DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is responsive to the Amendment filed on April 20, 2026. Claims 1, 10, and 11 are amended. Claims 1-20 are pending in the case. Claims 1, 10, and 11 are the independent claims.
This action is final.
Applicant’s Response
In the response filed on April 20, 2026, Applicant amended the claims and provided arguments in response to the interpretation of the claims under 35 USC 112(f) and rejections of the claims under 35 USC 101 and 103 in the previous office action.
Response to Argument/Amendment
Applicant’s amendments to the claims in response to the interpretation of the claims under 35 USC 112(f) in the previous office action are acknowledged, and Applicant’s associated arguments have been fully considered. Applicant’s amendments appear to recite structure for implementing the various recited modules of the explanation engine. Therefore, the interpretation appears to be moot, and is withdrawn.
Applicant’s amendments to the claims in response to the rejection of the claims under 35 USC 101 in the previous office action are acknowledged, and Applicant’s associated arguments have been fully considered.
Applicant argues that the claims do not recite a mathematical concept. In addition, Applicant argues that the claims do not recite a mere mental process because “[e]valuating the layers of a hierarchical machine learning architecture and executing ablation cycles to alter inputs and calculate impacts on intermediate and top-level results are not merely mental steps or disembodied mathematical concepts. A human mind cannot feasibly trace through the complex, multi-layered hidden networks of a machine learning reasoning process to calculate the weighted impact of individual nodes. This is a process performed by the explanation engine interacting directly with the computational structure of the reasoning engine. This is fundamentally tied to the physical operation and computational execution of machine learning models, not a disembodied mental process.”
This argument is not persuasive. First, Examiner notes that the combination of limitations argued by Applicant do not appear to be recited in the independent claims, and therefore Applicant’s arguments do not appear to apply to the independent claims or various of the dependent claims which do not recite these limitations. For example, Applicant argues that executing ablation cycles to alter inputs and calculate impacts on intermediate and top level results are not merely mental steps or disembodied mathematical concepts. However, nothing in the independent claims recites any such ablation process. Moreover, of the dependent claims that do recite an ablation module, only claims 7, 8, 18, and 19 appear to recite that it is to alter inputs to layers, and only claims 8 and 19 appear to indicate that it is used to determine an effect on top layer results. Therefore, Applicant’s arguments regarding the ablation process do not appear to be applicable to the majority of the claims. With respect to claims 8 and 19, Examiner notes that the claims only require conducting ablation cycles to alter input layers of a machine-based reasoning process to determine an effect of that layer on a top level result. Nothing in the claims appears to actually require that the recited machine-based reasoning process be a complex hierarchical machine learning architecture, as opposed to a reasoning process which is machine-based. A human can mentally determine alterations of inputs to layers of reasoning processes and determine effects of such alterations, absent any further definition of the reasoning process, even if the reasoning process is based in some way on a machine.
Applicant further argues that evaluating layers of a hierarchical machine learning architecture and tracing through “complex, multi-layered hidden networks of a machine learning reasoning process to calculate the weighted impact of individual nodes,” requiring direct interaction with the computational structure of the reasoning engine and being tied to physical operation/execution of machine learning models, cannot be performed in the human mind. However, the independent claims do not appear to recite the combination of limitations as argued. The independent claims only recite that there is an explanation engine having cooperating modules configured to evaluate layers in a hierarchical architecture of a machine-based reasoning process that uses machine learning. Other than indicating that the reasoning process has layers and has a hierarchical architecture (i.e. the architecture of the reasoning process may have at least two layers with an ordering/organization), nothing in the independent claims appears to require any particular level of complexity for the architecture of the reasoning process. Moreover, nothing requires that the architecture of the reasoning process include any networks, hidden or otherwise. While the independent claim do recite that the process “uses” machine learning, they do not provide any detail regarding this use, such that this use would require machine learning or neural network computational structures within the reasoning process itself, as opposed to only being an external component which is utilized for some purposes. The independent claims do not even require that the reasoning process itself be executed on any machine or computer. Instead the claims only recite that the reasoning process is “machine-based.” Therefore, as noted above, the independent claims only recite that the explanation engine/its modules evaluate layers in a hierarchical architecture of a reasoning process which is machine based and “uses” machine learning, for the intended purpose of supporting an explanation of how the reasoning process arrived at its reported top-level and intermediate results. A human is capable of mentally evaluating a reasoning process having at least two ordered layers in order to support explanation of how the process arrives at top level and intermediate results, even if that process may be “machine-based” in some manner and even if that process might “use” machine learning in some manner, absent any further requirements as to the implementation of the process.
Applicant further argues that the claims integrate any alleged abstract idea into a practical application. “Paragraph 0003 of the specification details how the claims are directed to solving specific technical problems inherent in artificial intelligence, specifically the ‘black box’ problem where post-hoc explanations do not match the actual factors causing the system performance.” Applicant argues that the claims achieve this with “an explanation engine utilizes a set of modules cooperating to evaluate the specific layers in a hierarchical architecture of a machine-based reasoning process,” “a crawl back module and an ablation module interact with the intermediate layers of the reasoning process, altering inputs to determine the exact quantitative effect of that specific layer on the top-level result,” and “a messaging module collects these intermediate output results and maps them to problem-specific or domain-specific terminology.” Applicant further argues that this combination does not just perform generic data gathering or mathematical calculations, and instead executes a highly specific ablation and mapping protocol on the internal architecture of a machine learning model to translate opaque intermediate computational results into understandable domain terminology, where paragraph 0022 of the specification explains that this constitutes a practical application that effects an improvement in technology of artificial intelligence by allowing users to recognize when automated reasoning is in error.
These arguments are not persuasive. Applicant’s arguments are made with respect to limitations which are not actually reflected in the claims Although Applicant argues that the claims solve specific problems inherent in artificial intelligence, as discussed above, the claims only recite that the reasoning process is “machine based” and that it “uses” machine learning. Therefore, as presently claimed, the use of the recited explanation engine and messaging model does not appear to be targeted to artificial intelligence and machine learning models specifically, and is instead directed to any “reasoning process” that can be considered “machine based” (i.e. based on a “machine” in any manner) and to “use” machine learning (i.e. even if that use is incidental to the actual reasoning process and performed by components which are external from the claimed apparatus, reasoning process, explanation engine, etc.). Nothing in the claims appears to require evaluation of any “specific layer” of the reasoning process. Moreover, as previously discussed, the concepts of the crawl back module and ablation module are not recited in the independent claims at all. Only claims 8 and 19 recite the ablation module altering inputs to a layer of the reasoning process to determine an effect on the top-level result. However, even these claims do not appear to require that this be an interaction with intermediate layers of the reasoning process “to determine the exact quantitative effect of that specific [intermediate] layer.” With respect to the messaging module, while the independent claims do recite that it collects the top-level result as well as one or more intermediate output results, nothing in the independent claims appears to also recite or otherwise require that the messaging module also “maps” them to problem-specific or domain-specific terminology. Instead, the independent claims only recite that the intermediate output results collected by the messaging module are “associated with” such terminology. Examiner again notes that the independent claims and the vast majority of the dependent claims do not recite the “ablation and mapping protocol” argued by Applicant at all. With respect to claims 8 and 19, these claims only recite conduction one or more ablation cycles with respect to the machine-based reasoning process. Nothing in these claims appears to further require that this entail any sort of processing involving “the internal architecture of a machine learning model,” or that the intermediate output results be “opaque.” Therefore, even if the various technological solutions and improvements to technology are described in the specification, such solutions or improvements (or the components which realize such solutions and improvements) do not appear to be reflected in the claims as presently recited.
Applicant argues that the specific combination of elements in the claims provides an inventive concept, including (i) specific architectural interaction in which the interaction between a reasoning engine that breaks its process down into divisible layers and an explanation engine that evaluates those specific layers goes beyond the routine and conventional, (ii) ablation and extraction combination in which the ablation module selectively removes or alters layers while a crawl back module extracts the resulting intermediate data represents a non-conventional way to achieve transparency in machine learning systems and transforms the process from a simple input-output calculation into an interpretable analytical system, and (iii) a technological improvement as indicated in paragraph 0026 of the specification, where the overall system improves the underlying technology of automated reasoning by providing a reliable mechanism to build trust and verify the outputs of a machine learning process. Applicant concludes that these features provide significantly more than merely applying an abstract idea using generic computers and instead provide specific inventive solutions to technical problems in the artificial intelligence field.
For similar reasons to those already discussed, these arguments are also not persuasive (i.e. Applicant’s arguments are made with respect to limitations which are not actually reflected in the claims). The independent claims do not appear to recite any “reasoning engine that breaks its process down into divisible layers,” or any “specific architectural interaction” between such a reasoning engine and the explanation engine. While claims 4 and 14 do recite a first reasoning engine configured to break down its machine based reasoning process into divisible layers, this does not appear to be the same reasoning process as is recited in the independent claims. Moreover, even these claims do not a appear to recite that the explanation engine actually “evaluates those specific layers,” instead only reciting that the explanation engine “is configured to cooperate with” the first reasoning engine and allow a user to query intermediary output results. The majority of the claims do not recite any use of any “ablation and extraction combination.” Instead, in the dependent claims which do actually recite the ablation module and the crawl back module, only one of the two is recited (see claims 6 and 16, reciting the crawl back module without the ablation module; see claims 7, 8, and 17-19, reciting the ablation module without the crawl back module). Claims 5 and 15 appear to be the only claims which actually recite that the crawl back module and ablation module are configured to cooperate, and these claims only recite that this is to trace through intermediate layers of the reasoning process to record factors considered and importance in arriving at the top level result. These claims do not recite that the “ablation module selectively removes or alters layers while a crawl back module extracts the resulting intermediate data” as argued by Applicant. Finally, while paragraph 0026 does appear to describe some benefit of the overall system described in the specification of the instant application, it only appears to indicate that some “modules” of the explanation engine provide this benefit, with no indication of which particular modules or functions of the explanation engine provide the benefit, or any details regarding how the modules of functions of the explanation engine provide such benefits. Therefore, the paragraph appears to describe the idea of the solution instead of the implementation of the solution, and it cannot be said that such a solution or benefit is actually reflected as presently recited in the claims.
Applicant’s amendments to the claims in response to the rejection of the claims under 35 USC 103 in the previous office action are acknowledged, and Applicant’s associated arguments have been fully considered. Applicant notes that the independent claims have been amended to recite “a messaging module of the explanation engine configured to collect the top-level result as well as one or more intermediate output results from intermediate layers of the machine-based reasoning process, where the one or more intermediate output results from the multiple layers of reasoning are associated with terminology…” Applicant argues that although Forsyth is cited as teaching associating layers of reasoning with terminology, Forsyth “merely maps the final output of its neural network model to a text lexicon….its intermediate hidden layers…output intermediate vector outputs, but these intermediate outputs are strictly mathematical and are not associated with any human-readable terminology….Forsyth relies entirely on the final activation scores of the top k items to provide its explanations….Forsyth fails to teach associating the intermediate output results from multiple layers of reasoning with domain-specific or problem-specific terminology. Forsyth does not disclose ‘the one or more intermediate output results from the multiple layers of reasoning are associated with terminology use in (i) a problem…or (ii) a domain…’ as recited in the claim language.” Applicant further notes that does not disclose associating its intermediate outputs with domain- or problem-specific terminology and therefore does not cure this deficiency, because its explanatory rules are formulated in terms of mathematical and structural properties of the neural network itself and Chatterjee “entirely fails to teach associating these intermediate output results with terminology used in a problem to be solved or a domain pertinent to the problem. Chatterjee maps to mathematical and structural properties of the neural network itself and not to terminology of the problem statement.”
Applicant’s arguments are persuasive, and the rejection under 35 USC 103 is withdrawn. However, new grounds of rejection are provided below.
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 1-9 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claims do not fall within at least one of the four categories of patent eligible subject matter because the claims are directed to software per se. Independent claim 1 recites “an explanation engine having a set of modules…and a messaging module of the explanation engine…wherein the set of modules, including the messaging module, are implemented in i) software stored in one or more non-transitory machine readable mediums configured to be executed by one or more processors….” This appears to recite that the execution engine and its various modules are implemented in software, and further appears to describe that the software is stored in non-transitory machine-readable mediums configured to be executed by one or more processors. However, Examiner notes that the claim does not positively recite that the claimed apparatus itself comprises the non-transitory machine-readable mediums or the processors, and instead only appears to recite that the claimed apparatus comprises the explanation engine having the set of modules (which, as noted, are themselves implemented in software, in at least one recited configuration). In addition, while the claim additionally recites that the modules may also be implemented in logic gates, electronic circuits, or any combination of these, these appear to be recited as alternative limitations. Therefore, independent claim 1 appears to recite at least one embodiment which can be implemented entirely in software, such that the claimed invention of claims 1-9 is therefore is not a machine, an article of manufacture, a process, or a composition of matter as contemplated ty 35 U.S.C. 101.
Examiner respectfully suggests amending the claims to clarify that the claimed invention is not purely software, i.e., to positively recite that the apparatus itself comprises the structural components which implement the modules of claims 1-10, such as the processors and mediums, such that claims 1-10 no longer recite any embodiment which is a complete software embodiment or realized entirely by software.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claims recite mental processes and mathematical concepts. This judicial exception is not integrated into a practical application because the claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception, as explained below.
Step 1 for all Claims:
Claims 1-10 are directed to a machine. Claims 11-20 are directed to a method (process). Therefore, Claims 1-20 are directed to one of the four statutory categories of invention, i.e., process, machine, manufacture, or composition of matter.
Regarding Claim 1:
Step 2A, Prong 1:
[An apparatus, comprising: an explanation engine having a set of modules cooperating with each other configured to] evaluate layers in a hierarchical architecture of a machine based reasoning process that uses machine learning to support an explanation of how the machine-based reasoning process arrived at its reported results of both a top-level result as well as corresponding intermediate output results, As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses evaluating layers to provide an explanation which is making an evaluation based upon the inputs and results of the machine-based reasoning process which can be feasibly performed in the human mind (see MPEP 2106.04(a)(2)(III)).
Step 2A, Prong2:
An apparatus, comprising: an explanation engine having a set of modules cooperating with each other configured to…wherein the set of modules, including the messaging module, are implemented in i) software stored in one or more non-transitory machine-readable mediums configured to be executed by one or more processors; ii) logic gates configured to receive an input, perform a desired functionality, and output results; iii) electronic circuits; and iv) any combination of these. This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
and a messaging module of the explanation engine configured to collect the top- level result as well as one or more intermediate output results from intermediate layers of the machine-based reasoning process, where the one or more intermediate output results from the multiple layers of reasoning are associated with terminology used in at least one of i) a problem to be solved and ii) a domain pertinent to the problem in order to communicate how the machine-based reasoning process came to its reported results in a communication. This limitation amounts to extra-solution activity of gathering data for use in the claimed process. As described in MPEP2106.05(g), limitations that amount to merely adding insignificant extra-solution activity to a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Step 2B:
An apparatus, comprising: an explanation engine having a set of modules cooperating with each other configured to… wherein the set of modules, including the messaging module, are implemented in i) software stored in one or more non-transitory machine-readable mediums configured to be executed by one or more processors; ii) logic gates configured to receive an input, perform a desired functionality, and output results; iii) electronic circuits; and iv) any combination of these. This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
and a messaging module of the explanation engine configured to collect the top- level result as well as one or more intermediate output results from intermediate layers of the machine-based reasoning process, where the one or more intermediate output results from the multiple layers of reasoning are associated with terminology used in at least one of i) a problem to be solved and ii) a domain pertinent to the problem in order to communicate how the machine-based reasoning process came to its reported results in a communication. As discussed above, the additional elements of collecting the top-level result and intermediate output results which is recited at a high level of generality and amounts to extra-solution activity of gathering data. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), "receiving or transmitting data over a network", "electronic record keeping", and "storing and retrieving information in memory").
Regarding Claim 2:
Step 2A, Prong 1:
[The apparatus of claim 1, where the explanation engine has a terminology module configured to] assign terminology from any of i) the domain pertinent to the problem and ii)the specific problem to be solved, for the multiple layers in the hierarchical architecture of the machine-based reasoning process supplied from a reasoning engine, where the user is able to understand the results in terms of the specific problem or domain based on the way the communication is generated. As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses assigning terminology from a pertinent domain to the reasoning process to help the user to understand the results which is making observations or opinions based on descriptive terminology pertinent to the field of study in order to provide an explanation, which can be feasibly performed in the human mind (see MPEP 2106.04(a)(2)(III)).
Step 2A, Prong 2:
The apparatus of claim 1, where the explanation engine has a terminology module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Step 2B:
The apparatus of claim 1, where the explanation engine has a terminology module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Regarding Claim 3:
Step 2A, Prong 1:
and where the terminology module is configured to crawl through the hierarchical architecture of the machine-based reasoning process, to be created by a reasoning engine, and then associate i) the terminology specific to the problem to be solved supplied by the user and/or terminology specific to a relevant subject matter domain with ii) the multiple layers making up the hierarchical architecture of the machine-based reasoning process. As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses crawling through the reasoning process and associating terminology to layers in the reasoning process which is making a judgement based upon how well certain terminology match with certain layers in the reasoning process which can be feasibly performed in the human mind (see MPEP2106.04(a)(2)(III)).
Step 2A, Prong 2:
The apparatus of claim 1, where the explanation engine has a terminology module of the explanation engine configured to accept input of terminology for the problem to be solved that is supplied by at least one of i) a description of the problem to be solved ii) a description of preferred approach to solve the problem from a user, and iii) a database ofknown terminology specific to the domain pertinent to the problem, This limitation amounts to extra solution activity of gathering data for use in the claimed process. As described in MPEP2106.05(g), limitations that amount to merely adding insignificant extra-solution activity to a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Step 2B:
The apparatus of claim 1, where the explanation engine has a terminology module of the explanation engine configured to accept input of terminology for the problem to be solved that is supplied by at least one of i) a description of the problem to be solved ii) a description of preferred approach to solve the problem from a user, and iii) a database of known terminology specific to the domain pertinent to the problem, As discussed above, the additional elements of accepting terminology which is recited at a high level of generality and amounts to extra-solution activity of gathering data. The courts have found limitations directed to obtaining information electronically, recited at a high level ofgenerality, to bewell-understood, routine, and conventional (see MPEP2106.05(d)(II), "receiving or transmitting data over a network", "electronic record keeping", and "storing and retrieving information in memory").
Regarding Claim 4:
Step 2A, Prong 1:
breakdown its machine-based reasoning process into divisible layers that provide intermediary output results to other layers in order to determine the top level result from the machine-based reasoning process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses breaking down the reasoning process into layers to determine a top-level result which is making an observation and then a judgement based upon mentally categorizing parts of the reasoning process and then arriving at a result by observing the intermediate results which can be feasibly performed in the human mind (see MPEP2106.04(a)(2)(III)).
Step 2A, Prong2:
The apparatus of claim 1, where the explanation engine is configured to cooperate with a first reasoning engine that is configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a toolto perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
as opposed to a second reasoning engine that is configured to create one omnibus neural network that is compiled as a black box that merely outputs its final decision; This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The final decision outputting is recited at a high-level of generality with no detail of the decision-making process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
and where the explanation engine is configured to cooperate with the first reasoning engine to allow a user to query what the intermediary output results are for each layer of the machine-based reasoning process as well as what would happen when the intermediary output results were altered. This limitation amounts to extra-solution activity of gathering data and outputting for use in the claimed process. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity to a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Step 2B:
The apparatus of claim 1, where the explanation engine is configured to cooperate with a first reasoning engine that is configured to...This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
as opposed to a second reasoning engine that is configured to create one omnibus neural network that is compiled as a black box that merely outputs its final decision; This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with thejudicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a toolto perform an abstract idea. The final decision outputting is recited at a high-level of generality with no detail of the decision-making process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP2106.05(f)).
and where the explanation engine is configured to cooperate with the first reasoning engine to allow a user to query what the intermediary output results are for each layer of the machine-based reasoning process as well as what would happen when the intermediary output results were altered. As discussed above, the additional elements of accepting user queries which is recited at a high level of generality and amounts to extra-solution activity of gathering data. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), "receiving or transmitting data over a network", "electronic record keeping", and "storing and retrieving information in memory").
Regarding Claim 5:
Step 2A, Prong 1:
[The apparatus of claim 1, where the explanation engine has a crawl back module configured to]cooperate with an ablation modulet o trace through the intermediate layers of the machine-based reasoning process constructed by a reasoning engine o record factors being considered and how important that factor was into arriving at the top-level result from the machine-based reasoning process. As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses tracing through the intermediate layers of a reasoning process and making record of important factors which is making observations of the plurality of layers and making note of the importance of the factors which can be feasibly performed in the human mind (see MPEP 2106.04(a)(2)(III)).
Step 2A, Prong2:
The apparatus of claim 1, where the explanation engine has a crawl back module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Step 2B:
The apparatus of claim 1, where the explanation engine has a crawl back module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Regarding Claim 6:
Step 2A, Prong 2:
[The apparatus of claim 1, where the explanation engine has a crawl back module configured to] cooperate with the messaging module, where the crawl back module of the explanation engine is configured to crawl through a decomposition of the machine- based reasoning process to collect and then report the intermediate output results from the multiple layers of the reasoning process to explain the top-level result in terms of the intermediate output results. This limitation amounts to extra-solution activity of gathering data for use in the claimed process. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity to a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
The apparatus of claim 1, where the explanation engine has a crawl back module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Step 2B:
The apparatus of claim 1, where the explanation engine has a crawl back module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent)with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
[The apparatus of claim 1, where the explanation engine has a crawl back module configured to] cooperate with the messaging module, where the crawl back module of the explanation engine is configured to crawl through a decomposition of the machine- based reasoning process to collect and then report the intermediate output results from the multiple layers of the reasoning process to explain the top-level result in terms of the intermediate output results. As discussed above, the additional elements of collecting the top level result and intermediate output results and reporting them are recited at a high level of generality and amount to extra-solution activity of gathering data. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP2106.05(d)(II), "receiving or transmitting data over a network", "electronic record keeping", and "storing and retrieving information in memory").
Regarding Claim 7:
Step 2A, Prong 1:
[The apparatus of claim 1, where the explanation engine has an ablation module configured to] change the intermediate output results from layers of the machine-based reasoning process by altering an input for that layer and then output a new intermediate output result from that layer of the machine-based reasoning process as well as a new top level result. As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses changing inputs for layers to get different intermediate output results which is making a prediction or calculation based upon altering he initial circumstances in a hypothetical scenario which can be feasibly performed in the human mind (see MPEP 2106.04(a)(2)(III)).
Step 2A, Prong2:
The apparatus of claim 1, where the explanation engine has an ablation module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Step 2B:
The apparatus of claim 1, where the explanation engine has an ablation module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP2106.05(f)).
Regarding Claim 8:
Step 2A, Prong 1:
conduct one or more ablation cycles to alter an input to a layer of the machine-based reasoning process created by a reasoning engine to determine an effect of that layer on the top-level result and record the effect; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses determining and recording an effect of a layer on the result which is making a judgement and memory based upon observing the differences between the beginning and end of the ablation cycles and considering the input alterations which can be feasibly performed in the human mind (see MPEP 2106.04(a)(2)(III)).
Step 2A, Prong2:
The apparatus of claim 1, further comprising: an ablation module configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component(See MPEP2106.05(f)).
and where the messaging module is configured to take results of the ablation cycles and data generated with them in order to generate the reported results of an impact of each layer of machine-based reasoning process in the communication generated by the messaging module. This limitation amounts to extra-solution activity of gathering outputting for use in the claimed process. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity to a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Step 2B:
The apparatus of claim 1, further comprising: an ablation module configured to... This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component(See MPEP2106.05(f)).
and where the messaging module is configured to take results of the ablation cycles and data generated with them in order to generate the reported results of an impact of each layer of machine-based reasoning process in the communication generated by the messaging module. As discussed above, the additional elements of taking results and data to generate a report which is recited at a high level of generality and amounts to extra-solution activity of transmitting data. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), "receiving or transmitting data over a network", "electronic record keeping", and "storing and retrieving information in memory").
Regarding Claim 9:
Step 2A, Prong 1:
[The apparatus of claim 1, further comprising: where the messaging module of the explanation engine is configured to] 1) extract the intermediate output results from the multiple layers of the machine-based reasoning process created by a reasoning engine and 2) cooperate with a terminology module to associate the intermediate output results from the multiple layers with the terminology taken from the at least one of i) subject domain pertinent to the problem and ii) the problem specific terminology used in the problem to be solved. As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses extracting intermediate results and associating those results with terminology which is making a judgement based upon analyzing the multiple layers of the reasoning process and then matching results to their appropriate terminology which can be feasibly performed in the human mind (see MPEP 2106.04(a)(2)(III)).
Step 2A, Prong 2:
The apparatus of claim 1, further comprising: where the messaging module of the explanation engine is configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception usinga generic computer component (See MPEP2106.05(f)).
Step 2B:
The apparatus of claim 1, further comprising: where the messaging module of the explanation engine is configured to… This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The configuration is recited at a high-level of generality with no detail of the configuration process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Regarding Claim 10:
Step 2A, Prong 1:
causing an explanation engine having a set of modules to evaluate layers in a hierarchical architecture of a machine-based reasoning process that uses machine learning to support an explanation of how the machine-based reasoning process arrived at its reported results of both a top-level result as well as corresponding intermediate output results, As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses evaluating layers to provide an explanation which is making an evaluation based upon the inputs and results of the machine-based reasoning process which can be feasibly performed in the human mind (see MPEP 2106.04(a)(2)(III)).
Step 2A, Prong 2:
A non-transitory computer-readable medium including executable instructions that, when executed with one or more processors, cause an explanation engine to perform operations as follows, comprising: This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The performing of operations is recited at a high-level of generality with no detail of the operations performance process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
and causing a messaging module of the explanation engine to collect the top-level result as well as one or more intermediate output results from intermediate layers of the machine-based reasoning process, where the one or more intermediate output results from each layer of reasoning is associated with terminology used in at least one of i) a problem being solved and ii) a domain pertinent to the problem in order to communicate how the machine-based reasoning process came to its reported results in a communication. This limitation amounts to extra-solution activity of gathering data for use in the claimed process. As described in MPEP 2106.05(g), limitations that amount to merely adding insignificant extra-solution activity to a judicial exception do not amount to significantly more than the exception itself, and cannot integrate a judicial exception into a practical application.
Step 2B:
A non-transitory computer-readable medium including executable instructions that, when executed with one or more processors, cause an explanation engine to perform operations as follows, comprising: This limitation is recited at a high level of generality and amounts to no more than adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. The performing of operations is recited at a high-level of generality with no detail of the operations performance process such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
and causing a messaging module of the explanation engine to collect the top-level result as well as one or more intermediate output results from intermediate layers of the machine-based reasoning process, where the one or more intermediate output results from each layer of reasoning is associated with terminology used in at least one of i) a problem being solved and ii) a domain pertinent to the problem in order to communicate how the machine-based reasoning process came to its reported results in a communication. As discussed above, the additional elements of collecting results which is recited at a high level of generality and amounts to extra-solution activity of gathering data. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), "receiving or transmitting data over a network", "electronic record keeping", and "storing and retrieving information in memory").
Regarding Claim 11: The claim is rejected on the same grounds as Claim 1 for reciting substantially similar limitations.
Regarding Claim 12: The claim is rejected on the same grounds as Claim 2 for reciting substantially similar limitations.
Regarding Claim 13: The claim is rejected on the same grounds as Claim 3 for reciting substantially similar limitations.
Regarding Claim 14: The claim is rejected on the same grounds as Claim 4 for reciting substantially similar limitations.
Regarding Claim 15: The claim is rejected on the same grounds as Claim 5 for reciting substantially similar limitations.
Regarding Claim 16: The claim is rejected on the same grounds as Claim 6 for reciting substantially similar limitations.
Regarding Claim 17:
Step 2A, Prong 1:
The method of claim 11, further comprising: configuring an ablation module of the explanation engine to remove each intermediate layer of the machine-based reasoning process, one at a time, and evaluate an impact on the top-level result from the machine based reasoning process. As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses evaluating the impact of layers on the top level result to provide an explanation which is making an evaluation based upon manipulating the inputs of the machine-based reasoning process which can be feasibly performed in the human mind (see MPEP 2106.04(a)(2)(III)).
Regarding Claim 18: The claim is rejected on the same grounds as Claim 7 for reciting substantially similar limitations.
Regarding Claim 19: The claim is rejected on the same grounds as Claim 8 for reciting substantially similar limitations.
Regarding Claim 20: The claim is rejected on the same grounds as Claim 9 for reciting substantially similar limitations.
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 ofthe statutory basis (i.e., changing from AIA to pre-AIA ) forthe rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supportingthe rejection, would be the same under either status. The following is a quotation of 35 U.S.C. 103which forms the basis forall obviousness rejections set forth in this Office action: Apatent for a claimed invention may not be obtained, notwithstandingthatthe claimed invention is not identically disclosed as set forth in section 102, ifthe 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. Patentabilityshall not be negated bythe manner in which the invention was made.
Claims 1-3, 5-13, 15-16, 18-20 are rejected under 35 U.S.C. 103 as being unpatentable over Forsyth et al. (US 20200311798 A1) in view of Chatterjee et al. (US 10824959 B1), further in view of Battach et al. (US 20190266236 A1), hereinafter referred to as Forsyth, Chatterjee, and Battach, respectively.
Regarding Claim 1:
Forsyth teaches:
An apparatus, comprising: an explanation engine having a set of modules cooperating with each other configured to evaluate layers in a hierarchical architecture of a machine-based reasoning process... ([0046] "FIG. 4A is a flowchart 400 to illustrate a neural network regression flow use of visual semantic embeddings that employs a two step training procedure, according to an embodiment. As illustrated, the NN regressor model may be trained with use of a multi-layer neural network. As pre-processing steps, the search engine server 120 receives an input image 410 on which to train, which is submitted to the visual semantic embedder 124 (FIG. 1) in order to generate a visual semantic embedding for the input image 410. The search engine server 120 also receives, as an input, a group of words that represent labels describing the product represented by the input image 410 using the terms in the developed lexicon, which was discussed above. In one embodiment, the NN regressor model computes the visual semantic embedding based on a visual-semantic loss between a text-based image embedding of the input image and features represented within the group of words, as will be discussed in more detail with reference to FIG. 4B."
Examiner's Note: The search engine server is read as the explanation engine. The multi-layer neural network is read as hierarchical architecture, especially given FIG. 4A which displays the variety of modules/components organized in order and function. Computing the visual semantic embedding based on visual-semantic loss is read as evaluating layers... in a machine-based reasoning process.
where multiple layers of reasoning are associated with terminology used in at least one of i) a problem to be solved and ii) a domain pertinent to the problem in order to communicate how the machine-based reasoning process came to its reported results in a communication. ([0038] "To train the regressor, a dataset of more than 75,000 fashion products was captured from Net-A-Porter, a popular online fashion retailer, mining the product images and accompanying text descriptions foreach item. Through an iterative open encoding of frequently occurring unigrams, bigrams, and trigrams in the text descriptions, the search engine server 120 created a lexicon of 1,300 fashion characteristics broken down into eight categories: type, shape/silhouette, color, pattern/print, material, details/trim, brand, and style/context. Using this dataset, the search engine server 120 trains a two-layer neural network using a regression loss function over all 1,300 characteristics. Then, the search engine server 120 leverages these pre trained layers to train one additional neuron per characteristic, allowing the disclosed model to capture fashion characteristics with only few representative examples in the training set."
Examiner's Note: The search engine is read as the explanation engine, and the visual semantic embedder serves as the terminology module within its respective engine. The terminology in the art is all the collected lexicon of fashion characteristics that come from the fashion domain, such as the 75,000 fashion products from fashion retailer website Net-A-Porter, which is pertinent to the field and problem to be solved. These terminology assignments, AKA the trainings of one neuron per characteristic, are done through a multi-layer neural network to produce explanations.)
wherein the set of modules, includinq the messaqinq module, are implemented in i) software stored in one or more non-transitory machine-readable mediums confiqured to be executed by one or more processors; ii) loqic qates confiqured to receive an input, perform a desired functionality, and output results; iii) electronic circuits; and iv) any combination of these ([0166] computer system 1800 capable of executing disclosed embodiments includes instructions 1802 executed to cause the computer system to perform the methods/functions disclosed. [0168]-[0169] computer system includes memory and processors, etc. [0171] "The computer system 1800 may also include a disk (or optical) drive unit 1815. The disk drive unit 1815 may include a non-transitory computer readable medium 1840 in which one or more sets of instructions 1802, e.g., software, can be embedded.")
Forsyth fails to teach: as well as one or more intermediate output results from intermediate layers of the machine-based reasoning process…
However, Chatterjee teaches:
that uses machine learning to support an explanation of how the machine-based reasoning process arrived at its reported results of both a top-level result as well as corresponding intermediate output results, (Col 5 Lines 47-67 & Col 6 Lines 1-4 "Some classification techniques, such as those based on neural networks, may generate internal representations of the input data or intermediate data structures which are neither part of the input data, nor part of the model output. The predictions made by some such models may sometimes depend more directly on these internal representations (which may sometimes be referred to as "hidden" layers of the model) than on the raw input data values themselves. In such an embodiment, when a client requests an explanation for a particular prediction, a first level explanation may sometimes be provided in terms of the intermediate representations."
Examiner's Note: The particular prediction is read as the top-level result. The intermediate representations are read as the intermediate output results.)
and a messaging module of the explanation engine configured to collect the top-level result as well as one or more intermediate output results from intermediate layers of the machine-based reasoning process... (Col 5 Lines 47-67 & Col 6 Lines 1-4"Some classification techniques, such as those based on neural networks, may generate internal representations of the input data or intermediate data structures which are neither part of the input data, nor part of the model output. The predictions made by some such models may sometimes depend more directly on these internal representations (which may sometimes be referred to as "hidden" layers of the model) than on the raw input data values themselves. In such an embodiment, when a client requests an explanation for a particular prediction, a first level explanation may sometimes be provided in terms of the intermediate representations."
Examiner's Note: The particular prediction is read as the top-level result. The intermediate representations are read as the intermediate output results. According to Fig. 1, the Explainer selector 160 collects these results and is read as the messaging module.)
Forsyth and Chatterjee are considered to be analogous to each other as they are all in the field of machine learning. Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the explanation engine taught by Forsyth and the messaging module that collects top-level and intermediate results taught by Chatterjee in order to provide easier-to-understand explanations expressed through the rule-mining techniques that indicate relationships between predicates and the results. (Col 15 Lines 53-63 "In response to client requests for explanations, the first-level explainer may provide explanatory rules which indicate relationships between predicates on properties of the internal representations and the classifier's predictions. In the case of neural network classifiers, for example, the predicates of the first-level explanatory rules may be expressed in terms of the weights assigned to connections to or from one or more hidden-layer nodes, the number of input links leading into a given hidden-layer node, the number of output links emanating from a given hidden-layer node, etc.")
Forsyth and Chatterjee fail to teach: the one or more intermediate output results are associated with terminology.
However, Battach teaches:
the one or more intermediate output results are associated with terminology ([0023], Fig. 2, executing NLP model to generate intermediate result 125 based on processing input data using first selection of layers and lexicon A; [0027], NPL model 200 including lexicon A processing layers 210-1 and lexicon B processing layers 210-2; NLP model including classifiers for each vocabulary the NLP model is arranged to operate on; NLP model includes lexicon A classifier 220-1 and lexicon B classifier 220-2; [0031], Fig. 3, generating intermediate result based on processing input data with initial layers of NLP inference model, where intermediate result generated by processing input data through lexicon A processing layers 210-1 of NLP model 200 and classifier 220-1 of the NLP model).
Where the features of Battach, in which intermediate output results are collected from intermediate layers of the model and are associated with a vocabulary/lexicon, are incorporated into the system of Forsyth, in which the multiple layers of reasoning are associated with terminology used in at least one of i) a problem to be solved and ii) a domain pertinent to the problem in order to communicate how the machine-based reasoning process came to its reported results in a communication (as previously cited), these intermediate output results from the multiple layers of reasoning would be associated with terminology used in at least one of i) a problem to be solved and ii) a domain pertinent to the problem.
Forsyth and Battach are considered to be analogous to each other as they are all in the field of machine learning. Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the explanation engine taught by Forsyth and the natural language processing models which output intermediate results associated with terminology taught by Battach in order to provide for early exit in natural language processing tasks, saving computation cycles. (Battach abstract)
Regarding Claim 2:
Forsyth further teaches:
The apparatus of claim 1, where the explanation engine has a terminology module configured to assign terminology from any of i) the domain pertinent to the problem for the multiple layers in the hierarchical architecture of the machine-based reasoning process supplied from a reasoning engine... ([0038] "To train the regressor, a dataset of more than 75,000 fashion products was captured from Net-A-Porter, a popular online fashion retailer, mining the product images and accompanying text descriptions for each item. Through an iterative open encoding of frequently occurring unigrams, bigrams, and trigrams in the text descriptions, the search engine server 120 created a lexicon of 1,300fashion characteristics broken down into eight categories: type, shape/silhouette, color, pattern/print, material, details/trim, brand, and style/context. Using this dataset, the search engine server 120 trains a two-layer neural network using a regression loss function over all 1,300 characteristics. Then, the search engine server 120 leverages these pre-trained layers to train one additional neuron per characteristic, allowing the disclosed model to capture fashion characteristics with only few representative examples in the training set."
Examiner's Note: The search engine is read as the explanation engine, and the visual semantic embedder serves as the terminology module within its respective engine. The terminology in the art is all the collected lexicon of fashion characteristics that come from the fashion domain, such as the 75,000 fashion products from fashion retailer website Net-A-Porter, which is pertinent to the field and problem to be solved. These terminology assignments, AKA the trainings of one neuron per characteristic, are done through a multi-layer neural network to produce explanations.)
and ii) the specific problem to be solved, ([0031] "Given the abundance of annotated fashion image data that may be found online, previous research has leveraged deep learning techniques to build models that support multi-modal input queries. However, because these models are generally trained on text corpora that are too sparse to capture the complex and evolving semantic relationships that exist in fashion, far less attention has been paid to multi modal output: generating coherent linguistic justifications for a model's recommendations."
Examiner's Note: The general goal of the invention is to satisfy the demand of clients for fashion recommendations in the form of easily understandable explanations. The specific problem to be solved is for the model to capture the complex semantic relationships within fashion within its fashion recommendations which has not been addressed in previous research.)
where the user is able to understand the results in terms of the specific problem or domain based on the way the communication is generated. ([0032] "FIG. 2 illustrates an example of context-based product queries with stylist explanations, powered by a regression model, according to an embodiment." [0039] "The disclosed method of fashion characteristic extraction is based on a structured lexicon, allowing the search engine server 120to provide explanations by leveraging different categories of characteristics.")
Regarding Claim 3:
Forsyth further teaches:
The apparatus of claim 1, where the explanation engine has a terminology module of the explanation engine configured to accept input of terminology for the problem to be solved that is supplied by at least one of i) a description of the problem to be solved ii) a description of preferred approach to solve the problem from a user, and iii) a database of known terminology specific to the domain pertinent to the problem, ([0038] "To train the regressor, a dataset of more than 75,000 fashion products was captured from Net-A Porter, a popular online fashion retailer, mining the product images and accompanying text descriptions for each item. Through an iterative open encoding of frequently occurring unigrams, bigrams, and trigrams in the text descriptions, the search engine server 120 created a lexicon of 1,300 fashion characteristics broken down into eight categories: type, shape/silhouette, color, pattern/print, material, details/trim, brand, and style/context. Using this dataset, the search engine server 120 trains a two-layer neural network using a regression loss function over all 1,300 characteristics. Then, the search engine server 120 leverages these pre-trained layers to train one additional neuron per characteristic, allowing the disclosed model to capture fashion characteristics with only few representative examples in the training set. This architecture also makes the model extensible: as tastes change and fashion evolves, new characteristics can be added without having to retrain the entire network."
Examiner's Note: To remain consistent, the search engine (server) is again read as the explanation engine. [0038] satisfies the third requirement; the invention mines and accepts the terminology scraped from the database of Net-A-Porter online retailer which includes more than 75,000 fashion products. Given that the problem is related to fashion recommendations and the online retailer sells fashion-related, they both share the same domain.
and where the terminology module is configured to crawl through the hierarchical architecture of the machine-based reasoning process, to be created by a reasoning engine, and then associate i)the terminology specific to the problem to be solved supplied by the user and/or terminology specific to a relevant subject matter domain with ii) the multiple layers making up the hierarchical architecture of the machine-based reasoning process. ([0046] "FIG. 4A is a flowchart400 to illustrate a neural network regression flow use of visual semantic embeddings that employs a two-step training procedure, according to an embodiment. As illustrated, the NN regressor model may be trained with use of a multi-layer neural network. As pre-processing steps, the search engine server 120 receives an input image 410 on which to train, which is submitted to the visual semantic embedder 124 (FIG. 1) in order to generate a visual semantic embedding for the input image 410. The search engine server 120 also receives, as an input, a group of words that represent labels describing the product represented by the input image 410 using the terms in the developed lexicon, which was discussed above. In one embodiment, the NN regressor model computes the visual semantic embedding based on a visual-semantic loss between a text-based image embedding of the input image and features represented within the group of words, as will be discussed in more detail with reference to FIG. 4B."
[0119] "In various embodiments and with additional reference to FIG. 4B, the search engine server 120 begins by encoding each image in a general embedding space, which it uses to measure item similarity. The search engine server 120 (e.g., the visual semantic embedder 124) may then train a general embedding using a visual-semantic loss between the image embedding and features representing a text description of the corresponding item. This helps ensure that semantically similar items are projected in a nearby space."
Examiner's Note: The use of a neural regressor(neural network regressor model) is read as crawling back through the layers of the NN model. The visual semantic embedder is read as the terminology module. [0046-0049] goes on to describe how some of the NN regressor layers, specifically those within the vector predictor, are trained (AKA associated) to carry out the previously described function. Therefore, these specific layers are "associated" with the terminology.
Regarding Claim 5:
Forsyth further teaches:
The apparatus of claim 1, where the explanation engine has a crawl back module configured to cooperate with an ablation module to trace through the intermediate layers of the machine-based reasoning process constructed by a reasoning engine to record factors being considered and how important that factor was into arriving at the top-level result from the machine-based reasoning process. ([0095] "Using the neural regressor, the search engine server 120 may compute activations for all 1300 neurons for product images. Each activation may represent an "association score" of that term and the item. Users then query the NN regressor model for style and context with arbitrary combinations of these 1300 characteristics. Those queries may be submitted in a variety of different query types."
[0096] "Text-based queries: To compute how strongly a term, (i), is associated with a product, the search engine server 120 computes the visual semantic embedding of the product and then computes the "association score" for that term using the activation of the i.sup.th neuron. To retrieve a set of (k) items that best match a term of the query, the search engine server 120 samples a subset of fashion items, and returns the top k items with the largest "association score" for the term. If the query contains a set of terms, the search engine server 120 computes the combined association score as the product of the association scores of individual terms, so products with a very low association scores for any term in the set get penalized heavily." [0098] "To show and explain results for a text query with style or outfit level terms, the search engine server 120 first computes the top k most "associated" items for the query using the procedure described above. The search engine server 120 may then compute the average activation score for each term in the lexicon for these k products. The top element-level terms with the highest average activation scores serve as explanations for our results, e.g., may be formatted into phrases and sentences that explain the relevance of the image-based search results supplied in response to the query. Such explanations also allow us to better understand outfit or even brand-level styles."
Examiner's Note: The neural regressor (neural network regressor model) is read as the crawl back module crawling back through the layers of the NN model since they both collects the data from the layers. The terms and every term's activation score are read as the factors and how important that factor was to arrive at the end result using reasoning, respectively. The items are ranked according each item's combined association scores to individual terms. Then, the average activation score for each term for the ranked items are computed, the terms with the highest activation becoming the result's reasoning and most important factors.
[0151] "In Table 7, we provide an ablation study to supplement the results of Table 6."
Examiner's Note: Forsyth in [0151-0152] includes an ablation (oran ablation module)that the previously described process operates alongside.
Regarding Claim 6:
Forsyth further teaches:
The apparatus of claim 1, where the explanation engine has a crawl back module configured to cooperate with the messaging module, where the crawl back module of the explanation engine is configured to crawl through a decomposition of the machine-based reasoning process to collect.. ([0095]"Using the neural regressor, the search engine server 120 may compute activations for all 1300 neurons for product images. Each activation may represent an "association score" of that term and the item. Users then query the NN regressor model for style and context with arbitrary combinations of these 1300 characteristics. Those queries may be submitted in a variety of different query types."
Examiner's Note: The neural regressor (neural network regressor model) is read as the crawl back module crawling back through the layers of the NN model since they both collects the data from the layers.)
Forsyth fails to teach: and then report the intermediate output results from the multiple layers of the reasoning process to explain the top-level result in terms of the intermediate output results.
However, Chatterjee teaches:
and then report the intermediate output results from the multiple layers of the reasoning process to explain the top-level result in terms of the intermediate output results. (Col 5 Lines 47-67 & Col 6 Lines 1-4 "Some classification techniques, such as those based on neural networks, may generate internal representations of the input data or intermediate data structures which are neither part of the input data, nor part of the model output. The predictions made by some such models may sometimes depend more directly on these internal representations (which may sometimes be referred to as "hidden" layers of the model) than on the raw input data values themselves. In such an embodiment, when a client requests an explanation for a particular prediction, a first level explanation may sometimes be provided in terms of the intermediate representations."
Examiner's Note: The particular prediction is read as the top-level result. The intermediate representations are read as the intermediate output results. According to Fig. 1, the Explainer selector 160 collects these results and is read as the messaging module.)
Forsyth and Chatterjee are considered to be analogous to each other as they are all in the field of machine learning. Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the explanation engine taught by Forsyth and reporting the collected top-level and intermediate results taught by Chatterjee in order to provide easier-to-understand explanations expressed through the rule mining techniques that indicate relationships between predicates and the results. (Col 15 Lines 53 63 "In response to client requests for explanations, the first-level explainer may provide explanatory rules which indicate relationships between predicates on properties of the internal representations and the classifier's predictions. In the case of neural network classifiers, for example, the predicates of the first-level explanatory rules may be expressed in terms of the weights assigned to connections to or from one or more hidden-layer nodes, the number of input links leading into a given hidden-layer node, the number of output links emanating from a given hidden-layer node, etc.")
Regarding Claim 7:
Forsyth further teaches:
The apparatus of claim 1, where the explanation engine has an ablation module configured to change the intermediate output results from layers of the machine-based reasoning process by altering an input for that layer and then output a new intermediate output result from that layer of the machine-based reasoning process as well as a new top-level result. ([0151] "More specifically, Table 7 includes.. additional ablations including components illustrated in FIG. 4B. In Table 7, we provide an ablation study to supplement the results of Table 6. In further embodiments, additional layers are employed including the FC, which uses a fully connected layer for its type-specific projection rather than a learned diagonal projection, and Cosine, which uses cosine distance to train the NN regressor model rather than Euclidean distance."
[0152] "The first line of Table 7(b) illustrates that learning our type specific embeddings gives a consistent improvement over training a single embedding to make comparisons. We note that our relative performance using the entire dataset is higher than our disjoint set, which we attribute to likely being due to the additional training data for learning each type-specific embedding. Analogous to the Maryland dataset, the next three lines of Table 7(b) illustrate a consistent performance improvement as we add in the remaining pieces of our model."
Examiner's Note: The ablation study tests the different combinations of layers, embeddings, and datasets used. The varying usage of different database sets is read as altering the input data. The varying integrations of layers and embeddings is read as changing the intermediate output from the layers. Furthermore, the table demonstrates that different accuracies result from the removing and adding of layers and datasets; this is read as outputting a new intermediate output result from the layer of the machine-based reasoning process.
Regarding Claim 8:
Forsyth fails to teach:
The apparatus of claim 1, further comprising: an ablation module configured to conduct one or more ablation cycles to alter an input to a layer of the machine-based reasoning process created by a reasoning engine to determine an effect of that layer on the top level result and record the effect; and where the messaging module is configured to take results of the ablation cycles and data generated with them in order to generate the reported results of an impact of each layer of machine-based reasoning process in the communication generated by the messaging module.
However, Chatterjee teaches:
The apparatus of claim 1, further comprising: an ablation module configured to conduct one or more ablation cycles to alter an input to a layer of the machine-based reasoning process created by a reasoning engine to determine an effect of that layer on the top-level result and record the effect; (Col 14 Lines 9-19 "Using the contents of table 521, a selected rule mining algorithm has obtained a rule set525 comprising three rules RO, R1 and R2. Each rule indicates some set of predicates on the input attributes, and an implication about what the classifier would predict regarding the favorite sport if the attribute predicate conditions are met by a given observation record. Two example criteria for ranking the rules are shown: support or coverage (indicative of the fraction of the input data which meets the attribute predicate conditions of the different rules) and confidence or accuracy (indicative of the correctness of the implication)."
Col 15 Lines 43-56 "In an initial stage 720A of a potentially iterative or hierarchical rule-mining technique..."
Examiner's Note: The changing predicates on the input attributes is read as the altered input into layers of the machine-based reasoning process. The process is potentially iterative. The confidence measure given for each attribute predicate shows how much each attribute predicate's presence affects prediction, AKA top-level result.)
and where the messaging module is configured to take results of the ablation cycles and data generated with them in order to generate the reported results of an impact of each layer of machine-based reasoning process in the communication generated by the messaging module. (Col 3 Lines 26-33 "For example, the machine learning service may train, test and evaluate a wide variety of models (e.g., for supervised and/or unsupervised learning) in response to client requests received via a set of programmatic interfaces of the service (e.g., application programmatic interfaces or APIs, web-based consoles, command-line tools, or graphical user interfaces."
Col 15 Lines 43-56 "In an initial stage 720A of a potentially iterative or hierarchical rule mining technique, an explainer with a rule set 722 formulated in terms of predicates on the properties of the internal representations may be generated. In some cases, the initial rule set may include predicates on the raw input data as well as predicates on the internal representations; that is, the initial rule set may not be restricted to predicates on the properties of the internal representations alone. The initial set of rules may be ranked relative to one another, e.g., using similar types of metrics to those discussed earlier. In response to client requests for explanations, the first-level explainer may provide explanatory rules which indicate relationships between predicates on properties of the internal representations and the classifier's predictions."
Examiner's Note: The explainer provides explanatory rules that indicate the relationships between predicates and the predictions. Those results are then provided to the client according to their requests through communication apparatuses.)
Forsyth and Chatterjee are considered to be analogous to each other as they are all in the field of machine learning. Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the explanation engine taught by Forsyth and the ablation module that generates the results of each layer's impact on the reasoning process and the messaging module that reports the results of the ablation module taught by Chatterjee in order to provide easier-to-understand explanations expressed through the rule-mining techniques that indicate relationships between predicates and the prediction results. Here, the rules with predicates are read as the layers. (Col 15 Lines 53-63 "In response to client requests for explanations, the first-level explainer may provide explanatory rules which indicate relationships between predicates on properties of the internal representations and the classifier's predictions. In the case of neural network classifiers, for example, the predicates of the first-level explanatory rules may be expressed in terms of the weights assigned to connections to or from one or more hidden-layer nodes, the number of input links leading into a given hidden-layer node, the number of output links emanating from a given hidden-layer node, etc.")
Regarding Claim 9:
Forsyth teaches:
The apparatus of claim 1, further comprising: where the messaging module of the explanation engine is configured to 1) extract the intermediate output results from the multiple layers of the machine-based reasoning process created by a reasoning engine ([0047] "The multi-layer neural network may further include a first fully connected layer (FC) and rectifier linear unit (ReLU) 420 and a second FC and ReLU 430, which feeds two separate processing layers, namely the complete vector predictor 132 and the individual term predictor 134, which may be executed in parallel in one embodiment. More specifically, the second FC and ReLU 430 (or an additional and final sets of full connected NN layer and rectifier linear unit layers) may output intermediate NN vector outputs that become inputs into the complete vector predictor 132 and the individual term predictor 134."
Examiner's Note: To remain consistent, the search engine is again read as the explanation engine. The intermediate NN vector outputs are extracted from the layers of the NN, AKA the FCs and ReLUs. These extractions become inputs for the module in the next part of the reasoning process.)
and 2) cooperate with a terminology module to associate the intermediate output results from the multiple layers with the terminology taken from the at least one of i) subject domain pertinent to the problem and ii)the problem specific terminology used in the problem to be solved. ([0051] "In the second step, the search engine server 120 may stop training with the two hidden layers, e.g., the first and second FC and ReLU layers 420 and 430, and connect the last layer (e.g., the second FC and ReLU layer430) to 1,300 individual neurons with separate sigmoid activations, where each neuron predicts a specific characteristic term by using the individual term predictor 134... The individual term predictor 134, in order words, includes numerous neurons (e.g., 1300 in this example), each with a corresponding sigmoid activation, where each neuron of the numerous neurons is separately trainable for a respective individual term (e.g., characteristic term) corresponding to the input image."
Examiner's Note: The individual term predictor is read as the terminology module since it trains or associates the intermediate output results from the last connected FC and ReLU layers to each neuron with a specific characteristic term. These terms are extracted from fashion websites which are from domains pertinent to the problem being solved; the process of obtaining these terms is outlined in [0042-0044].
Regarding Claim 10:
The claim is rejected on the same grounds as Claim 1 for reciting substantially similar limitations, with the exception of one limitation further taught by Forsyth:
A non-transitory computer-readable medium including executable instructions that, when executed with one or more processors, cause an explanation engine to perform operations as follows, comprising: ([0171] "The computer system 1800 may also include a disk (or optical) drive unit 1815. The disk drive unit 1815 may include a non-transitory computer readable medium 1840 in which one or more sets of instructions 1802, e.g., software, can be embedded.")
Regarding Claim 11: The claim is rejected on the same grounds as Claim 1 for reciting substantially similar limitations.
Regarding Claim 12: The claim is rejected on the same grounds as Claim 2 for reciting substantially similar limitations.
Regarding Claim 13: The claim is rejected on the same grounds as Claim 3 for reciting substantially similar limitations.
Regarding Claim 15: The claim is rejected on the same grounds as Claim 5 for reciting substantially similar limitations.
Regarding Claim 16: The claim is rejected on the same grounds as Claim 6 for reciting substantially similar limitations.
Regarding Claim 18: The claim is rejected on the same grounds as Claim 7 for reciting substantially similar limitations.
Regarding Claim 19: The claim is rejected on the same grounds as Claim 8 for reciting substantially similar limitations.
Regarding Claim 20: The claim is rejected on the same grounds as Claim 9 for reciting substantially similar limitations.
Claims 4, 14 are rejected under35 U.S.C. 103 as being unpatentable over Forsyth in view of Chatterjee, further in view of Battach, in further view of Donaldson et al. (US 20220067557A1), hereinafter referred to as Donaldson.
Regarding Claim 4:
Forsyth further teaches:
The apparatus of claim 1, where the explanation engine is configured to cooperate with a first reasoning engine that is configured to breakdown its machine-based reasoning process into divisible layers that provide intermediary output results to other layers in order to determine the top level result from the machine-based reasoning process; ([0036] "The NN regressor 130 may further include a complete vector predictor 132, an individual term predictor 134, and a mean square loss calculator 136, which will be discussed with reference to FIG. 4A, as part of use of the visual semantic embeddings (vectors) to train the NN regressor 130 to generate fashion-based recommendations and other multi-modals outputs."
[0047] "The multi-layer neural network may further include a first fully connected layer (FC) and rectifier linear unit (ReLU) 420 and a second FC and ReLU 430, which feeds two separate processing layers, namely the complete vector predictor 132and the individual term predictor 134, which may be executed in parallel in one embodiment. More specifically, the second FC and ReLU 430 (or an additional and final sets of full connected NN layer and rectifier linear unit layers) may output intermediate NN vector outputs that become inputs into the complete vector predictor 132 and the individual term predictor 134."
Examiner's Note: To remain consistent, the search engine is again read as the explanation engine. The multi-layer neural network feeds into two separate processing layers which is read as breaking down its machine-based reasoning process into divisible layers. The two processing layers are used to make fashion-based recommendations which is read as determining the top-level result from the machine-based reasoning process.
Forsyth fails to teach: as opposed to a second reasoning engine that is configured to create one omnibus neural network that is compiled as a black box that merely outputs its final decision;
However, Donaldson teaches:
as opposed to a second reasoning engine that is configured to create one omnibus neural network that is compiled as a black box that merely outputs its final decision; ([0013] "A final layer combines the lowest resolution output to make a prediction about the class of the image being considered. The family of Class Activation Mappings (CAM) examine each spatial resolution-particularly the penultimate low-resolution layer-to highlight the areas of the image important to the CNN's classification decision. 'Black box' methods, such as Randomized Input Sampling for Explanation (RISE), do not require access or knowledge of internal machine learning function processes to highlight influential image regions, though their results are not presented in context of example reference images.")
Forsyth and Donaldson are considered to be analogous to each other as they are all in the field of machine learning. Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the explanation engine taught by Forsyth and the supposed implementation of a "black box" neural network taught by Donaldson in order to not require access or knowledge of the internal machine learning function processes to provide explanations. ([0013] "Blackbox' methods, such as Randomized Input Sampling for Explanation (RISE), do not require access or knowledge of internal machine learning function processes to highlight influential image regions, though their results are not presented in context of example reference images.")
Forsyth and Donaldson fail to teach: and where the explanation engine is configured to cooperate with the first reasoning engine to allow a user to query what the intermediary output results are for each layer of the machine-based reasoning process as well as what would happen when the intermediary output results were altered.
However, Chatterjee teaches:
and where the explanation engine is configured to cooperate with the first reasoning engine to allow a user to query what the intermediary output results are for each layer of the machine-based reasoning process as well as what would happen when the intermediary output results were altered. ([0067] "The respective in memory database instances may receive the corresponding query execution instructions from the query coordinator. The respective in-memory database instances may execute the corresponding query execution instructions to obtain, process, or both, data (intermediate results data) from the low-latency data. The respective in-memory database instances may output, or otherwise make available, the intermediate results data, such as to the query coordinator."
[0068] "The query coordinator may execute a respective portion of query execution instructions (allocated to the query coordinator) to obtain, process, or both, data (intermediate results data) from the low-latency data. The query coordinator may receive, or otherwise access, the intermediate results data from the respective in-memory database instances. The query coordinator may combine, aggregate, or otherwise process, the intermediate results data to obtain results data.)"
Forsyth and Chatterjee are considered to be analogous to each other as they are all in the field of machine learning. Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the explanation engine taught by Forsyth and the function to allow a user to query the intermediary output results of a reasoning process in order to provide to the user easier-to-understand explanations expressed through the rule-mining techniques that indicate relationships between predicates and the prediction results. Here, the rules with predicates are read as the layers, and the effect of the rules on the prediction outcome is presented to the user. (Col 15 Lines 53-63 "In response to client requests for explanations, the first-level explainer may provide explanatory rules which indicate relationships between predicates on properties of the internal representations and the classifier's predictions. In the case of neural network classifiers, for example, the predicates of the first-level explanatory rules may be expressed in terms of the weights assigned to connections to or from one or more hidden-layer nodes, the number of input links leading into a given hidden-layer node, the number of output links emanating from a given hidden-layer node, etc.")
Regarding Claim 14: The claim is rejected on the same grounds as Claim 4 for reciting substantially similar limitations.
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Forsyth in view of Chatterjee, further in view of Battach, in further view of Takeuchi et al. (US 20190221311 A1), hereinafter referred to as Takeuchi.
Regarding Claim 17:
Forsyth fails to teach: The method of claim 11, further comprising: configuring an ablation module of the explanation engine to remove each intermediate layer of the machine-based reasoning process, one at a time, and evaluate an impact on the top-level result from the machine based reasoning process.
However, Takeuchi teaches:
The method of claim 11, further comprising: configuring an ablation module of the explanation engine to remove each intermediate layer of the machine based reasoning process, one at a time, and evaluate an impact on the top-level result from the machine-based reasoning process. ([0095] "FIG. 5 is a block diagram for illustrating an example of functional components of the neural network NN. The neural network NN includes the input layer 401, the intermediate layer 402, an output layer403, a conversion module 501, a rearrangement module 502, a prediction data calculation module 503, an importance calculation module 504, a setting module 505, a unification module 506, a dimensionality reduction module 507, and a selection module 508."
[0103] "The dimensionality reduction module 507 reduces the number d of dimensions of the output vector h.sup.l.sub.D based on the output vector h.sup.l.sub.D from the intermediate layer 402 and the matrixW.sup.l.sub.R as expressed by Expression (2) to output the output vector h.sup.l.sub.R subjected to the dimensionality reduction. The dimensionality reduction module 507 corresponds to the above-mentioned reporting unit group RU. In this case, the setting module 505 sets the weight a for the intermediate layer 402 based on the output vector h.sup.l.sub.R subjected to the dimensionality reduction from the dimensionality reduction module 507 and the matrix W.sub.A."
[0191] "Further, the analysis apparatus 320 includes the dimensionality reduction module 507, to thereby allow data analysis to become more efficient through dimensionality reduction.
[0135] "The predictive function 910 executes the predictive processing based on the neural network NN (Step S916). Specifically, for example, the predictive function 910 selects the corresponding neural network NN from among the neural network group NNs based on the model parameter MP. Then, the predictive function 910 calculates the prediction result 353 and the importance of the item by supplying the selected neural network NN with the feature vectorx.sub.n and the explainable vector and outputs the prediction result 353 and the importance of the item to the explanation function 920 (Step S917)."
Examiner's Note: The dimensionality reduction module serves as the ablation module and removes each intermediate layer from the output vector. The output vector is then passed on to the analysis apparatus for the predictive function to calculate the prediction result and importance of each item on the final result. The importance of each item is read as the impact of each input of each intermediate layer on the top-level result. Furthermore, [0052] of the Specification states that removing a layer from the reasoning flow by altering an input can be done as a "zero weight". [0048] of Takeuchi also states the weight of the feature vectors range from 0 to 1.)
Forsyth and Takeuchi are considered to be analogous to each other as they are all in the field of machine learning. Therefore, before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to combine the explanation engine taught by Forsyth and the function to remove intermediate layers of a reasoning process and observe the impact on the top-level result in order to provide to the user easier-to-understand explanations that displays the contribution of a feature to the prediction. ([0018] "This invention has been made in view of the above-mentioned points, and therefore has an object to achieve improvement in interpretability of relevance between a feature that has contributed to a prediction based on machine learning and a clinical pathway." [0041] "A normal neural network outputs only the prediction result, but the neural network NN in this embodiment outputs not only the prediction result but also importance of a feature item (hereinafter referred to simply as "item").")
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
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 JEREMY L STANLEY whose telephone number is (469)295-9105. The examiner can normally be reached on Monday-Friday from 9:00 AM to 5:00 PM CST.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Al Kawsar, can be reached at telephone number (571) 270-3169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center and Private PAIR for authorized users only. Should you have questions about access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
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) Form at https://www.uspto.gov/patents/uspto-automated- interview-request-air-form.
/JEREMY L STANLEY/
Primary Examiner, Art Unit 2127
Prosecution Insights