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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR
1.17(e), was filed in this application after final rejection. Since this application is eligible for continued
examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the
finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's
submission filed on 12/22/2025 has been entered.
Response to Amendment
The amendments and arguments filed 12/22/2025 have been entered. Claims 1 remain pending in the application.
Applicant’s amendments and arguments, with respect to claim rejections of claims 1-20 under 35 U.S.C 101 filed 05/26/2026 have been considered and they are persuasive. Therefore, the previous rejections have been removed.
Applicant’s amendments and arguments, with respect to claim rejections of claims 1-20 under 35 U.S.C 103 filed 05/26/2026 have been considered some of them are persuasive.
Applicant argues that Ploennigs does not teach generating an executable modified version of the AI metamodel based on target-node details, because Ploennigs merely generates/uses a knowledge graph to configure or enrich a conversational agent and does not generate a node-specific executable metamodel according to the target node’s capabilities or resource usage. Applicant argues that Ploennigs does not teach pruning, based on the target-node details, concepts and relationships of a knowledge graph to obtain a pruned knowledge graph tailored to the target node prior to deployment, nor using that pruned knowledge graph to generate the executable modified version. Applicant also argues that Ploennigs does not teach deploying the executable modified version to the target node for execution, because Ploennigs’ linking/enriching of an IoT system is different from generating and deploying a customized executable metamodel to the node itself.
Applicant further argues that Fenoglio does not cure Ploennigs’ deficiencies because, although Fenoglio obtains information such as CPU/resource loads, that information is allegedly used for network/QoE analysis, rather than for pruning a knowledge graph based on node resources, generating an executable modified metamodel, or deploying that metamodel to the target node. Applicant argues that the combination of Ploennigs and Fenoglio as a whole still fails to teach the amended sequence of obtaining target-node resource information. Finally, Applicant argues that the Office’s previous motivation to combine Ploennigs and Fenoglio does not provide a reason to perform the newly claimed resource-aware KG pruning and deployment operations and that modifying the references in the asserted manner would require impermissible hindsight.
With regard to Applicant’s arguments that Fenoglio does not cure the deficiencies of Ploennigs because Fenoglio does not teach generating a node-specific executable metamodel by pruning a knowledge graph based on target-node details or deploying such executable metamodel to the target node for execution. The argument is persuasive only to the extent discussed below. Fenoglio nevertheless expressly teaches an artificial-intelligence architecture having symbolic layer and sub-symbolic layer at paragraph 74 “At the lowest layer of hierarchy 500 is sub-symbolic layer 502 that processes the multimodal network data 512 collected from the network under scrutiny. For example, at the core of sub-symbolic layer 502 may be one or more DNNs 508 or other machine learning-based model that processes the multimodal network data 512 collected from the network”, and paragraph 78 “At the top of hierarchy 500 may be symbolic layer 506 that may leverage symbolic learning to perform the functions described herein. In general, symbolic learning includes a set of symbolic grammar rules specifying the representation language of the system, a set of symbolic inference rules specifying the reasoning competence of the system, and a semantic theory containing the definitions of “meaning.””, Fenoglio discloses symbolic layer and sub-symbolic layer including machine-learning/deep-learning processing at the sub-symbolic layer and symbolic reasoning at the symbolic layer, which corresponds to the sub-symbolic layer and symbolic layer, as claimed. Accordingly, Fenoglio continues to teach or at least suggest the claimed symbolic and sub-symbolic layers and is relied upon for those particular limitations addressed by the additional prior art applied in the present rejection.
Applicants’ arguments concerning Ploennigs’ failure to teach the newly amended limitations have been considered and are persuasive. In particular, Ploennigs does not expressly teach pruning, based on target-node details, concepts and relationships of a knowledge graph to obtain a pruned knowledge graph tailored to the target node prior to deployment, using the pruned knowledge graph to generate an executable modified version of the artificial-intelligence metamodel, and deploying that executable modified version to the target node for execution. Accordingly, the previous mapping and rationale with respect to these limitations are not maintained.
However, upon further consideration, new ground(s) of rejection has been raised (See Below).
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 1-4, 6, 11-14, 16, 20 are rejected under 35 U.S.C. 103 as being unpatentable over Ryabinin et.al (NPL: Ontology-Driven Edge Computing) in view of Fenoglio et.al (US 20200022016 A1)
Regarding claim 1,
Ryabinin teaches or at least suggests a part of the method as disclosed within the preamble “A method comprising: providing, by a device, information for display regarding an artificial intelligence metamodel …” (page 1 section Abstract “We propose the new approach to utilize ontology reasoning mechanism right on the extreme resource-constrained Edge devices, not in the Fog or Cloud … smart leverage of on-demand automated transformation of Machine-to-Machine to Human-Centric IoT becomes possible. We demonstrate the practical usefulness of our solution by the implementation of ontology-driven Smart Home edge device” page 12 section 4.7 “enables building ad-hoc user interface with Edge device and opens a gate for the on-demand transformation of M2M functioning into the Human-Centric one” Ryabinin discloses an ontology-driven knowledge representation including domain ontology D and task ontology T, wherein domain ontology D provides reusable semantic knowledge and structure used in generating task ontology T, and task ontology T is represented as a graph comprising ontology nodes and relations. Accordingly, Ryabinin teaches or at least suggests the claimed artificial-intelligence metamodel and an associated knowledge graph, because the ontology graph provides semantic concepts and relationships used to represent the knowledge and behavior of the system. Ryabinin further discloses presenting the ontology-based representation as a DFD shown to a user, thereby teaching or at least suggesting providing information for display regarding the metamodel.)
Ryabinin teaches or at least suggests “receiving, at the device, an indication of a target node in a network to which the artificial intelligence metamodel is to be deployed” (page 6 section 4.2 “We propose the following lifecycle of the ontology-driven Edge device … DFD is automatically transformed into the task ontology that is cognitively compressed and stored in the concise binary format we call EON (Embedded or Edge ONtology). EON-encoded ontology is transferred to Edge device using wired or wireless connection”, and page 10 section 4.6 “Task ontology T should be uploaded to the target device to govern its functioning. However, it should first be compressed to fit the Edge device memory” Ryabinin discloses a particular target Edge device to which task ontology T is uploaded to govern the device’s functioning. The target edge device teaches or at least suggests the claimed target node because it is the network computing device selected to receive and execute the deployed ontology-based model. Accordingly, identification of that particular Edge device for deployment teaches or at least suggests receiving an indication of the target node.)
Ryabinin teaches or at least suggests “obtaining, by the device, node details for the target node indicative of resources available at the target node including capabilities and resource usage at the target node” (page 2 section 1 “Our goal was to make a step towards hardware implementation of task ontologies by organizing the ontology-driven functioning of the light Edge devices based on very resource-constrained microcontroller units (MCUs) like ESP8266 (80 KiB RAM, 80 MHz CPU), ATmega328 (2 KiB RAM, 16 MHz CPU) or even ATtiny45 (256 B RAM, 8 MHz CPU)”, and page 4 section 3 “Main problems of Edge devices, which hinder straightforward use of ontology-driven techniques, are the following … Low RAM capacity … Low CPU frequency … Low power”, and page 5 section 3 “We suggest full-fledged ontology-driven Edge Computing solution, assuming the behavior of Edge Computing devices (e.g. sensing, data processing, actuation and communication) is fully controlled by task ontologies. Thereby we a make a step towards hardware implementation of ontologies”. Ryabinin discloses RAM capacity, CPU frequency, power limitations, and memory constraints of the target Edge device. These hardware characteristics teach the claimed node details because they identify the computational resources and capabilities available to the target node. Ryabinin further discloses determining that task ontology T must be compressed to fit the target-device memory, thereby at least suggesting resource usage because the amount of memory required by T is considered relative to the memory resources available at the target node.)
Ryabinin teaches or at least suggests “generating, by the device, an executable modified version of the artificial intelligence metamodel by pruning, based on the node details, concepts and relationships of a knowledge graph associated with the artificial intelligence metamodel to obtain a pruned knowledge graph tailored to the node details of the target node, prior to deployment, and using the pruned knowledge graph to generate the executable modified version” (page 3 section 2 “The ontology “cognitive compression” method, whereby all the redundant information is trimmed, yet the essential structure of ontology remains retrievable and preserves its semantic power. Removing the excessive ontology nodes and relations, using the topological sorting for the remaining ones and applying multilevel structure layout to describe data flow chains in observable and concise form we managed to fit them in the RAM of tiny MCUs”, and page 10 section 4.6 “size of the ontology is the most significant parameter that should be optimized when it comes to reasoning on the Edge … we developed our own representation format called EON. It is a concise binary format highly optimized for Edge Computing tasks. The EON format is created special for task ontologies (T) trimming all the redundant information and assuming that, if required, this information can be unambiguously restored with a help of the domain ontology (D)”, and page 11 section 4.6 “Next, when preparing T to dumping to EON, we trim all the redundant knowledge that can be unambiguously restored by traversing the domain ontology D … The names of nodes and relations are also trimmed” Ryabinin discloses cognitively compressing task ontology T to fit the resource constraints of the target edge device by removing redundant ontology nodes and relations, which teach or at least suggest the claimed concepts because the nodes represent semantic elements of the ontology, while the ontology relations teach the claimed relationships because they define connections between those semantic elements. Removing those nodes and relations therefore teaches pruning concepts and relationships because the compression is performed so that T fits the target Edge-device resources, whereas the pruning is based on the node details and produces a pruned graph tailored to the target node prior to deployment. Ryabinin further discloses serializing the resulting compressed ontology into EON binary form for execution by the Edge-device reasoner, thereby teaching or at least suggesting using the pruned knowledge graph to generate the executable modified version.)
Ryabinin teaches or at least suggests “deploying, by the device, the executable modified version of the artificial intelligence metamodel to the target node for execution by the target node” (page 6 section 4.2 “DFD is automatically transformed into the task ontology that is cognitively compressed and stored in the concise binary format we call EON (Embedded or Edge ONtology). EON-encoded ontology is transferred to Edge device using wired or wireless connection. Afterwards, the reasoner on the device’ side starts working, so the device performs described actions”, and page 12 section 4.7 “The ontology T is stored in the MCU’s EEPROM. The reading and writing of EON format is ensured by the EON Reader/Writer. TheEvaluation Core performs traversing the EON-encoded ontology and calls operators provided by Function Module whenever needed … When the ontology T encoded to EON format and uploaded to Edge device, its embedded reasoner starts executing the tasks described … For this, ontology T is retrieved back to SciVi, decoded from EON with help of the ontology D and transformed into the DFD shown to the user” Ryabinin discloses transferring and uploading the cognitively compressed, EON-encoded task ontology to the target Edge device. The EON-encoded task ontology teaches or at least suggests the claimed executable modified version because the device-side reasoner traverses that representation, invokes the encoded operators, and executes the tasks represented therein. Accordingly, uploading the EON-encoded ontology to the Edge device and executing it there teaches deploying the executable modified version to the target node for execution by the target node.)
Ryabinin does not teach a part of the method within the preamble “A method comprising: an artificial intelligence … that includes at least a symbolic layer and a sub-symbolic layer”. However, Fenoglio teaches or at least suggests this (paragraph 78 “At the top of hierarchy 500 may be symbolic layer 506 that may leverage symbolic learning to perform the functions described herein. In general, symbolic learning includes a set of symbolic grammar rules specifying the representation language of the system, a set of symbolic inference rules specifying the reasoning competence of the system, and a semantic theory containing the definitions of “meaning.””, and paragraph 74 “At the lowest layer of hierarchy 500 is sub-symbolic layer 502 that processes the multimodal network data 512 collected from the network under scrutiny. For example, at the core of sub-symbolic layer 502 may be one or more DNNs 508 or other machine learning-based model that processes the multimodal network data 512 collected from the network”. Fenoglio discloses the symbolic layer, which performs symbolic inference and semantic reasoning, and the sub-symbolic layer, which includes DNNs or other machine-learning models. The symbolic layer, therefore, teaches the claimed symbolic layer because it performs symbolic and semantic reasoning, while sub-symbolic layer teaches the claimed sub-symbolic layer because it performs learned, machine-learning-based processing. Accordingly, Fenoglio teaches or at least suggests an artificial-intelligence metamodel having the claimed symbolic and sub-symbolic layers.)
Before the effective filling date. It would have been obvious to a person of ordinary skill in the art to combine the teaching of ontology-driven reasoning and control on resource-constrained Edge devices, including using graph representation to prune knowledge and relations at nodes by Ryabinin with the teaching of the symbolic layer and the sub-symbolic layer by Fenoglio. The motivation to do so is referred to in Fenoglio’s disclosure (paragraph 60 “In some aspects, the DFRE may employ a sub-symbolic layer to project raw, multimodal measurements from the network into conceptual spaces from which symbolic information can be obtained. In turn, a symbolic reasoner of the DFRE can use this symbolic information to provide explainable QoE metrics. The overall goal is to optimize the QoE of the network, from the perspective of the user, while making efficient use of networking resources (e.g., QoS) and maintaining a satisfactory level of service, from the standpoint of the service provider (e.g., SLA)”, and paragraph 78 “the symbolic learning and generalized intelligence performed on a human scale at symbolic layer 506 requires a variety of reasoning and learning paradigms that more closely follows how humans learn and are able to explain why a particular conclusion was reached” Fenoglio discloses that the sub-symbolic layer processes raw multimodal network data into symbolic information and that the symbolic layer uses that symbolic information for higher-level reasoning and explainable conclusions. Ryabinin already employs ontology-based reasoning to control the functioning of resource-constrained Edge devices. A person of ordinary skill in the art would therefore have been motivated to incorporate Fenoglio’s symbolic and sub-symbolic layered architecture into Ryabinin’s ontology-driven Edge system so that machine-learning-based processing could extract useful information from raw device/network data and provide that information in symbolic form to Ryabinin’s ontology-based reasoning framework, thereby enabling complementary learned processing and symbolic reasoning at the Edge while retaining Ryabinin’s resource-aware ontology compression and deployment scheme. Such a modification would have involved incorporating a known layered AI-processing architecture into Ryabinin’s existing ontology-driven reasoning system for its established purpose, without changing Ryabinin’s underlying resource-aware compression and Edge-device deployment operation.)
Regarding claim 2 depends on claim 1, thus the rejection of claim 1 is incorporated.
Fenoglio teaches or at least suggests “The method as in claim 1, wherein the sub-symbolic layer of the artificial intelligence metamodel includes one or more machine learning models” (paragraph 74 “at the core of sub-symbolic layer 502 may be one or more DNNs 508 or other machine learning-based model that processes the multimodal network data 512 collected from the network, service providers, network operators, and/or network subscribers” Fenoglio discloses sub-symbolic layer 502 including one or more DNNs 508 or other machine-learning-based models for processing network data. The DNNs and other machine-learning-based models therefore teach the claimed one or more machine learning models included in the sub-symbolic layer.)
Regarding claim 3 depends on claim 1, thus the rejection of claim 1 is incorporated.
Fenoglio teaches or at least suggests “The method as in claim 1, wherein the modified version of the artificial intelligence metamodel is configured to use a semantic reasoning engine to make inferences based on sensor data” (paragraph 69 “Sensor/Telemetry data—e.g., any or all information regarding the traffic flows in the monitored network and/or the environmental conditions of the network”, paragraph 78 “At the top of hierarchy 500 may be symbolic layer 506 that may leverage symbolic learning to perform the functions described herein. In general, symbolic learning includes a set of symbolic grammar rules specifying the representation language of the system, a set of symbolic inference rules specifying the reasoning competence of the system, and a semantic theory containing the definitions of “meaning.””, and paragraph 99 “Also as shown, DFRE architecture 600 may include a symbolic reasoning engine 604 at the symbolic layer 506 that leverages conceptual spaces to map the outputs of sub-symbolic processor 602 into a symbolic format”. Fenoglio discloses sensor/telemetry data obtained from the monitored network and symbolic reasoning engine 604 at symbolic layer 506, which performs symbolic inference and semantic reasoning on information received from the sub-symbolic processing. Thus, the sensor/telemetry data teaches the claimed sensor data, and symbolic reasoning engine 604 teaches the claimed semantic reasoning engine configured to make inferences based on that data.)
Regarding claim 4 depends on claim 1, thus the rejection of claim 1 is incorporated.
Ryabinin in view of Fenoglio teaches or at least suggests “The method as in claim 1, wherein the information includes a representation of a knowledge graph at the symbolic layer of the artificial intelligence metamodel” (page 10 section 4.6 “we apply topological sorting to the nodes, so the data flow represented by use for relations is aligned from left to right, build ing a multilevel structure layout of ontology graph representation. This layout is both human- and machine-readable, incorporating the knowledge about the operations order. When dumped to EON format, this partial order of nodes is preserved, so the reasoner can evaluate the nodes one by one, and any time the next node requires the data from the previous ones, these data are already computed.” Ryabinin discloses a multilevel ontology graph representation comprising nodes and relations and being both human- and machine-readable, thereby teaching or at least suggesting the claimed representation of a knowledge graph. Fenoglio further discloses symbolic layer 506 for symbolic knowledge representation, inference, and semantic reasoning. Thus, Ryabinin in view of Fenoglio teaches or at least suggests that the information includes a representation of the knowledge graph at the symbolic layer of the artificial-intelligence metamodel. It would have been obvious to a person of ordinary skill in the art to incorporate Fenoglio’s symbolic-layer architecture into Ryabinin’s ontology-graph representation so that the graph-based knowledge can be represented and processed at a symbolic layer using known symbolic inference and semantic-reasoning techniques.)
Regarding claim 6 depends on claim 1, thus the rejection of claim 1 is incorporated.
Fenoglio teaches or at least suggests “The method as in claim 1, wherein generating the modified version of the artificial intelligence metamodel based on the node details comprises: converting a semantic reasoning task at the symbolic layer of the artificial intelligence metamodel into a machine learning model at its sub-symbolic layer” (paragraph 78 “At the top of hierarchy 500 may be symbolic layer 506 that may leverage symbolic learning to perform the functions described herein. In general, symbolic learning includes a set of symbolic grammar rules specifying the representation language of the system, a set of symbolic inference rules specifying the reasoning competence of the system, and a semantic theory containing the definitions of “meaning.””, and paragraph 106 “If symbolic reasoning engine 604 concludes the importance of adding new sub-symbolic processing to sub-symbolic processor 602, to handle this new piece of data, engine 604 may activate various mechanisms, including training its own model using its own experiences and potentially online resources for ground truth” Fenoglio discloses symbolic reasoning engine at symbolic layer and further discloses that, when the symbolic reasoning engine determines that new sub-symbolic processing is needed, the engine may activate mechanisms including training a model for sub-symbolic processor. Thus, the semantic reasoning task performed at the symbolic layer leads to generation of a machine-learning model for processing at the sub-symbolic layer, thereby teaching or at least suggesting converting the semantic reasoning task into a machine-learning model at the sub-symbolic layer.)
Regarding claim 11,
Fenoglio teaches or at least suggests “a network interface to communicate with a computer network” (paragraph 32 “The device 200 may also be any other suitable type of device depending upon the type of network architecture in place, such as IoT nodes, etc. Device 200 comprises one or more network interfaces” Fenoglio discloses a device having one or more network interfaces for communicating over a network, thereby teaching the claimed network interface configured to communicate with a computer network.)
Fenoglio teaches or at least suggests “a processor coupled to the network interface and configured to execute one or more processes” (paragraph 32 “The device 200 may also be any other suitable type of device depending upon the type of network architecture in place, such as IoT nodes, etc. Device 200 comprises one or more network interfaces 210, one or more processors 220” Fenoglio further discloses one or more processors associated with the network interface and configured to execute software processes, thereby teaching the claimed processor coupled to the network interface and configured to execute one or more processes.)
Fenoglio teaches or at least suggests “a memory configured to store a process that is executed by the processor, the process when executed configured” (paragraph 34 “The memory 240 comprises a plurality of storage locations that are addressable by the processor(s) 220 and the network interfaces 210 for storing software programs and data structures associated with the embodiments described herein.” Fenoglio discloses memory having storage locations for storing software programs and data structures that are accessible and executable by the processor, thereby teaching or at least suggesting the claimed memory configured to store a process executed by the processor. The operations performed by the stored process are taught or suggested for the same reasons set forth with respect to the corresponding method limitations of claim 1 above.)
The Applicant is further directed to the rejection of claim 1 above. The claim is further rejected under the same rationale as claim 1 because the claim recites similar limitations and processing steps.
Regarding claim 12 depends on claim 11, thus the rejection of claim 11 is incorporated. The Applicant is further directed to the rejection of claim 2 above. The claim is further rejected under the same rationale as claim 2 because the claim recites similar limitations and processing steps.
Regarding claim 13 depends on claim 11, thus the rejection of claim 11 is incorporated. The Applicant is further directed to the rejection of claim 3 above. The claim is further rejected under the same rationale as claim 3 because the claim recites similar limitations and processing steps.
Regarding claim 14 depends on claim 11, thus the rejection of claim 11 is incorporated. The Applicant is further directed to the rejection of claim 4 above. The claim is further rejected under the same rationale as claim 4 because the claim recites similar limitations and processing steps.
Regarding claim 16 depends on claim 11, thus the rejection of claim 11 is incorporated. The Applicant is further directed to the rejection of claim 6 above. The claim is further rejected under the same rationale as claim 6 because the claim recites similar limitations and processing steps.
Regarding claim 20,
Fenoglio teaches or at least suggests “A tangible, non-transitory, computer-readable medium storing program instructions that cause a device to execute a process comprising” (paragraph 34 “The memory 240 comprises a plurality of storage locations that are addressable by the processor(s) 220 and the network interfaces 210 for storing software programs and data structures associated with the embodiments described herein.” Fenoglio discloses physical memory having storage locations for storing software programs and data structures executable by a processor. The disclosed physical memory teaches or at least suggests the claimed tangible, non-transitory computer-readable medium because it provides persistent physical storage for the program instructions rather than a transitory propagating signal. The stored software programs further teach the claimed program instructions that cause a device to execute a process, while the recited processing steps are taught or suggested for the same reasons set forth with respect to the corresponding method limitations of claim 1 above.)
The Applicant is further directed to the rejection of claim 1 above. The claim is further rejected under the same rationale as claim 1 because the claim recites similar limitations and processing steps.
Claims 5, 15 are rejected under 35 U.S.C. 103 as being unpatentable over Ryabinin et.al (NPL: Ontology-Driven Edge Computing) in view of Fenoglio et.al (US 20200022016 A1), further in view of Tung et.al (US 20200050605 A1)
Regarding claim 5 depends on claim 4, thus the rejection of claim 4 is incorporated.
Ryabinin in view of Fenoglio does not expressly teach or suggest “The method as in claim 4, further comprising: receiving, at the device and after providing the information for display, a selection of one or more concepts in the knowledge graph to be pruned, wherein those one or more concepts are pruned from the knowledge graph of the artificial intelligence metamodel when the device generates the modified version of the artificial intelligence metamodel”. However, Tung teaches or at least suggests this limitation (paragraph 29 “These candidate nodes and/or edges comprise a candidate traversal path … According to some embodiments, these candidate traversal paths may be flagged for manual inspection for removal. According to some embodiments, the candidate traversal paths may be further analyzed in view of other testing analysis (e.g., schema pruning)”, paragraph 34 “According to other embodiments, the final removal decision may be a manual process in case the information is actually highly valuable”, and paragraph 44 “the graphical user interfaces (GUIs) 210 displayed by the display circuitry 208 may be representative of GUIs generated by the knowledge graph system 20 to receive query requests or present the query results to the enterprise application”. Tung discloses displaying a knowledge graph through a graphical user interface, identifying candidate nodes and/or edges for pruning, flagging the candidate graph portions for manual inspection for removal, and allowing the final removal decision to be made manually. The candidate nodes teach or at least suggest the claimed concepts in the knowledge graph because the nodes represent entities or information elements of the graph, while the manual removal decision teaches or at least suggests receiving a selection of those concepts to be pruned. Tung further discloses removing the selected graph information during the pruning process, thereby teaching or at least suggesting pruning the selected concepts from the knowledge graph.)
Before the effective filling date. It would have been obvious to a person of ordinary skill in the art to combine the teaching of ontology-driven reasoning and control on resource-constrained Edge devices, including using graph representation to prune knowledge and relations at nodes by Ryabinin and the teaching of the symbolic layer and the sub-symbolic layer by Fenoglio with the teaching of manual knowledge graph pruning by Tung. The motivation to do so is referred to in Tung’s disclosure (paragraph 9 “The knowledge graph disclosed herein improves the quality of the knowledge graph dataset by reducing information stored on the knowledge graph that are determined not to be helpful to the query result analysis (i.e., data “pruning”). The pruning process removes irrelevant information from the knowledge graph when analyzing the query result. The improvements offered by the pruning process includes produces faster processing times and conservation of resources, as resources are not wasted in analyzing irrelevant information that are determined to offer little, to no, relevance to the query result”, and paragraph 34 “According to other embodiments, the final removal decision may be a manual process in case the information is actually highly valuable”. Tung discloses that pruning irrelevant information from a knowledge graph improves the quality of the graph, reduces processing time, and conserves computing resources by avoiding processing of information that provides little or no relevance to the desired result. Tung further permits the final removal decision to be performed manually. Thus, a person of ordinary skill in the art would have been motivated to incorporate Tung’s user-assisted knowledge-graph pruning into the ontology-driven system of Ryabinin, as modified by Fenoglio, so that a user could select unnecessary concepts for removal before generation of the modified model, thereby reducing irrelevant graph information, conserving limited Edge-device resources, and improving processing efficiency.)
Regarding claim 15 depends on claim 14, thus the rejection of claim 14 is incorporated. The Applicant is further directed to the rejection of claim 5 above. The claim is further rejected under the same rationale as claim 5 because the claim recites similar limitations and processing steps.
Claims 7, 10, 17 are rejected under 35 U.S.C. 103 as being unpatentable over Ryabinin et.al (NPL: Ontology-Driven Edge Computing) in view of Fenoglio et.al (US 20200022016 A1), further in view of Ostergaard et.al (US 20220277231 A1)
Regarding claim 7 depends on claim 1, thus the rejection of claim 1 is incorporated.
Ryabinin in view of Fenoglio does not expressly teach or suggest “The method as in claim 1, wherein the device generates the modified version of the artificial intelligence metamodel to satisfy one or more key performance indicators when executed by the target node”. However, Ostergaard teaches or at least suggests this (paragraph 54 “the key performance indicators (KPI) of the inference processes of the edge computers are analyzed at 1406 to determine which models have had the best performance in these similar events. This analysis is done by calculating a quantifiable adjusted value based on each key performance indicator for each model used during each similar event on each edge computer in the subset, and then combining all adjusted values to determine which model has performed best under events similar to the current one historically. Based on this analysis, and an analysis of all historical model performance in similar events from the subset as a whole, a single model is selected that will give the highest average performance for the edge computers”, and paragraph 55 “the results can be compiled into a final table of edge computers and what model each computer should receive 404 as derived from the original list of device groups 1401. The machine learning orchestration 404 is then invoked at 1409 using an interface to call for sending new models to the edge computers”. Ostergaard discloses evaluating models based on key performance indicators associated with execution on Edge computers, selecting the model providing the highest performance for the respective Edge computers, and sending the selected model to those Edge computers for execution. A person of ordinary skill in the art would have been motivated to incorporate Ostergaard’s KPI-based model evaluation and selection technique into Ryabinin’s ontology-driven Edge-device system so that, in addition to tailoring the ontology-based model to the resource constraints of the target Edge device, the modified version could be generated or selected based on performance criteria associated with execution at that target device. Such a combination would predictably allow the target-specific ontology-based model to satisfy desired performance objectives when executed by the target node.)
Before the effective filling date. It would have been obvious to a person of ordinary skill in the art to combine the teaching of ontology-driven reasoning and control on resource-constrained Edge devices, including using graph representation to prune knowledge and relations at nodes by Ryabinin and the teaching of the symbolic layer and the sub-symbolic layer by Fenoglio with the teaching of … by Ostergaard. The motivation to do so is referred to in Ostergaard’s disclosure (paragraph 29 “By detecting a change at the edge computer (“refresh event”) automatically, the delay between the refresh event and the completion of the model refresh can be reduced beyond the capability of related art implementations. By algorithmically determining how to perform the model refresh, the performance of a model orchestration system 404 can be increased in terms of accuracy and adaptiveness by removing the need for manual configuration. The proposed model refresh invocation system accomplishes these feats by considering data from edge computers related to the refresh event, edge computer device information from resource management systems, and time series information on inference processes on edge computers”, and paragraph 57 “the present disclosure can identify and react to a new event faster, improve accuracy by correcting non-optimal model selection in subsequent switches of models, and reduce the labor needed to manually categorize the relationship between devices and between events and models by using data analysis”. Ostergaard discloses considering data from Edge computers, including information regarding inference processes executing at the Edge computers, when determining whether and how to refresh a model, and further teaches evaluating such inference-process performance using key performance indicators as discussed above. Ostergaard further discloses that this approach reduces model-refresh delay, increases accuracy and adaptiveness, enables faster response to new events, and corrects non-optimal model selections. Thus, a person of ordinary skill in the art would have been motivated to incorporate Ostergaard’s use of Edge-device performance feedback and KPI-based model evaluation into Ryabinin’s ontology-driven Edge system so that the modified ontology-based version could be selected or updated based on its performance at the target Edge device, thereby improving accuracy and responsiveness to changing operating conditions.)
Regarding claim 10 depends on claim 1, thus the rejection of claim 1 is incorporated.
Ryabinin in view of Fenoglio does not expressly teach or suggest “The method as in claim 1, further comprising: receiving, at the device and from the target node, performance feedback regarding the modified version of the artificial intelligence metamodel” However, Ostergaard teaches or at least suggests this (paragraph 33 “Data for the process is collected by an edge data collection function 613 on the edge controller. Two kinds of data are collected—statistics 612 from the machine learning inference process 611 running on the same computer, and external edge data 610 that can be observed by the controller 420. The inference process statistics 612 are collected from available log files, application programming interface (API) or standard output from the local inference process running on the computer”, and paragraph 53 “As described in the first process 501, the current ‘live’ data 1400 is sent by the edge device when a refresh event is triggered, including statistics from the edge computer machine learning inference process. This data is then compared to historical statistics from machine learning inference processes in the same deployment 1404, and these statistics can include key performance indicators 1402 of the inference process, along with the ‘live’ data captured 1403 when changes to these inference processes were invoked”. Ostergaard discloses collecting statistics regarding the machine-learning inference process executing on an Edge computer and transmitting live data including those statistics from the Edge computer. The inference-process statistics teach or at least suggest the claimed performance feedback because they reflect performance of the model executing at the Edge computer. Ryabinin, as set forth in claim 1 above, teaches generating a compressed/modified ontology-based task representation and deploying that modified version to a target Edge device for execution. Thus, applying Ostergaard’s known performance-feedback technique to Ryabinin’s ontology-driven Edge system teaches or at least suggests receiving, at the device and from the target node, performance feedback regarding the modified ontology-based version executing at the target node.)
The motivation to combine the teaching is similar to the motivation as recited in claim 7 above.
Regarding claim 17 depends on claim 11, thus the rejection of claim 11 is incorporated. The Applicant is further directed to the rejection of claim 7 above. The claim is further rejected under the same rationale as claim 7 because the claim recites similar limitations and processing steps.
Claims 8, 9, 18, 19 are rejected under 35 U.S.C. 103 as being unpatentable over Ryabinin et.al (NPL: Ontology-Driven Edge Computing) in view of Fenoglio et.al (US 20200022016 A1), further in view of Donatelli et.al (US 20220188630 A1)
Regarding claim 8 depends on claim 1, thus the rejection of claim 1 is incorporated.
Ryabinin in view of Fenoglio does not expressly teach or suggest “The method as in claim 1, further comprising: forming, by the device, the artificial intelligence metamodel by consolidating metamodels from a plurality of nodes in the network”. However, Donatelli teaches or at least suggests this (paragraph 17 “The common models with mapped input are federated by the model augmentation service, generating an aggregate model, effectively the result of the combination of all of the clients' feature data, without the need for clients to share respective client data”, and paragraph 22 “Model augmentation program 300 augments the model of a client using models from the plurality of clients within a domain, generating a single aggregate model federated from all participating client models in the domain. The augmented model benefits from the aggregate feature input data from all clients in the domain without having to share feature data considered proprietary or for internal use by clients”. Donatelli discloses a plurality of networked clients having respective models and federating the client models to generate a single aggregate model. Thus, the clients teach or at least suggest the claimed plurality of network nodes, and federation of the respective client models teaches consolidating models associated with those nodes. When applied to Ryabinin’s ontology-driven system, Donatelli’s federation technique would allow ontology-based representations associated with multiple Edge devices to contribute information to a common aggregate representation, which could then provide a common model/knowledge basis from which Ryabinin’s device-specific task ontology T is generated, compressed according to the resources of a particular Edge device, and deployed to that device.)
Before the effective filling date. It would have been obvious to a person of ordinary skill in the art to combine the teaching of ontology-driven reasoning and control on resource-constrained Edge devices, including using graph representation to prune knowledge and relations at nodes by Ryabinin and the teaching of the symbolic layer and the sub-symbolic layer by Fenoglio with the teaching of federating models from multiple clients into an aggregate model by Donatelli. The motivation to do so is referred to in Donatelli’s disclosure (paragraph 17 “The common models with mapped input are federated by the model augmentation service, generating an aggregate model, effectively the result of the combination of all of the clients' feature data, without the need for clients to share respective client data. The augmented model is distributed to the client members of the grouping of the asset class, and a client-side module prepends the feature mapper to the common model, producing an improved model customized to the respective client and resulting in a model with a more accurate and effective prediction of the output, such as the asset state or condition”. Donatelli discloses that federating models from multiple clients combines information contributed by the respective client models without requiring the clients to share their underlying proprietary data and produces an improved model providing more accurate and effective prediction. Thus, a person of ordinary skill in the art would have been motivated to incorporate Donatelli’s model-federation technique into Ryabinin’s ontology-driven Edge system so that knowledge/model information associated with multiple Edge devices could be combined into a common aggregate representation, thereby improving the information available for generating Ryabinin’s device-specific ontology representation while retaining distributed processing and avoiding unnecessary sharing of underlying client data.)
Regarding claim 9 depends on claim 8, thus the rejection of claim 8 is incorporated.
Ryabinin in view of Donatelli teaches or at least suggests “The method as in claim 8, wherein each of the metamodels from the plurality of nodes in the network use a seed knowledge base that is common across the metamodels from the plurality of nodes in the network” (page 7 section 4.3 “First of all, we propose an extensible domain ontology of Edge Computing functions (hereafter denoted as D) … The ontology D describes available actions, data processing filters, etc”, page 8 section 4.5 “The user-defined DFD is than used by SciVi Smart system to automatically create the task ontology (hereafter denoted as T) …”, and page 9 section 4.6 “we developed our own representation format called EON. It is a concise binary format highly optimized for Edge Computing tasks. The EON format is created special for task ontologies (T) trimming all the redundant information and assuming that, if required, this information can be unambiguously restored with a help of the domain ontology (D)”. Ryabinin discloses domain ontology D as a reusable knowledge source describing available Edge-computing functions and further discloses generating task ontology T for controlling an Edge device, with information removed from T being recoverable using domain ontology D. Thus, domain ontology D teaches or at least suggests the claimed seed knowledge base because it provides the underlying common knowledge used in forming and interpreting the task ontology. Donatelli, as set forth in claim 8 above, supplies the plurality of models associated with respective network nodes. Accordingly, Ryabinin in view of Donatelli teaches or at least suggests that the respective node-associated models use a common underlying seed knowledge base.)
Regarding claim 18 depends on claim 11, thus the rejection of claim 11 is incorporated. The Applicant is further directed to the rejection of 8 above. The claim is further rejected under the same rationale as claim 8 because the claim recites similar limitations and processing steps.
Regarding claim 19 depends on claim 18, thus the rejection of claim 18 is incorporated. The Applicant is further directed to the rejection of claim 9 above. The claim is further rejected under the same rationale as claim 9 because the claim recites similar limitations and processing steps.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to DUY TU DIEP whose telephone number is (703)756-1738. The examiner can normally be reached M-F 8-4:30.
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, Alexey Shmatov can be reached at (571) 270-3428. 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.
/DUY T DIEP/Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123
Prosecution Insights