal et al (US20200143206A1) in further view of Mishra et al (US20180338250A1), in further view of Drori et al “Automatic Machine Learning by Pipeline Synthesis using Model-Based Reinforcement Learning and a Grammar” (2019).
Regarding claim 14, Ouyang, Kartal and Mishra fail to teach but Drori teaches the method of claim 1, comprising retraining the machine learning model when a predetermined number N of new network configurations are gathered (pg. 2-3 & 5, Section 2 Methods & Section 2.1, section 3.2 pre-trained Model/Results, teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation; formulates pipeline synthesis as a single-player game in which a pipeline is a state, an action modifies the pipeline into a new state, and pipeline performance is the reward; Drori’s neural network receives s state and outputs both action probabilities and a value approximating the actial evaluation of the state, and the model is trained by minimizing error between predicted performance and actual evaluation; further teaches using MCTS with a UCB rule that includes visit counts and selecting the action/state that maximizes the UCB value, adding the new state to the tree with the neural-network estimates; generated pipeline is then applied to the data to obtain actual evaluation, thereby teaching selective actual evaluation of promising configurations identifies by neural-network guided MCTS; supports retraining/updating an ML model based on new actual evaluations, visit-count based MCTS selection, selecting a highest-valued child/state for evaluation, and initializing new node values using predicted reward/evaluation).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. Lastly, Drori teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 15, Ouyang, Kartal and Mishra fail to teach but Drori teaches the method of claim 14, wherein N = 1 (pg. 2-3 & 5, Section 2 Methods & Section 2.1, section 3.2 pre-trained Model/Results, teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation; formulates pipeline synthesis as a single-player game in which a pipeline is a state, an action modifies the pipeline into a new state, and pipeline performance is the reward; Drori’s neural network receives s state and outputs both action probabilities and a value approximating the actial evaluation of the state, and the model is trained by minimizing error between predicted performance and actual evaluation; further teaches using MCTS with a UCB rule that includes visit counts and selecting the action/state that maximizes the UCB value, adding the new state to the tree with the neural-network estimates; generated pipeline is then applied to the data to obtain actual evaluation, thereby teaching selective actual evaluation of promising configurations identifies by neural-network guided MCTS; supports retraining/updating an ML model based on new actual evaluations, visit-count based MCTS selection, selecting a highest-valued child/state for evaluation, and initializing new node values using predicted reward/evaluation).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. Lastly, Drori teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 16, Ouyang, Kartal and Mishra fail to teach but Drori teaches the method of claim 1, wherein during expansion, a visit count of each of the new tree child nodes is set to one (pg. 2-3 & 5, Section 2 Methods & Section 2.1, section 3.2 pre-trained Model/Results, teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation; formulates pipeline synthesis as a single-player game in which a pipeline is a state, an action modifies the pipeline into a new state, and pipeline performance is the reward; Drori’s neural network receives s state and outputs both action probabilities and a value approximating the actial evaluation of the state, and the model is trained by minimizing error between predicted performance and actual evaluation; further teaches using MCTS with a UCB rule that includes visit counts and selecting the action/state that maximizes the UCB value, adding the new state to the tree with the neural-network estimates; generated pipeline is then applied to the data to obtain actual evaluation, thereby teaching selective actual evaluation of promising configurations identifies by neural-network guided MCTS; supports retraining/updating an ML model based on new actual evaluations, visit-count based MCTS selection, selecting a highest-valued child/state for evaluation, and initializing new node values using predicted reward/evaluation).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. Lastly, Drori teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 17, Ouyang, Kartal and Mishra fail to teach but Drori teaches the method of claim 1, wherein terminating said iterating occurs after a predetermined number of iterations where no improvements in the reward function value are obtained (pg. 2-3 & 5, Section 2 Methods & Section 2.1, section 3.2 pre-trained Model/Results, teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation; formulates pipeline synthesis as a single-player game in which a pipeline is a state, an action modifies the pipeline into a new state, and pipeline performance is the reward; Drori’s neural network receives s state and outputs both action probabilities and a value approximating the actial evaluation of the state, and the model is trained by minimizing error between predicted performance and actual evaluation; further teaches using MCTS with a UCB rule that includes visit counts and selecting the action/state that maximizes the UCB value, adding the new state to the tree with the neural-network estimates; generated pipeline is then applied to the data to obtain actual evaluation, thereby teaching selective actual evaluation of promising configurations identifies by neural-network guided MCTS; supports retraining/updating an ML model based on new actual evaluations, visit-count based MCTS selection, selecting a highest-valued child/state for evaluation, and initializing new node values using predicted reward/evaluation).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. Lastly, Drori teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 18, Ouyang, Kartal and Mishra fail to teach but Drori teaches the method of claim 1, wherein assessing the goodness comprises determining a total network traffic constraint such that a total network traffic of the different configuration of network cells corresponding to the selected at least one new tree child node is greater than or equal to a total network traffic of the current configuration of network cells (pg. 2-3 & 5, Section 2 Methods & Section 2.1, section 3.2 pre-trained Model/Results, teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation; formulates pipeline synthesis as a single-player game in which a pipeline is a state, an action modifies the pipeline into a new state, and pipeline performance is the reward; Drori’s neural network receives s state and outputs both action probabilities and a value approximating the actial evaluation of the state, and the model is trained by minimizing error between predicted performance and actual evaluation; further teaches using MCTS with a UCB rule that includes visit counts and selecting the action/state that maximizes the UCB value, adding the new state to the tree with the neural-network estimates; generated pipeline is then applied to the data to obtain actual evaluation, thereby teaching selective actual evaluation of promising configurations identifies by neural-network guided MCTS; supports retraining/updating an ML model based on new actual evaluations, visit-count based MCTS selection, selecting a highest-valued child/state for evaluation, and initializing new node values using predicted reward/evaluation).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. Lastly, Drori teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 19, Ouyang, Kartal and Mishra fail to teach but Drori teaches the method of claim 1, wherein selecting the at least one new tree child node comprises: selecting a single new tree child node having a highest calculated value among the new tree child nodes, such that the simulation by the network simulator is restricted to the single new tree child node (pg. 2-3 & 5, Section 2 Methods & Section 2.1, section 3.2 pre-trained Model/Results, teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation; formulates pipeline synthesis as a single-player game in which a pipeline is a state, an action modifies the pipeline into a new state, and pipeline performance is the reward; Drori’s neural network receives s state and outputs both action probabilities and a value approximating the actial evaluation of the state, and the model is trained by minimizing error between predicted performance and actual evaluation; further teaches using MCTS with a UCB rule that includes visit counts and selecting the action/state that maximizes the UCB value, adding the new state to the tree with the neural-network estimates; generated pipeline is then applied to the data to obtain actual evaluation, thereby teaching selective actual evaluation of promising configurations identifies by neural-network guided MCTS; supports retraining/updating an ML model based on new actual evaluations, visit-count based MCTS selection, selecting a highest-valued child/state for evaluation, and initializing new node values using predicted reward/evaluation).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. Lastly, Drori teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
Regarding claim 20, Ouyang, Kartal and Mishra fail to teach but Drori teaches the method of claim 1, wherein assigning to each new tree child node the respective child node value comprises: initializing the respective child node value with the calculated value of the reward function in respect of the predicted values of the network performance indicators (pg. 2-3 & 5, Section 2 Methods & Section 2.1, section 3.2 pre-trained Model/Results, teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation; formulates pipeline synthesis as a single-player game in which a pipeline is a state, an action modifies the pipeline into a new state, and pipeline performance is the reward; Drori’s neural network receives s state and outputs both action probabilities and a value approximating the actial evaluation of the state, and the model is trained by minimizing error between predicted performance and actual evaluation; further teaches using MCTS with a UCB rule that includes visit counts and selecting the action/state that maximizes the UCB value, adding the new state to the tree with the neural-network estimates; generated pipeline is then applied to the data to obtain actual evaluation, thereby teaching selective actual evaluation of promising configurations identifies by neural-network guided MCTS; supports retraining/updating an ML model based on new actual evaluations, visit-count based MCTS selection, selecting a highest-valued child/state for evaluation, and initializing new node values using predicted reward/evaluation).
It would have been obvious to a person of ordinary skill in the art at the time of the invention to combine Ouyang, which teaches a self-organizing cellular network in which a data processing system applies machine-learning-based analytics to predict network performance indicators and to adjust modifiable operating parameters of deployed network cells based on such predictions, with Kartal, which teaches exploring alternative system configurations using a search tree with node selection policies, simulation-based evaluation, reward calculation, and back-propagation, and further in view of Mishra, which explicitly teaches a coordinating server that, upon determining that a configuration change is beneficial, automatically transmits instructions to update configuration parameters of base stations in an operational cellular network without human intervention. Lastly, Drori teaches the claimed ML-guided MCTS child-node value assignment and selective evaluation. A skilled artisan would have been motivated to incorporate the explicit deployment and closed-loop configuration update mechanism of Mishra into the combined system of Ouyang and Kartal in order to ensure that the optimal configuration identifies through the tree-based simulation and reward evaluation is automatically applied to the live network, thereby improving operational efficiency and responsiveness.
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 MICHAEL WILLIAM ABBATINE whose telephone number is (571)272-0192. The examiner can normally be reached Monday-Friday 0830-1700 EST.
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, Nishant Divecha can be reached at (571) 270-3125. 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.
/MICHAEL WILLIAM ABBATINE JR./Examiner, Art Unit 2419
/Nishant Divecha/Supervisory Patent Examiner, Art Unit 2419
Prosecution Insights