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 .
The following action is in response to the amendment and remarks of 10/24/2025.
By the amendment, claims 1 and 5 have been amended. Claims 1-20 are pending and have been considered below.
Response to Arguments
The specification objection (Non-Final Rejection 07/24/2025) has been withdrawn in light of the specification amendment and corresponding remarks.
The 35 USC 101 rejection of claims 1-20 (Non-Final Rejection 07/24/2025) have been withdrawn in light of the claims amendment and corresponding remarks (Remarks 10/24/2025 pages 9-10).
Regarding the 35 USC 103 rejection of claims 1, 9 and 15, Applicant argues (Remarks 10/24/2025 pages 10-11) that neither Jawagal nor Polleri teach or disclose “a training corpus that associates at least one training functionality description with a corresponding at least one training computing service” as originally claimed in claims 1, 9 and 15. The Examiner respectfully disagrees.
Jawagal was relied on to teach this limitation. Contrary to Applicant’s argument, Jawagal does teach this limitation in at least the cited paragraph:
“[0020] The candidate answers and the questions thus generated are provided to a machine comprehension model (MC) to train the MC model to identify an answer span to a given user query within a specified context. In an example, the MC model is trained on a generic dataset such as the Stanford Question Answer dataset (SQuAD) and including the candidate answers and the questions that were automatically generated. Further, the MC model implements a 3-layer deep question answering (QA) architecture before the final output layer with varying hidden size having a predetermined number of neurons that better captures the starting and the ending positions of an answer span within a given context. The answer span includes one or more sentences in a given context (e.g., a paragraph) that include information responsive to a user query. One configuration of the 3-layer architecture can be made up of linear layers where all the layers have a hidden size of 384 or 384 hidden neurons while the second configuration can include 2 layers with a hidden size of 384 followed by a third layer with a hidden size of 192. As part of the extension of fine-tuning pooling of the 7th, 9th and the 10th layer of the encoder can also be implemented.”
As can be seen, Jawagal discloses that the trained model is trained on a generic training dataset comprising association between description of functionality and service in the representative questions and answers of the SQuAD corpus. The argument is not persuasive. See also Jawagal paragraphs 25-29 for additional details on the training corpus.
The rejection of original claims 9-11, 14-17 and 20 over Jawagal and Polleri and original claims 12-13 and 18-19 over Jawagal, Polleri and Shashanka have been maintained below.
Regarding amended claim 1, Jawagal and Polleri fail to explicitly disclose determining an “edge” probability. Accordingly, the 35 USC 103 rejections of claims 1-5 and 7-8 over Jawagal and Polleri and claims 6 and 7 over Jawagal, Polleri and Shashanka have been withdrawn. However, this “edge” probability is taught by Shashanka as previously applied in the rejections of claims 6 and 7. Accordingly, new ground for rejection of claims 1-8 over Jawagal, Polleri and Shashanka have been applied below.
Regarding the 35 USC 103 rejection of claims 5, 11 and 17, Applicant argues (Remarks 10/24/2025 pages 11-12) that Jawagal does not disclose at least “determining the at least one determined computing service is determined based on the at least one edge probability being determined to satisfy at least one functionality probability criterion corresponding to the at least one training computing service” as required by amended claim 5. The Examiner respectfully disagrees.
Firstly, the Examiner disagrees that original claims 11 and 17 recite subject matter similar to amended claim 5. Neither claim 11 nor claim 17 recite that at least one determined computer service is based on at least one edge probability, which corresponds to at least one training computing service in parent claim 1, and therefore recite different scope from claim 5.
Regarding claim 5, Jawagal does disclose the amended limitations of claim 5 in at least the cited paragraph:
“[0031] When the user query 192 is received, it is analyzed by the query processor 108 which can tokenize and parse the user query 192. The resulting query words are employed by an information retriever 182 included in the answer generator 112 for an inverted index lookup followed by the scoring of the plurality of documents 110 using a term vector model scoring. The information retriever 182 thus returns the top N contexts (where N is a natural number and N=3 in an example) for the user query 192 and the top-scoring context 164 is provided to the MC model 106. The MC model 106 identifies an answer span 166 to the user query 192 from the top-scoring context 164. The answer span 166 can be provided as the response to the user query 192. In an example, the answer span 166 can be employed by an answer composer 184 to frame a response 194 including the answer span 166 which can be returned to the user who initially posed the user query 192. In an example, the response 194 can be provided via a chat window coupled to the QA system 100 or other virtual agents 196 such as bots. In an example, the QA system 100 can be coupled to existing virtual agents or bots in order to enable automatic domain-based question answering as detailed herein.”
As can be seen, Jawagal discloses determining a probability corresponding to a training computing service to result in a determined computing service through term vector scoring and ranking. The deficiency regarding “edge” probabilities is previously met by the combination of Jawagal, Polleri and Shashanka. The argument is not persuasive.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 9-11, 14-17 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Jawagal (US 20210240776 A1) (previously presented) in view of Polleri (US 20210081819 A1) (previously presented).
Regarding claim 9, Jawagal recites “A system” (Jawagal at 0057: FIG. 13 illustrates a computer system 1300 that may be used to implement the automatic question answering system 100. More particularly, computing machines such as desktops, laptops, smartphones, tablets and wearables which may be used to generate or access the data from the automatic question answering system 100 may have the structure of the computer system 1300.) and “a device comprising a processor configured to:” (Jawagal at 0057: The computer system 1300 may include additional components not shown and that some of the process components described may be removed and/or modified. In another example, a computer system 1300 can sit on external-cloud platforms such as Amazon Web Services, AZURE® cloud or internal corporate cloud computing clusters, or organizational computing resources, etc. See also Jawagal at 0058.)
“apply an embedding function to the description to result in an embedded description, wherein the embedded description comprises at least one description item vector corresponding to the at least one description item;” (Jawagal at 0019: The sequence of the context and answer elements input to the context encoder and the answer encoder are provided by a pre-trained initialized 300-dimensional Global Vectors (GLOVE) embedding which is an unsupervised learning algorithm for obtaining vector representations for words.)
“input the at least one description item vector to a long short-term memory function to result in a context vector;” (Jawagal at 0019: The sequence of the context and answer elements input to the context encoder and the answer encoder are provided by a pre-trained initialized 300-dimensional Global Vectors (GLOVE) embedding which is an unsupervised learning algorithm for obtaining vector representations for words.),
“analyze the context vector using a trained supervised learning model to result in an analyzed context vector, wherein the trained supervised learning model is trained using a training corpus that associates at least one training functionality description with a corresponding at least one training computing service; and” (Jawagal at 0020: The candidate answers and the questions thus generated are provided to a machine comprehension model (MC) to train the MC model to identify an answer span to a given user query within a specified context. In an example, the MC model is trained on a generic dataset such as the Stanford Question Answer dataset (SQuAD) and including the candidate answers and the questions that were automatically generated., see also 0026, 0028, 0032)
“output, to the user interface, at least one computing service” (Jawagal at 0021: The query and the context determined to be relevant to the query are provided to the MC model which identifies an answer span that includes information requested in the query. A response to the query is generated in a complete sentence framed in accordance with the grammar and which includes the information from the identified answer span. The generated response can be provided to the user via a user interface which can include, for example, a virtual agent or a bot.)
However, Jawagal does not recite “receive a description from a user interface of at least one computing service functionality, wherein the description comprises at least one description item;”
“a computing service recommendation of at least one recommended computing service of the at least one determined computing service usable to perform the at least one computing service functionality.”
On the other hand, Polleri recites “receiving, by a system comprising a processor via a user interface, a description of at least one computing service functionality, wherein the description comprises at least one description item;” (Polleri at 0105: In one aspect, techniques can be used for defining a machine learning solution, including receiving a first input (e.g., aural, textual, or GUI) describing a problem for the machine learning solution. A model composition engine 132, as shown in FIG. 1, can transcribe the first input into one or more text fragments. See also Polleri at 0107: The techniques can receive multiple inputs from the user. Based on the multiple inputs, the techniques can determine the intentions of the user to establish a machine learning architecture. In the technique, the intelligent assistant can analyze the inputs and recommend various options for a user based on the analysis. The techniques can generate code for the machine learning architecture. The code can be stored and reused for one or more different machine learning processes. The disclosed techniques simplify the process of developing intelligent applications.)
“a computing service recommendation of at least one recommended computing service of the at least one determined computing service usable to perform the at least one computing service functionality.” (Polleri at 0041: The machine learning techniques can employ a chatbot to indicate the location of data, select a type of machine learning solution, display optimal solutions that best meet the constraints, and recommend the best environment to deploy the solution. See also Polleri at 0047: The library components 168 can be scalable to allows for the definition of multiple environments (e.g., different Kubernetes clusters) where the various portions of the application can be deployed to achieve any Quality of Service (QoS) or Key Performance Indicators (KPIs) specified. A Kubernetes cluster is a set of node machines for running containerized applications.... The monitoring engine 156 can recommend various adjustments to the machine learning application 112 by signaling needed changes to the model composition engine 132.)
Jawagal and Polleri are analogous arts in machine learning in the context of receiving a natural language processing request from the user to retrieve or generate a response. Therefore it would have been obvious to one having ordinary skill in the art and the teachings of Jawagal and Polleri before them before the effective filing of the claimed invention modify Jawagal with Polleri to recite receiving, by a system comprising a processor via a user interface, a description of at least one computing service functionality, wherein the description comprises at least one description item and a computing service recommendation of the at least one determined computing service usable to perform the at least one computing service functionality. One would have been motivated to make this modification to enable generation of models and applications by users without having detailed knowledge of generating code or models, as suggested by Polleri. (Polleri at 0009: [t]he machine learning infrastructure system allows a user (i.e., a data scientist) to generate machine learning applications without having detailed knowledge of the cloud-based network infrastructure or knowledge of how to generate code for building the model. The machine learning platform can analyze the identified data and the user provided desired prediction and performance characteristics to select one or more library components and associated API to generate a machine learning application. The machine learning techniques can monitor and evaluate the outputs of the machine learning model to allow for feedback and adjustments to the model. The machine learning application can be trained, tested, and compiled for export as stand-alone executable code.)
Regarding claim 10, Jawagal and Polleri recites “The system of claim 9,” and Polleri further recites “wherein the computing service recommendation comprises an arrangement of recommended computing services to perform the at least one computing service functionality.” (Polleri at 0342: With reference to FIG. 16, an embodiment of a block diagram that uses ontologies to produce machine learning product graphs used in designing a machine learning model or application. In various embodiments, context aware semantic Machine Learning (ML) services can enables a user to perform a high-precision search of services and automated composition of machine learning deployment services based on formal ontology-based representations of service semantics that can include QoS and product KPIs. … QoS and KPIs constraints are also used as part of architecture selection (e.g., given that latency cannot exceed X, compose services infrastructure which handle a quantity of input data at a specified time in the pipeline that makes up the machine learning model). See also Polleri at 0343: The architecture adoption engine 1612 uses meta-learning to connect various blocks for generation of the product graph 1620. An interpolation process selects the best software functions and hardware for a particular machine learning problem query… Before deployment of production graph, the user can be provided with a proposed product graph 1620 with metrics of model performance and a set of compromises between QoS and other user requirements and constraints (requirements risk).) [A product graph, which is visual representation of nodes, edges and properties that represent and store data and their relationships (see Polleri at 0346), i.e., arrangement of recommended computing services to perform the at least one computing service functionality.] The motivation rationale applied to claim 1 is similarly applicable for claim 3 (additionally Polleri at 0346: [r]elationships can be intuitively visualized using graph databases, making them useful for heavily inter-connected data… Graph databases, by design, allow simple and fast retrieval of complex hierarchical structures that can be difficult to model in relational systems.)
Regarding claim 11, Jawagal and Polleri recites “The system of claim 9,” and Jawagal further recites “wherein the analyzed context vector comprises at least one computing service probability corresponding to at least one computing service to perform the at least one computing service functionality, and the method further comprising determining, by the system, the computing service recommendation based on the at least one computing service probability being determined to satisfy at least one functionality probability criterion corresponding to the at least one training computing service.” (Jawagal at 0031: When the user query 192 is received, it is analyzed by the query processor 108 which can tokenize and parse the user query 192. The resulting query words are employed by an information retriever 182 included in the answer generator 112 for an inverted index lookup followed by the scoring of the plurality of documents 110 using a term vector model scoring. The information retriever 182 thus returns the top N contexts (where N is a natural number and N=3 in an example) for the user query 192 and the top-scoring context 164 is provided to the MC model 106. The MC model 106 identifies an answer span 166 to the user query 192 from the top-scoring context 164.)
Regarding claim 14, Jawagal and Polleri recites “The system of claim 9,” and Jawagal further recites “wherein the trained supervised learning model is updated to result in an updated trained supervised learning model,” (Jawagal at 0050: At 722, the question generator 144 is fine tuned using a set of rewards which can include question answering quality which determines effectiveness of the question generator 144 in answering unseen questions. _Another reward aspect can include fluency. Fluency refers to the syntactic and semantic aspects of a word sequence (i.e., the question). The rewards functions can be concurrently trained on the output from the answer generator 112 with user queries and the corresponding answers in order to improve accuracy of the question generator 144. Therefore, the automatic question answering system 100 incorporates aspects of reinforcement learning to retrain the question generator 144 with a set of reward functions.)
“and wherein the training functionality description and corresponding at least one training computing service are updated with the description of the at least one computing service functionality and the computing service recommendation.” (Jawagal at 0050: At 722, the question generator 144 is fine tuned using a set of rewards which can include question answering quality which determines effectiveness of the question generator 144 in answering unseen questions. _Another reward aspect can include fluency. Fluency refers to the syntactic and semantic aspects of a word sequence (i.e., the question). The rewards functions can be concurrently trained on the output from the answer generator 112 with user queries and the corresponding answers in order to improve accuracy of the question generator 144. Therefore, the automatic question answering system 100 incorporates aspects of reinforcement learning to retrain the question generator 144 with a set of reward functions.)
Regarding claim 15, claim 15 is the non-transitory embodiment of claim 9. Therefore, claim 15 is rejected for the same rationale as claim 9. Additionally, claim 15 recites “A non-transitory machine-readable medium, comprising executable instructions that, when executed by a processor, facilitate performance of operations, comprising:” (Jawagal at 0059: The automatic question answering system 100 may be implemented as software stored on a non-transitory processor-readable medium and executed by the one or more processors 1302.)
Regarding claim 16, claim 16 is the non-transitory embodiment of claim 10. Therefore, claim 16 is rejected for the same rationale as claim 10.
Regarding claim 17, claim 17 is the non-transitory embodiment of claim 11. Therefore, claim 17 is rejected for the same rationale as claim 11.
Regarding claim 20, claim 20 is the non-transitory embodiment of claim 14. Therefore, claim 20 is rejected for the same rationale as claim 14.
Claims 1-8, 12-13 and 18-19 are rejected under 35 U.S.C. 103 as being unpatentable over Jawagal in view of Polleri and in further view of Shashanka, et al. (US 20210256115 A1) (previously presented).
Regarding claim 1, Jawagal recites “A method, comprising:” (Jawagal at 0010-0011: FIG. 9 shows a flowchart that details a method of generating a natural language like response incorporating an answer span in accordance with the examples disclosed herein.)
“applying, by the system, an embedding function to the description to result in an embedded description, wherein the embedded description comprises at least one description item vector corresponding to the at least one description item;” (Jawagal at 0019: The sequence of the context and answer elements input to the context encoder and the answer encoder are provided by a pre-trained initialized 300-dimensional Global Vectors (GLOVE) embedding which is an unsupervised learning algorithm for obtaining vector representations for words.)
“inputting, by the system, the at least one description item vector to a recurrent neural network function to result in a context vector;” (Jawagal at 0019: The sequence of the context and answer elements input to the context encoder and the answer encoder are provided by a pre-trained initialized 300-dimensional Global Vectors (GLOVE) embedding which is an unsupervised learning algorithm for obtaining vector representations for words.), “wherein the context vector is at least based on a number of occurrences of at least one word corresponding to the description or based on at least one placement of the at least one word in the description;” (Jawagal at 0033: During the encoding process, each embedding vector is augmented with a binary feature to illustrate the presence of a given word token in a context.)
“analyzing, by the system, the context vector using a trained supervised learning model to result in an analyzed context vector, wherein the trained supervised learning model is trained using a training corpus that associates at least one training functionality description with a corresponding at least one training computing service; and” (Jawagal at 0020: The candidate answers and the questions thus generated are provided to a machine comprehension model (MC) to train the MC model to identify an answer span to a given user query within a specified context. In an example, the MC model is trained on a generic dataset such as the Stanford Question Answer dataset (SQuAD) and including the candidate answers and the questions that were automatically generated., see also 0026, 0028, 0032)
“based on the analyzed context vector, determining by the system, at least one probability corresponding to the at least one training computing service;
based on the at least one probability, determining at least one computing service, of the at least one training service, to result in at least one determined computing service; and” (Jawagal at 0031: When the user query 192 is received, it is analyzed by the query processor 108 which can tokenize and parse the user query 192. The resulting query words are employed by an information retriever 182 included in the answer generator 112 for an inverted index lookup followed by the scoring of the plurality of documents 110 using a term vector model scoring. The information retriever 182 thus returns the top N contexts (where N is a natural number and N=3 in an example) for the user query 192 and the top-scoring context 164 is provided to the MC model 106. The MC model 106 identifies an answer span 166 to the user query 192 from the top-scoring context 164.)
“based on the at least one determined computing service, outputting, via the user interface,” (Jawagal at 0021: The query and the context determined to be relevant to the query are provided to the MC model which identifies an answer span that includes information requested in the query. A response to the query is generated in a complete sentence framed in accordance with the grammar and which includes the information from the identified answer span. The generated response can be provided to the user via a user interface which can include, for example, a virtual agent or a bot.)
However, Jawagal does not recite “receiving, by a system comprising at least one processor via a user interface, a description of at least one computing service functionality, wherein the description comprises at least one description item;”
“a computing service recommendation of at least one recommended computing service of the at least one determined computing service usable to perform the at least one computing service functionality.”
On the other hand, Polleri recites “receiving, by a system comprising a processor via a user interface, a description of at least one computing service functionality, wherein the description comprises at least one description item;” (Polleri at 0105: In one aspect, techniques can be used for defining a machine learning solution, including receiving a first input (e.g., aural, textual, or GUI) describing a problem for the machine learning solution. A model composition engine 132, as shown in FIG. 1, can transcribe the first input into one or more text fragments. See also Polleri at 0107: The techniques can receive multiple inputs from the user. Based on the multiple inputs, the techniques can determine the intentions of the user to establish a machine learning architecture. In the technique, the intelligent assistant can analyze the inputs and recommend various options for a user based on the analysis. The techniques can generate code for the machine learning architecture. The code can be stored and reused for one or more different machine learning processes. The disclosed techniques simplify the process of developing intelligent applications.)
“a computing service recommendation of at least one recommended computing service of the at least one determined computing service usable to perform the at least one computing service functionality.” (Polleri at 0041: The machine learning techniques can employ a chatbot to indicate the location of data, select a type of machine learning solution, display optimal solutions that best meet the constraints, and recommend the best environment to deploy the solution. See also Polleri at 0047: The library components 168 can be scalable to allows for the definition of multiple environments (e.g., different Kubernetes clusters) where the various portions of the application can be deployed to achieve any Quality of Service (QoS) or Key Performance Indicators (KPIs) specified. A Kubernetes cluster is a set of node machines for running containerized applications.... The monitoring engine 156 can recommend various adjustments to the machine learning application 112 by signaling needed changes to the model composition engine 132.)
Jawagal and Polleri are analogous arts in machine learning in the context of receiving a natural language processing request from the user to retrieve or generate a response. Therefore it would have been obvious to one having ordinary skill in the art and the teachings of Jawagal and Polleri before them before the effective filing of the claimed invention modify Jawagal with Polleri to recite receiving, by a system comprising a processor via a user interface, a description of at least one computing service functionality, wherein the description comprises at least one description item and a computing service recommendation of the at least one determined computing service usable to perform the at least one computing service functionality. One would have been motivated to make this modification to enable generation of models and applications by users without having detailed knowledge of generating code or models, as suggested by Polleri. (Polleri at 0009: [t]he machine learning infrastructure system allows a user (i.e., a data scientist) to generate machine learning applications without having detailed knowledge of the cloud-based network infrastructure or knowledge of how to generate code for building the model. The machine learning platform can analyze the identified data and the user provided desired prediction and performance characteristics to select one or more library components and associated API to generate a machine learning application. The machine learning techniques can monitor and evaluate the outputs of the machine learning model to allow for feedback and adjustments to the model. The machine learning application can be trained, tested, and compiled for export as stand-alone executable code.)
Polleri further teaches determining edges for the recommended computing services (Polleri at 0346: Product graphs 1620 are visual representations of data that represent mathematical structures used to study pairwise relationships between objects and entities. In computing, a graph database (GDB) is a database that uses graph structures for semantic queries with nodes, edges, and properties to represent and store data. A key concept of the system is the graph (or edge or relationship). The graph relates the data items in the store to a collection of nodes and edges, the edges representing the relationships between the nodes. The relationships allow data in the store to be linked together directly and, in many cases, retrieved with one operation. Graph databases hold the relationships between data as a priority.)
However, neither Jawagal nor Polleri explicitly disclose wherein the determined at least one probability is an edge probability.
On the other hand, Shashanka recites “determining, at least one edge probability corresponding to at least one computing service” (Shashanka at 0082: In another embodiment, the semantic representation of the document based on the document embedding is generated using representative sentences. In the representative sentences technique, the electronic device (100) generates a similarity graph. Each node in the similarity graph is a sentence block and edge-weights in the similarity graph are provided by angular cosine similarities. Further, the electronic device (100) determines a page-rank score for each of the sentence block of the plurality of sentence blocks using a page-rank algorithm. Shashanka at 0088: As, the maximum margin relevance scoring for the plurality of sentence blocks is determined, the electronic device (100) then selects a predetermined percentage of sentence blocks of the plurality of sentence blocks with the highest maximum margin relevance scoring for further processing.) [The edge-weights, i.e., probability criterion, are used to determine the sentence blocks with the highest maximum margin relevance scoring, i.e., service based on an edge.]
Jawagal, Polleri and Shashanka are analogous arts in machine learning in the context of receiving a natural language processing request from the user to retrieve or generate a response. Therefore it would have been obvious to one having ordinary skill in the art and the teachings of Jawagal, Polleri and Shashank before them before the effective filing of the claimed invention to modify Jawagal and Polleri with Shashanka to recite determining, at least one edge probability to determine at least one computing service. One would have been motivated to make this modification to better leverage state-of-the-art models to reduce data security risk, as suggested by Shashanka (Shashanka at 0077: processes the plurality of sentence blocks to determine the embeddings for each of the plurality of sentence blocks. The embeddings for each of the plurality of sentence blocks are in the form of vectors and are extracted using a deep-learning language model (step 342). Further, the proposed method leverages state-of-the-art model architectures called attention networks (i.e., Transformers) and bidirectional transformers. For example, the models include but are not limited to: Google Universal Sentence Encoder Large v3, Google BERT and S-BERT and all the other variants of BERT. Shashanka at 0026: Further, the method includes determining, by the document semantics controller, the semantic representation of the document based on the embeddings for each of the sentence blocks and generating, by the document semantics controller, the semantic representation of the document to determine the data security risk associated with the document.)
Regarding claim 2, Jawagal, Polleri and Shashanka recites “The method of claim 1,” and Polleri further recites “wherein the at least one description item comprises at least one of: at least one word, at least one image signal representing at least one portion of at least one image, or at least one audio signal representing at least one portion of at least one sound.” (Polleri at 0105: Machine learning models are trained for generating predictive outcomes for code integration requests. In one aspect, techniques can be used for defining a machine learning solution, including receiving a first input (e.g., aural, textual, or GUI) describing a problem for the machine learning solution. A model composition engine 132, as shown in FIG. 1, can transcribe the first input into one or more text fragments. The model composition engine 132 can determine an intent of a user to create a machine learning architecture based at least in part on the one or more text fragments. The techniques can include correlating the one or more text fragments to one or more machine learning frameworks of a plurality of models. The techniques can include presenting (e.g., interface or audio) the one or more machine learning model to the user.) The motivation rationale applied to claim 1 is similarly applicable for claim 2.
Regarding claim 3, Jawagal, Polleri and Shashanka recites “The method of claim 1,” and Polleri further recites “wherein the computing service recommendation comprises an arrangement of recommended computing services to perform the at least one computing service functionality.” (Polleri at 0342: With reference to FIG. 16, an embodiment of a block diagram that uses ontologies to produce machine learning product graphs used in designing a machine learning model or application. In various embodiments, context aware semantic Machine Learning (ML) services can enables a user to perform a high-precision search of services and automated composition of machine learning deployment services based on formal ontology-based representations of service semantics that can include QoS and product KPIs. … QoS and KPIs constraints are also used as part of architecture selection (e.g., given that latency cannot exceed X, compose services infrastructure which handle a quantity of input data at a specified time in the pipeline that makes up the machine learning model). See also Polleri at 0343: The architecture adoption engine 1612 uses meta-learning to connect various blocks for generation of the product graph 1620. An interpolation process selects the best software functions and hardware for a particular machine learning problem query… Before deployment of production graph, the user can be provided with a proposed product graph 1620 with metrics of model performance and a set of compromises between QoS and other user requirements and constraints (requirements risk).) [A product graph, which is visual representation of nodes, edges and properties that represent and store data and their relationships (see Polleri at 0346), i.e., arrangement of recommended computing services to perform the at least one computing service functionality.] The motivation rationale applied to claim 1 is similarly applicable for claim 3 (additionally Polleri at 0346: [r]elationships can be intuitively visualized using graph databases, making them useful for heavily inter-connected data… Graph databases, by design, allow simple and fast retrieval of complex hierarchical structures that can be difficult to model in relational systems.)
Regarding claim 4, Jawagal, Polleri and Shashanka recites “The method of claim 1,” and Jawagal further recites “wherein the context vector comprises at least one of: a usage frequency of the at least one description item or an order of the at least one description item.” (Jawagal at 0018: The text extracted from the plurality of documents is tokenized into word tokens wherein each word forms a token and word-specific statistical features are considered such as positioning and frequency of the word tokens, word relatedness etc. The extracted features are used to generate a single term score (TS) for each of the word tokens.)
Regarding claim 5, Jawagal, Polleri and Shashanka recites “The method of claim 1,” and Jawagal further recites “determining, the at least one determined computing service is based on the least one edge probability being determined to satisfy at least one functionality probability criterion corresponding to the at least one training computing service.” (Jawagal at 0031: When the user query 192 is received, it is analyzed by the query processor 108 which can tokenize and parse the user query 192. The resulting query words are employed by an information retriever 182 included in the answer generator 112 for an inverted index lookup followed by the scoring of the plurality of documents 110 using a term vector model scoring. The information retriever 182 thus returns the top N contexts (where N is a natural number and N=3 in an example) for the user query 192 and the top-scoring context 164 is provided to the MC model 106. The MC model 106 identifies an answer span 166 to the user query 192 from the top-scoring context 164.)
Regarding claim 6, Jawagal, Polleri and Shashanka recites “The method of claim 5”, and Polleri recites “determining, by the system, edges for at least one pair of the at least one recommended computing service; and” (Polleri at 0346: Product graphs 1620 are visual representations of data that represent mathematical structures used to study pairwise relationships between objects and entities. In computing, a graph database (GDB) is a database that uses graph structures for semantic queries with nodes, edges, and properties to represent and store data. A key concept of the system is the graph (or edge or relationship). The graph relates the data items in the store to a collection of nodes and edges, the edges representing the relationships between the nodes. The relationships allow data in the store to be linked together directly and, in many cases, retrieved with one operation. Graph databases hold the relationships between data as a priority.)
Further, Shashanka recites “determining, by the system, the computing service recommendation based on at least one edge, of the edges, having an edge probability that is determined to satisfy an edge probability criterion corresponding to the at least one training computing service.” (Shashanka at 0082: In another embodiment, the semantic representation of the document based on the document embedding is generated using representative sentences. In the representative sentences technique, the electronic device (100) generates a similarity graph. Each node in the similarity graph is a sentence block and edge-weights in the similarity graph are provided by angular cosine similarities. Further, the electronic device (100) determines a page-rank score for each of the sentence block of the plurality of sentence blocks using a page-rank algorithm. Shashanka at 0088: As, the maximum margin relevance scoring for the plurality of sentence blocks is determined, the electronic device (100) then selects a predetermined percentage of sentence blocks of the plurality of sentence blocks with the highest maximum margin relevance scoring for further processing.) [The edge-weights, i.e., probability criterion, are used to determine the sentence blocks with the highest maximum margin relevance scoring, i.e., recommendation based on an edge.] The motivation rationale of claim 1 is similarly applicable to claim 6.
Regarding claim 7, claim 7 has materially similar limitations as claim 6. Therefore, claim 7 is rejected for the same rationale as claim 6.
Regarding claim 8, Jawagal, Polleri and Shashanka recites “The method of claim 1,” and Jawagal further recites “wherein the trained supervised learning model is updated to result in an updated trained supervised learning model,” (Jawagal at 0050: At 722, the question generator 144 is fine tuned using a set of rewards which can include question answering quality which determines effectiveness of the question generator 144 in answering unseen questions. _Another reward aspect can include fluency. Fluency refers to the syntactic and semantic aspects of a word sequence (i.e., the question). The rewards functions can be concurrently trained on the output from the answer generator 112 with user queries and the corresponding answers in order to improve accuracy of the question generator 144. Therefore, the automatic question answering system 100 incorporates aspects of reinforcement learning to retrain the question generator 144 with a set of reward functions.)
“and wherein the training functionality description and corresponding at least one training computing service are updated with the description of the at least one computing service functionality and the computing service recommendation.” (Jawagal at 0050: At 722, the question generator 144 is fine tuned using a set of rewards which can include question answering quality which determines effectiveness of the question generator 144 in answering unseen questions. _Another reward aspect can include fluency. Fluency refers to the syntactic and semantic aspects of a word sequence (i.e., the question). The rewards functions can be concurrently trained on the output from the answer generator 112 with user queries and the corresponding answers in order to improve accuracy of the question generator 144. Therefore, the automatic question answering system 100 incorporates aspects of reinforcement learning to retrain the question generator 144 with a set of reward functions.)
Regarding claim 12, Jawagal and Polleri recites “The system of claim 11”, Polleri further teaches determining edges for at least one pair of the at least one recommended computing service (Polleri at 0346: Product graphs 1620 are visual representations of data that represent mathematical structures used to study pairwise relationships between objects and entities. In computing, a graph database (GDB) is a database that uses graph structures for semantic queries with nodes, edges, and properties to represent and store data. A key concept of the system is the graph (or edge or relationship). The graph relates the data items in the store to a collection of nodes and edges, the edges representing the relationships between the nodes. The relationships allow data in the store to be linked together directly and, in many cases, retrieved with one operation. Graph databases hold the relationships between data as a priority.)
However, neither Jawagal nor Polleri recite “determining, by the system, the computing service recommendation based on at least one edge, of the edges, having an edge probability that is determined to satisfy an edge probability criterion corresponding to the at least one training computing service.”
On the other hand, Shashanka recites “determining, by the system, the computing service recommendation based on at least one edge, of the edges, having an edge probability that is determined to satisfy an edge probability criterion corresponding to the at least one training computing service.” (Shashanka at 0082: In another embodiment, the semantic representation of the document based on the document embedding is generated using representative sentences. In the representative sentences technique, the electronic device (100) generates a similarity graph. Each node in the similarity graph is a sentence block and edge-weights in the similarity graph are provided by angular cosine similarities. Further, the electronic device (100) determines a page-rank score for each of the sentence block of the plurality of sentence blocks using a page-rank algorithm. Shashanka at 0088: As, the maximum margin relevance scoring for the plurality of sentence blocks is determined, the electronic device (100) then selects a predetermined percentage of sentence blocks of the plurality of sentence blocks with the highest maximum margin relevance scoring for further processing.) [The edge-weights, i.e., probability criterion, are used to determine the sentence blocks with the highest maximum margin relevance scoring, i.e., recommendation based on an edge.]
Jawagal, Polleri and Shashanka are analogous arts in machine learning in the context of receiving a natural language processing request from the user to retrieve or generate a response. Therefore it would have been obvious to one having ordinary skill in the art and the teachings of Jawagal, Polleri and Shashank before them before the effective filing of the claimed invention to modify Jawagal and Polleri with Shashanka to recite determining, by the system, the computing service recommendation based on at least one edge, of the edges, having an edge probability that is determined to satisfy an edge probability criterion corresponding to the at least one training computing service. One would have been motivated to make this modification to better leverage state-of-the-art models to reduce data security risk, as suggested by Shashanka (Shashanka at 0077: processes the plurality of sentence blocks to determine the embeddings for each of the plurality of sentence blocks. The embeddings for each of the plurality of sentence blocks are in the form of vectors and are extracted using a deep-learning language model (step 342). Further, the proposed method leverages state-of-the-art model architectures called attention networks (i.e., Transformers) and bidirectional transformers. For example, the models include but are not limited to: Google Universal Sentence Encoder Large v3, Google BERT and S-BERT and all the other variants of BERT. Shashanka at 0026: Further, the method includes determining, by the document semantics controller, the semantic representation of the document based on the embeddings for each of the sentence blocks and generating, by the document semantics controller, the semantic representation of the document to determine the data security risk associated with the document.)
Regarding claim 13, claim 13 has materially similar limitations as claim 12. Therefore, claim 13 is rejected for the same rationale as claim 12.
Regarding claim 18, claim 18 is the non-transitory embodiment of claim 12. Therefore, claim 18 is rejected for the same rationale as claim 12.
Regarding claim 19, claim 19 is the non-transitory embodiment of claim 13. Therefore, claim 19 is rejected for the same rationale as claim 13.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ANDREW L TANK whose telephone number is (571)270-1692. The examiner can normally be reached Monday-Thursday 9a-6p.
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, Matthew Ell can be reached at 571-270-3264. 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.
/ANDREW L TANK/ Primary Examiner, Art Unit 2141
Prosecution Insights