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 .
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 09/02/2025 has been entered.
Response to Amendment
The Amendment filed 09/02/2025 has been entered. Claim 5 is canceled. Claims 1-4 and 6-14 remain pending in the application.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
Claims 13 and 14 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, or for pre-AIA the applicant regards as the invention.
Claims 13 and 14 recite “at least one processor of a machine learning model”. However, it is unclear what is meant by the above limitation. The specification mentions Device 108 includes one or multiple processor(s) and at least one memory for instructions and is designed to carry out a method described in the following. In the present example, model 102 is designed to determine triple 112 for knowledge graph 100 and includes the first entity, a second entity a1t, and their relationship a1t ([0024]) and Fig. 3 shows steps in a method for training a model for determining the knowledge graph ([0021]). For the purpose of examination, examiner will interpret as at least one processor that implements a machine learning model.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-4 and 6-14 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Claims 1-4 and 6-12 are directed to a method, claim 13 is directed to a device and claim 14 is directed to a medium. Therefore, the claims are eligible under Step 1 for being directed to a process, a machine and a manufacture respectively.
Step 2A Prong 1:
Independent claims 1, 13 and 14 recite:
determining a prediction for a second entity, a prediction for a relationship for a triple for the knowledge graph, and a prediction for an explanation for the triple using the model as a function of the input data - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of analyzing data, which is observing, evaluating and judging that is practically capable of being performed in the human mind with the assistance of pen and paper;
determining a first probability that the model assigns to the triple and a second probability that the model assigns to the prediction for the explanation;
determining a classification for the triple as a function of the first probability and of the second probability - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of analyzing data, which is observing, evaluating and judging that is practically capable of being performed in the human mind with the assistance of pen and paper; and
when the classification meets a condition:
determining the explanation as a function of the prediction for the explanation and of the triple for the knowledge graph as a function of the first entity, of the prediction for the second entity, and of the prediction for the relationship, - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of analyzing data, which is observing, evaluating and judging that is practically capable of being performed in the human mind with the assistance of pen and paper;
a function being defined depending on a weighted sum of the first probability and of the second probability and at least one parameter being trained for the model depending on the function - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses a mathematical calculation which falls within the mathematical concepts grouping of abstract ideas;
Dependent claim 2 recites:
determining a first measure, which characterizes a difference between two probability distributions, between the prediction for the second entity and the second entity - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of analyzing data, which is observing, evaluating and judging that is practically capable of being performed in the human mind with the assistance of pen and paper;
determining a second measure, which characterizes a difference between two probability distributions, between the prediction for the relationship and the relationship - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of analyzing data, which is observing, evaluating and judging that is practically capable of being performed in the human mind with the assistance of pen and paper;
determining a third measure, which characterizes a difference between two probability distributions, of a weighted third cross entropy between the prediction for the explanation and the explanation - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of analyzing data, which is observing, evaluating and judging that is practically capable of being performed in the human mind with the assistance of pen and paper,
the function being defined depending on the first measure, of the second measure, and of the third measure and of a weighted sum of the first probability and of the second probability - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses a mathematical calculation which falls within the mathematical concepts grouping of abstract ideas.
Dependent claim 3 recites:
wherein the function is also defined depending on a sum of the first measure, the second measure, and the third measure - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses a mathematical calculation which falls within the mathematical concepts grouping of abstract ideas.
Dependent claim 6 recites:
wherein a vector representation that defines at least a portion of the input data is determined for at least one word of the text body or for at least one sentence of the text body, as a function of at least one other word or as a function of at least one other sentence - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of analyzing data, which is observing, evaluating and judging that is practically capable of being performed in the human mind with the assistance of pen and paper.
Dependent claim 7 recites:
wherein a first vector is assigned to a first word from a sentence from the text body, a second vector is assigned to a second word from the sentence of the text body, the vector representation being computed as a weighted sum from the first vector and the second vector - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses a mathematical calculation which falls within the mathematical concepts grouping of abstract ideas.
Dependent claim 11 recites:
wherein metadata that are assigned to a triple in the knowledge graph are determined as a function of the prediction for the explanation or as a function of the explanation - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses a mathematical calculation which falls within the mathematical concepts grouping of abstract ideas.
Dependent claim 12 recites:
wherein the classification meets the condition when the first probability exceeds a first threshold value and when the second probability exceeds a second threshold value - Under its broadest reasonable interpretation in light of the specification, this limitation encompasses a mathematical calculation which falls within the mathematical concepts grouping of abstract ideas.
Step 2A Prong 2: This judicial exception is not integrated into a practical application because they recite the additional elements:
Independent claims 1, 13 and 14:
providing a first entity for the knowledge graph - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g));
providing a text body that includes a text collection or a document collection that includes a sentence - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g));
providing input data for a model that are defined as a function of the text body and the first entity of the knowledge graph - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g));
wherein the method is performed by at least one processor that implements a machine learning model - the “implements” step recited at a high level of generality, and amounts to more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f) and amounts to components of a general purpose computer that applies a judicial exception, by use of conventional computer functions (see MPEP § 2106.05(b));
A device for determining a knowledge graph - These limitations amount to components of a general purpose computer that applies a judicial exception, by use of conventional computer functions (see MPEP § 2106.05(b));
A non-transitory machine-readable storage medium on which is stored a computer program for determining a knowledge graph, the computer program, when executed by a computer, causing the computer to perform - these limitations amount to components of a general purpose computer that applies a judicial exception, by use of conventional computer functions (see MPEP § 2106.05(b)). The claimed non-transitory machine-readable medium is recited at a high level of generality and are merely invokes as tool to store instructions to perform the abstract idea.
wherein training data are provided - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)),
wherein the explanation is an excerpt from the text body; the training data including a plurality of pairs of a triple and an explanation assigned to the triple, the model including a classifier that is trained as a function of the training data for determining for the first entity from the triple the prediction for the relationship and the prediction for the explanation for the triple - the description of the explanation, the training data and a classifier amount to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Dependent claim 2 recites:
providing the second entity and the relationship - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g));
Dependent claim 4 recites:
the first measure is at least one from a cross entropy, a Kullback-Leibler divergence, and an f-divergence - the description of the first measure amounts to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)), and/or
the second measure is at least one from a cross entropy, a Kullback-Leibler divergence, and an f-divergence - the description of the second measure amounts to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)), and/or
the third measure is at least one from a cross entropy, a weighted cross entropy of a Kullback-Leibler divergence, and an f-divergence - the description of the third measure amounts to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Dependent claim 8 recites:
wherein an output including the triple is output at a first output of the model - the “output” step recited at a high level of generality, and amounts to mere data outputting, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Dependent claim 9 recites:
wherein an output that defines a start and an end of at least one section in the text body is output at a second output of the model - the “output” step recited at a high level of generality, and amounts to mere data outputting, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Dependent claim 10 recites:
wherein the prediction for the second entity, the prediction for the relationship or the prediction for the explanation is defined by a value of a distribution of values across a plurality of vectors - the “define” step recited at a high level of generality, and amounts to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Accordingly, these additional elements do not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea. The claims are thus directed to the abstract idea.
Step 2B: The claims do not include additional elements that amount to significantly more than the judicial exception.
The additional elements:
Independent claims 1, 13 and 14:
providing a first entity for the knowledge graph - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g));
providing a text body that includes a text collection or a document collection that includes a sentence - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g));
providing input data for a model that are defined as a function of the text body and the first entity of the knowledge graph - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g));
wherein the method is performed by at least one processor that implements a machine learning model - the “implements” step recited at a high level of generality, and amounts to more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f) and amounts to components of a general purpose computer that applies a judicial exception, by use of conventional computer functions (see MPEP § 2106.05(b));
A device for determining a knowledge graph - These limitations amount to components of a general purpose computer that applies a judicial exception, by use of conventional computer functions (see MPEP § 2106.05(b));
A non-transitory machine-readable storage medium on which is stored a computer program for determining a knowledge graph, the computer program, when executed by a computer, causing the computer to perform - these limitations amount to components of a general purpose computer that applies a judicial exception, by use of conventional computer functions (see MPEP § 2106.05(b)).
wherein training data are provided - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)),
wherein the explanation is an excerpt from the text body; the training data including a plurality of pairs of a triple and an explanation assigned to the triple, the model including a classifier that is trained as a function of the training data for determining for the first entity from the triple the prediction for the relationship and the prediction for the explanation for the triple - the description of the explanation, the training data and a classifier amount to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Dependent claim 2 recites:
providing the second entity and the relationship - the “providing” step recited at a high level of generality, and amounts to mere data gathering, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g));
Dependent claim 4 recites:
the first measure is at least one from a cross entropy, a Kullback-Leibler divergence, and an f-divergence - the description of the first measure amounts to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)), and/or
the second measure is at least one from a cross entropy, a Kullback-Leibler divergence, and an f-divergence - the description of the second measure amounts to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)), and/or
the third measure is at least one from a cross entropy, a weighted cross entropy of a Kullback-Leibler divergence, and an f-divergence - the description of the third measure amounts to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Dependent claim 8 recites:
wherein an output including the triple is output at a first output of the model - the “output” step recited at a high level of generality, and amounts to mere data outputting, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Dependent claim 9 recites:
wherein an output that defines a start and an end of at least one section in the text body is output at a second output of the model - the “output” step recited at a high level of generality, and amounts to mere data outputting, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Dependent claim 10 recites:
wherein the prediction for the second entity, the prediction for the relationship or the prediction for the explanation is defined by a value of a distribution of values across a plurality of vectors - the “define” step recited at a high level of generality, and amounts to mere data description, which is a form of insignificant extra-solution activity (see MPEP § 2106.05(g)).
Accordingly, these additional elements do not amount to significantly more than the judicial exception. As such, the claims are ineligible.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1, 13 and 14 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by COSTABELLO et al. (hereinafter COSTABELLO), US 20190220524 A1.
Regarding independent claim 1, COSTABELLO teaches a method for determining a knowledge graph ([0003] generate a reasoning graph based on the ontology and the particular neighborhood, and may generate an explanation of the particular candidate response based on the reasoning graph), comprising the following steps:
providing a first entity for the knowledge graph ([0013] As shown in FIG. 1A, and by reference number 105, a user of the user device (e.g., via a user interface provided to the user) may cause the user device to provide, to the prediction platform, training data for training a knowledge graph associated with a particular disease (e.g., severe acute respiratory syndrome (SARS)));
providing a text body that includes a text collection or a document collection that includes a sentence ([0013] As further shown in FIG. 1A, and by reference number 110, the user may cause the user device to provide, to the prediction platform, an ontology for the training data. In some implementations, the training data and the ontology may not be stored in the user device, but the user device may cause the training data and the ontology to be provided from a resource, storing the training data and the ontology, to the prediction platform; [0014] the training data may include information associated with a subject of the ontology. For example, example implementation 100 relates to an ontology associated with the SARS disease. Thus, the training data may include data associated with the SARS disease that is received from relationship database management systems (RDBMS), comma-separated values (CSV) data stores, and/or the like. As shown in FIG. 1A, the training data may include data indicating a disease (e.g., SARS), a cause of the disease (e.g., virus_XYZ), what organ the disease affects (e.g., lungs), symptoms of the disease (e.g., high fever), a virus identifier (e.g., virus_XYZ), a protein sequence associated with the virus (e.g., ACARBAC), a drug identifier associated with a drug that treats the disease (e.g., SARS vaccine), a drug type (e.g., vaccine), what the drug treats (e.g., SARS), and/or the like; [0020] As shown in FIG. 1C, a fit/train engine, of the prediction platform, may receive the knowledge graph. In some implementations, the knowledge graph may be serialized into a list of statements, and the list of statements may be received by the fit/train engine; [0023] As shown in FIG. 1D, assume that the prediction platform receives (e.g., from the user device) new statements, indicating among all that SARS Type B is a disease);
providing input data for a model that are defined as a function of the text body and the first entity of the knowledge graph ([0016] As shown in FIG. 1B, and by reference numbers 105 and 110, the training data and the ontology may be provided to a knowledge graph converter of the prediction platform);
determining a prediction for a second entity, a prediction for a relationship for a triple for the knowledge graph ([0017] As further shown in FIG. 1B, and by reference number 115, the knowledge graph converter may generate a knowledge graph based on the training data and the ontology (e.g., based on the converted and aggregated training data and the ontology); [0019] As further shown in FIG. 1B, the knowledge graph may include the training data integrated within the ontology as nodes that represent concepts, and edges or links that show interrelationships (e.g., relations) between the concepts. For example, in addition to the information conveyed by the ontology, the knowledge graph may indicate that SARS is an instance of Disease, that ACARBAC is an instance of a protein sequence, that virus_XYZ node is an instance of a Virus, that lungs are an instance of an Organ, and that SARS vaccine is an instance of Vaccine. Thus, the knowledge graph may indicate that SARS is a disease with a vaccine treatment by the SARS vaccine), and a prediction for an explanation for the triple using the model as a function of the input data ([0055] As shown in FIG. 1K, and by reference numbers 110 and 165, the prediction platform may process the ontology and the particular neighborhood with the smallest loss of quality, with a reasoning model, to generate a reasoning graph for the particular candidate drug. In some implementations, the reasoning graph may include a knowledge graph that provides explanations of why the particular candidate drug (e.g., Drug 1) has an effect on SARS Type B and achieved a score of 0.85);
determining a first probability that the model assigns to the triple ([0032] As shown in FIG. 1G, a prediction engine, of the prediction platform, may receive the revised knowledge graph embeddings (e.g., described above in connection with to FIG. 1E) and the candidate drugs (e.g., described above in connection with FIG. 1F). In some implementations, and as shown by reference number 145 in FIG. 1G, the prediction engine may score the candidate drugs based on the revised knowledge graph embeddings. In such implementations, the prediction engine may utilize a relational learning model (e.g., TransE, RESCAL, ComplEx, DistMult, HolE, and/or the like) to determine values associated with the candidate statements (e.g., scores). The prediction engine may then utilize the values to calculate the probability estimates for the candidate drugs) and a second probability that the model assigns to the prediction for the explanation ([0055] the explanations provided in the reasoning graph may become more abstract (e.g., and less relevant) when moving in a particular direction (e.g., to the right) through the reasoning graph. In other words, a degree of abstraction of the explanations, in the reasoning graph, increases when moving to the right through the reasoning graph. Thus, the reasoning graph may be divided into levels of abstraction, with a first level of abstraction (e.g., less abstract and more relevant) being provided to the left of the dashed line in FIG. 1K and a second level of abstraction (e.g., more abstract and less relevant) being provided to the right of the dashed line in FIG. 1K);
determining a classification for the triple as a function of the first probability and of the second probability ([0054] As further shown in FIG. 1J, the particular neighborhood with a smallest loss of quality for the particular candidate drug may include a knowledge graph indicating that Drug 1 has an effect on SARS Type B and is a treatment; and that SARS Type B affects the lungs and is caused by Virus_XYZ. In some implementations, the neighborhood with the smallest loss of quality may include less information than the neighborhood of the particular candidate drug, described above in connection with FIG. 1H); and
when the classification meets a condition ([0061] In some implementations, the reasoning graph may indicate (e.g., at the first level of abstraction) that Drug 1 fights against coronaviruses, SARS Type B is caused by virus_XYZ, and SARS Type B causes a high fever. In some implementations, the reasoning graph may indicate (e.g., at the second level of abstraction) virus_XYZ is a coronavirus, a coronavirus causes a high fever, and a coronavirus affects the lungs):
determining the explanation as a function of the prediction for the explanation and of the triple for the knowledge graph as a function of the first entity, of the prediction for the second entity, and of the prediction for the relationship ([0062] As shown in FIG. 1L, and by reference number 170, the prediction platform may generate, based on the reasoning graph, a text explanation (e.g., for end users) of why the particular candidate drug (e.g., Drug 1) has an effect on SARS Type B and achieved a score of 0.85. In some implementations, the prediction platform may convert the reasoning graph into the text explanation. In some implementations, the text explanation may include two or more different levels of abstraction, such as a first text explanation of the first level of abstraction of the reasoning graph, and a second text explanation of the second level of abstraction of the reasoning graph. For example, as shown in FIG. 1L, the text explanation may indicate that the “prediction that Drug 1 has an effect on SARS Type B (Score: 0.85) is because: (1) Drug 1 fights against coronaviruses, SARS Type B is caused by Virus_XYZ, and SARS Type B causes a high fever; and (2) Virus_XYZ is a coronavirus, a coronavirus causes a high fever, and a coronavirus affects the lungs.”), a function being defined depending on a weighted sum of the first probability and of the second probability and at least one parameter being trained for the model depending on the function ([0021] In some implementations, the fit/train engine may convert entities (e.g., nodes) and relations (e.g., links or edges) of the knowledge graph into points in a k-dimensional metric space. For example, as shown in FIG. 1C, the knowledge graph embeddings may include points in a k-dimensional metric space (e.g., shown as a graph in two dimensions for simplicity). In some implementations, the fit/train engine may minimize a loss function to learn model parameters that best discriminate positive statements from negative statements. In such implementations, the loss function may include a function that maps a statement onto a real number that represents the likelihood of that statement to be true. In such implementations, the loss function may include a pairwise margin-based loss function, a negative log-likelihood loss function, and/or the like. In some implementations, the fit/train engine may assign scores to statements of the knowledge graph in order to aid the loss function in determining how well the knowledge graph tells positive statements from negative statements. In some implementations, the fit/train engine may minimize the loss function in order to determine optimal parameters of the knowledge graph (e.g., the knowledge graph embeddings)), wherein the explanation is an excerpt from the text body ([0062] As shown in FIG. 1L, and by reference number 170, the prediction platform may generate, based on the reasoning graph, a text explanation (e.g., for end users) of why the particular candidate drug (e.g., Drug 1) has an effect on SARS Type B and achieved a score of 0.85. In some implementations, the prediction platform may convert the reasoning graph into the text explanation. In some implementations, the text explanation may include two or more different levels of abstraction, such as a first text explanation of the first level of abstraction of the reasoning graph, and a second text explanation of the second level of abstraction of the reasoning graph. For example, as shown in FIG. 1L, the text explanation may indicate that the “prediction that Drug 1 has an effect on SARS Type B (Score: 0.85) is because: (1) Drug 1 fights against coronaviruses, SARS Type B is caused by Virus_XYZ, and SARS Type B causes a high fever; and (2) Virus_XYZ is a coronavirus, a coronavirus causes a high fever, and a coronavirus affects the lungs.”); and
wherein training data are provided, the training data including a plurality of pairs of a triple and an explanation assigned to the triple (Fig. 1A, 110; [0013] by reference number 110, the user may cause the user device to provide, to the prediction platform, an ontology for the training data; [0055] the reasoning graph may include a knowledge graph that provides explanations of why the particular candidate drug (e.g., Drug 1) has an effect on SARS Type B and achieved a score of 0.85; [0101] The prediction platform may generate a reasoning graph based on the ontology and the particular neighborhood, and may generate an explanation of the particular candidate response based on the reasoning graph), the model including a classifier that is trained as a function of the training data for determining for the first entity from the triple ([0059] The OWL model may include a family of knowledge representation languages for authoring ontologies. Ontologies are a formal way to describe taxonomies and classification networks, essentially defining a structure of knowledge for various domains; [0060] the prediction platform may utilize one or more of the reasoning models to generate the reasoning graph for the particular candidate drug; Fig. 4; [0087] FIG. 4 is a flow chart of an example process 400 for determining explanations for predicted links in knowledge graphs).
Regarding independent claim 13, it is a device claim that corresponding to the method of claim 1. Therefore, it is rejected for the same reason as claim 1 above. COSTABELLO further teaches a device including at least one processor of a machine learning model for determining a knowledge graph ([0003] According to some implementations, a device may include one or more memories, and one or more processors, communicatively coupled to the one or more memories; Fig. 2, 220; [0068] Prediction platform 220 includes one or more devices that determine explanations for predicted links in knowledge graphs. In some implementations, prediction platform 220 may be designed to be modular such that certain software components may be swapped in or out depending on a particular need. As such, prediction platform 220 may be easily and/or quickly reconfigured for different uses. In some implementations, prediction platform 220 may receive information from and/or transmit information to one or more user devices 210)
Regarding independent claim 14, it is a medium claim that corresponding to the method of claim 1. Therefore, it is rejected for the same reason as claim 1 above. COSTABELLO further teaches a non-transitory machine-readable storage medium on which is stored a computer program for determining a knowledge graph, the computer program, when executed by at least one processor of a machine learning model, causing the at least one processor to perform the operations ([0004] According to some implementations, a non-transitory computer-readable medium may store instructions that include one or more instructions that, when executed by one or more processors, cause the one or more processors to receive a knowledge graph generated based on training data and an ontology for the training data, where the training data may include information associated with a subject of the ontology. The one or more instructions may cause the one or more processors to receive a query for information associated with the knowledge graph, and generate candidate responses to the query based on the knowledge graph. The one or more instructions may cause the one or more processors to identify a particular candidate response, of the candidate responses, based on scoring the candidate responses based on the knowledge graph, and determine, based on the knowledge graph, a neighborhood of the particular candidate response. The one or more instructions may cause the one or more processors to generate knowledge graph embeddings for the neighborhood of the particular candidate response, and identify, based on the knowledge graph embeddings, a portion of the neighborhood with a smallest loss of quality. The one or more instructions may cause the one or more processors to generate a reasoning graph based on the ontology and the portion of the neighborhood, and generate an explanation of the particular candidate response based on the reasoning graph; Fig. 3; [0081] Storage component 340 stores information and/or software related to the operation and use of device 300. For example, storage component 340 may include a hard disk (e.g., a magnetic disk, an optical disk, a magneto-optic disk, and/or a solid state disk), a compact disc (CD), a digital versatile disc (DVD), a floppy disk, a cartridge, a magnetic tape, and/or another type of non-transitory computer-readable medium, along with a corresponding drive.).
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 2-4 and 6-12 are rejected under 35 U.S.C. 103 as being unpatentable over COSTABELLO as applied in claim 1, in view of Zhou et al. (hereinafter Zhou), US 11,132,994 B1.
Regarding dependent claim 2, COSTABELLO teaches all the limitations as set forth in the rejection of claim 1 that is incorporated. COSTABELLO does not explicitly disclose
providing the second entity and the relationship;
determining a first measure, which characterizes a difference between two probability distributions, between the prediction for the second entity and the second entity;
determining a second measure, which characterizes a difference between two probability distributions, between the prediction for the relationship and the relationship;
determining a third measure, which characterizes a difference between two probability distributions, of a weighted third cross entropy between the prediction for the explanation and the explanation, the function being defined depending on the first measure, of the second measure, and of the third measure and of a weighted sum of the first probability and of the second probability.
However, in the same field of endeavor, Zhou teaches
providing the second entity and the relationship (Col 21 lines 6-17 As shown in FIG. 10, the question generation component 965 may determine word-level embeddings and character-level embeddings for the domain name and the slot name (1060) and perform an element-wise addition for the domain name embeddings and the slot name embeddings to determine domain-slot embeddings (1064). The value prediction component 960 may determine word-level embeddings and char-level embeddings for a potential slot value v1 (1062), perform an element-wise addition for the value v1 and the domain-slot embeddings (1064), and process the resulting embedding using a neural network or other type of machine learning model.);
determining a first measure, which characterizes a difference between two probability distributions, between the prediction for the second entity and the second entity (Col 22, lines 21-35 the value prediction component 960 calculates the score of each value v at turn t byp t v=Softmax(BiLinearΦ 1 (B q ,B C T ·αb)where αb=Attβ 2 (Bc, wd+ws) ∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
L c is the attention score over Bc, and pt v∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
L v ; The value prediction component 960 calculates the cross entropy loss of the predicted scores by Lossv=ΣtΣd∈D,s∈Ŝ d CrossEntropy (pt v,yt v) where yt v∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
L v is the label, which is the one-hot encoding of the true value of domain d and slot s, and Ŝd is the set of slots in domain d that has pre-defined Vs. The value prediction component 960 may thus determine, using the softmax output 1068, probability distributions for the potential slot values);
determining a second measure, which characterizes a difference between two probability distributions, between the prediction for the relationship and the relationship (Col 22, lines 53-67; Col 23 lines 1-2 When the value set Vs is unknown or too large to enumerate, such as pick up time in taxi domain, the span prediction component 980 predicts the answer to a question Qd,s as either a span in the dialog context data 905 or two special types: ‘not mentioned’ and ‘don't care’. The span prediction layer has two components. The span prediction component 980 first predicts the answer type of Qd,s. The type of the answer is either ‘not mentioned’, ‘don't care’ or ‘span’, and is calculated byp t st=Softmax(φ1·(w d +w s +E c T ·αe))
where αe=Attβ 3 (Ec, dw+ws)∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
L c , Θ1 ∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
3*D w is a model parameter to learn, and pt st
PNG
media_image1.png
33
33
media_image1.png
Greyscale
3. The loss of span type prediction is Lossst=ΣtΣd∈D,s∈Ŝ d CrossEntropy (pt st,yt st) where yt st∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
3 is the one-hot encoding of the true span type label, and S d is the set of slots in domain d that has no predefined Vs);
determining a third measure, which characterizes a difference between two probability distributions, of a weighted third cross entropy between the prediction for the explanation and the explanation, the function being defined depending on the first measure, of the second measure, and of the third measure and of a weighted sum of the first probability and of the second probability (Col 23, lines 3-20 To get the probability distribution of a span's start index, the span prediction component 980 applies a bilinear function between the dialog context data and the (domain, slot) pair. More specifically,p t st=Softmax(BiLinearΦ 2 (Relu(E c·Θ2),(w d +w s +E C T ·αe)))
where Θ2∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
D w *D w and pt st∈∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
L c . The bilinear function's first argument is a non-linear transformation of the context embedding, and its second argument is a context-dependent (domain, slot) pair embedding. The prediction loss is Lossss=ΣtΣd∈D,s∈S d CrossEntropy(p t ss ,y t ss) where yt ss∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
L c is one-hot encodings of true start indices. The probability distribution of a span's end index pt se and the loss Lossse is calculated in a similar way. The score of a span is the multiplication of probabilities of its start and end index. The final loss function is: Loss=Lossv+Lossst+Lossss+Lossse. The span prediction component 980 may thus determine, using the softmax output 1088, probability distributions of the span start and end for the slot value, if the slot value is determined to not be one of the special values (‘not mentioned’ or ‘don't care’)).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of the calculation of the cross entropy loss of the predicted scores as suggested in Zhou into COSTABELLO’s system because both of these systems are addressing determining a knowledge graph to track predicted pair values. This modification would have been motivated by the desire to enhance the model’s performance (Zhou, Col 3, lines 51-53).
Regarding dependent claim 3, the combination of COSTABELLO and Zhou teaches all the limitations as set forth in the rejection of claim 2 that is incorporated. Zhou further teaches wherein the function is also defined depending on a sum of the first measure, the second measure, and the third measure (Col 23, lines 22-28 The final loss function is: Loss=Lossv+Lossst+Lossss+Lossse. The span prediction component 980 may thus determine, using the softmax output 1088, probability distributions of the span start and end for the slot value, if the slot value is determined to not be one of the special values (‘not mentioned’ or ‘don't care’)).
Regarding dependent claim 4, the combination of COSTABELLO and Zhou teaches all the limitations as set forth in the rejection of claim 2 that is incorporated. Zhou further teaches wherein:
the first measure is at least one from a cross entropy, a Kullback-Leibler divergence, and an f-divergence (Col 22, lines 27-32 The value prediction component 960 calculates the cross entropy loss of the predicted scores by Lossv=ΣtΣd∈D,s∈Ŝ d CrossEntropy (pt v,yt v) where yt v∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
L v is the label, which is the one-hot encoding of the true value of domain d and slot s, and Ŝd is the set of slots in domain d that has pre-defined Vs), and/or
the second measure is at least one from a cross entropy, a Kullback-Leibler divergence, and an f-divergence (Col 22, lines 65-67 The loss of span type prediction is Lossst=ΣtΣd∈D,s∈Ŝ d CrossEntropy (pt st,yt st) where yt st∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
3 is the one-hot encoding of the true span type label, and S d is the set of slots in domain d that has no predefined Vs), and/or
the third measure is at least one from a cross entropy, a weighted cross entropy of a Kullback-Leibler divergence, and an f-divergence (Col 23, lines 14-20 The prediction loss is Lossss=ΣtΣd∈D,s∈S d CrossEntropy(p t ss ,y t ss) where yt ss∈
PNG
media_image1.png
33
33
media_image1.png
Greyscale
L c is one-hot encodings of true start indices. The probability distribution of a span's end index pt se and the loss Lossse is calculated in a similar way).
Regarding dependent claim 6, the combination of COSTABELLO and Zhou teaches all the limitations as set forth in the rejection of claim 2 that is incorporated. Zhou further teaches wherein a vector representation that defines at least a portion of the input data is determined for at least one word of the text body or for at least one sentence of the text body, as a function of at least one other word or as a function of at least one other sentence (Col 15, lines 49-59 A word sequence is usually represented as a series of one-hot vectors (i.e., a N-sized vector representing the N available words in a lexicon, with one bit high to represent the particular word in the sequence). The one-hot vector is often augmented with information from other models, which have been trained on large amounts of generic data, including but not limited to word embeddings that represent how individual words are used in a text corpus, labels from a tagger (e.g., part-of-speech (POS) or named entity tagger), labels from a parser (e.g., semantic or dependency parser), etc.).
Regarding dependent claim 7, the combination of COSTABELLO and Zhou teaches all the limitations as set forth in the rejection of claim 6 that is incorporated. Zhou further teaches wherein a first vector is assigned to a first word from a sentence from the text body, a second vector is assigned to a second word from the sentence of the text body, the vector representation being computed as a weighted sum from the first vector and the second vector (Col 14, lines 61-67, Col 15, lines 1-2 FIG. 7 illustrates feature data values 702-706 being processed by an encoder 750 to generate an encoded feature vector y. In mathematical notation, given a sequence of feature data values x1, . . . xn, . . . xN, with xn being a D-dimensional vector, an encoder E(x1, . . . xN)=y projects the feature sequence to y, with y being a F-dimensional vector. F is a fixed length of the vector and is configurable depending on user of the encoded vector and other system configurations; Col 15, lines 62-67, Col 16, lines 1-12 A word embedding is a representation of a word in the form of a multi-dimensional data vector, where each dimension represents some data point regarding the word, its usage, or other information about the word. To create word embeddings a text corpus is obtained that represents how words are used. The text corpus may include information such as books, news content, internet articles, etc. The system then creates multiple vectors, each corresponding to the usage of a particular word (e.g., the usage of the word in the source text corpus), and map those vectors into a vector space. Given the number of dimensions for each vector, the vector space may be a high dimensional vector space. The different dimensions for each data vector may correspond to how a word is used in the source text corpus. Thus the word embedding data represented in the individual values of a word embedding data vector may correspond to how the respective word is used in the corpus).
Regarding dependent claim 8, the combination of COSTABELLO and Zhou teaches all the limitations as set forth in the rejection of claim 2 that is incorporated. COSTABELLO further teaches wherein an output including the triple is output at a first output of the model ([0055] As shown in FIG. 1K, and by reference numbers 110 and 165, the prediction platform may process the ontology and the particular neighborhood with the smallest loss of quality, with a reasoning model, to generate a reasoning graph for the particular candidate drug. In some implementations, the reasoning graph may include a knowledge graph that provides explanations of why the particular candidate drug (e.g., Drug 1) has an effect on SARS Type B and achieved a score of 0.85).
Regarding dependent claim 9, the combination of COSTABELLO and Zhou teaches all the limitations as set forth in the rejection of claim 2 that is incorporated. COSTABELLO further teaches wherein an output that defines a start and an end of at least one section in the text body is output at a second output of the model ([0012] The prediction platform may generate a reasoning graph based on the ontology and the particular neighborhood, and may generate