DETAILED ACTION
Status of Claims
This Office action is responsive to communications filed on 2025-10-22. Claim(s) 2 was/were cancelled, and claim(s) 10-12 was/were added. Claim(s) 1 and 3-12 is/are pending and are examined herein.
Claim(s) 1 and 3-12 is/are rejected under 35 USC 101.
Claim(s) 1 and 3-12 is/are rejected under 35 USC 103.
Notice of Pre-AIA or AIA Status
The present application, filed on or after 2013-03-16, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The attached information disclosure statement(s) (IDS), submitted on 2025-10-17, is/are in compliance with the provisions of 37 CFR 1.97. Accordingly, the attached information disclosure statement(s) is/are being considered by the examiner.
Response to Arguments
Regarding rejections under 35 USC 101, the applicant’s arguments have been fully considered but they are not persuasive.
The applicant asserts that the amended claims “recite a specific configuration of hardware and machine-learned components” [remarks, page 9]. The examiner respectfully disagrees with this characterization. The claims recite merely generic computing hardware (a memory and a processor) alongside software functionalities to be implemented on this generic computing hardware, where the functionalities, as recited, are feasible to perform in the human mind. MPEP 2106.04(d) indicates that “merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea” is insufficient to integrate an abstract idea into a practical application.
The applicant asserts that “it is not possible for a person to divide time-series tensor data into multiple datasets, route them to different trained neural network components, combine them via a trained coupling section, and then tune the results using error-based correction parameters -- all in real time” [remarks, page 9]. However, the applicant provides no rationale to justify why the steps which were indicated as being abstract ideas cannot be performed in the human mind. This assertion fails to comply with 37 CFR 1.111(b) because it amounts to a general allegation that the claims define a patentable invention without specifically pointing out how the language of the claims prevents them from being performed by a human being. The examiner notes that the independent claims do not at present describe dividing the time-series data, but the point is anyway moot since it is evidently possible for a human being to divide data (cf. [Office action of 2025-07-22, page 7]). Routing data to neural networks is merely a step of data transfer, which means that it is an additional element which does not integrate the abstract idea into a practical application or provide significantly more than an abstract idea. Combining and tuning (cf. examiner’s remarks) are both practical to perform in the human mind.
The applicant argues that the claims “improve the functioning of inference systems” [remarks, page 10] and that they are not well-understood, routine, or conventional [remarks, page 11], but, in each case, the features listed by the applicant in their remarks [remarks, pages 10-11] are not recited in the independent claim and are not clearly supported in the specification. There is no “industrial time-series data” or “parallel learned networks” or “learned coupling” or “error-based tuning” or “dynamic division” or “simultaneous training” or “optimiz[ing] output” or “dynamic partitioning” in the claims (or in the originally filed specification). MPEP 2105.05(a) indicates that the requirements of the improvements analysis include that “a technical explanation as to how to implement the invention should be present in the specification” and that “the claim must be evaluated to ensure the claim itself reflects the disclosed improvement in technology”. In the present instance, the specification does not describe the purported improvements and/or purportedly non-routine elements described in the applicant’s remarks, and these elements are also not reflected in the pending independent claims.
The complete 101 analysis, updated in view of the applicant’s amendments, is given below.
Regarding rejections under 35 USC 103, the applicant’s arguments have been fully considered but they are not persuasive. The applicant asserts that Agostinelli does not disclose training “in an integrated manner” as recited in the amended claim [remarks, pages 12-14] but this is not persuasive. The applicant’s specification does not describe any special meaning to be attributed to the phrase “in an integrated manner”. In other words, nothing in the originally filed specification would lead a person of ordinary skill in the art to conclude that that the broadest reasonable interpretation of the phrase “in an integrated manner” precludes the type of training performed in Agostinelli. For example, the fact that the same noisy image is fed to each of the columns is one way in which the training is “integrated” (the columns are “integrated” via their inputs), and the fact that the columns are trained in parallel is yet another (the training is “integrated” temporally). Similarly, the fact that output of the columns is used by the component which computes the weighted average is still another way in which the training of the “plurality of the network sections and the coupling section” occurs “in an integrated manner” (the training of these sections is “integrated” procedurally). The complete prior art mappings, updated in view of the applicant’s amendments, are given below.
Specification
The specification is objected to as failing to provide proper antecedent basis for the claimed subject matter. See 37 CFR 1.75(d)(1) and MPEP 608.01(o). Specifically, MPEP 608.01(o) indicates that claims should be “scrutinized not only for new matter but also for new terminology” and that the “use of a confusing variety of terms for the same thing should not be permitted”. Claims 11-12 recite divide the acquired time series data group into a plurality of data sets and to cause at least one of the plurality of data sets to be processed by a plurality of different network sections among the plurality of network sections [claim 11; emphasis added] and cause each of the at least one of the plurality of data sets to be processed by a plurality of normalization processes that use different methods [claim 12; emphasis added]. However, the specification recites instead “dividing the time series data group into a plurality of groups” [specification, 0077; emphasis added]. This constitutes a “use of a confusing variety of terms for the same thing” and the specification is therefore objected to. The examiner suggests replacing all of the indicated instances of “plurality of data sets” in the claims with “plurality of groups” so that the specification provides clear and consistent antecedent basis for claim terminology.
Appropriate correction is required.
Examiner’s Remarks
Claim 1 recites tune the respective output data. The examiner notes that this usage of the word “tune” appears to deviate from its standard usage in the art: in machine learning, the verb “tune” is typically used with its direct object being a model and it refers to optimizing the performance of that model. However, the direct object of “tune” in the limitation above is the “respective output data” of the claim, which refers to (intermediate) output of a model, not to a model itself. Consequently, the word “tune” is interpreted broadly in accordance with its plain meaning as referring to any adjustment.
Claim Rejections - 35 USC 101
35 USC 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.
Claim(s) 1 and 3-12 is/are rejected under 35 USC 101 because the claimed invention(s) is/are directed to abstract ideas without significantly more.
Claim 1
Step 1. The claim and its dependents 3-7 and 10-12 fall under the statutory category of machines. An analysis of step 2 for each of these claims follows.
Step 2A Prong 1. The claim recites the following abstract ideas:
process the acquired time series data group to output respective output data, (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “process” time series data. See MPEP 2106.04(a)(2)(III).)
combine the respective output data to output a combined result, (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “combine” data. See MPEP 2106.04(a)(2)(III).)
tune the respective output data… tune the respective output data using a correction parameter corresponding to an error included in the inference result. (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can adjust/tune data based on an analysis of certain parameters. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
An inference device comprising: a memory and a processor that is coupled to the memory and that is configured to: … function as… wherein the processor is configured to (This recites generic computing components for performing an abstract idea. See MPEP 2106.05(f)(2).)
acquire a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to… the coupling section being configured to (This merely recites using “machine-learned versions of sections” to perform mental processes identified above. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data that is obtained from a resultant object obtained by processing the target object; (This recites a generic property of training a supervised machine learning model, since training results in the model’s predictions converging to (“approaching”) expected output. In other this, this merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[the respective output data] that is output by processing the acquired time series data group using the machine-learned version of the plurality of network sections and that is not combined by the machine-learned version of the coupling section, (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
and output an inference result by combining the respective tuned output data; (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
An inference device comprising: a memory and a processor that is coupled to the memory and that is configured to: … function as… wherein the processor is configured to (This recites generic computing components for performing an abstract idea. See MPEP 2106.05(f)(2).)
acquire a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; (This insignificant extra-solution activity is well-understood, routine, conventional as it is mere data transfer. See MPEP 2106.05(d)(II), “Receiving or transmitting data over a network” and/or “Storing and retrieving information in memory”.)
a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to… the coupling section being configured to (This merely recites using “machine-learned versions of network sections” to perform mental processes identified above. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data that is obtained from a resultant object obtained by processing the target object; (This recites a generic property of training a supervised machine learning model, since training results in the model’s predictions converging to (“approaching”) expected output. In other this, this merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[the respective output data] that is output by processing the acquired time series data group using the machine-learned version of the plurality of network sections and that is not combined by the machine-learned version of the coupling section, (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
and output an inference result by combining the respective tuned output data; (The insignificant extra-solution activity is well-understood, routine, conventional as it is merely presenting output. See MPEP 2106.05(d)(II), “Presenting offers”.)
Claim 3
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[The inference device according to claim 1, wherein the processor is configured to] generate a first time series data group and a second time series data group by processing the acquired time series data group according to a first criterion and a second criterion, respectively, (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can generate data groups according to criteria. See MPEP 2106.04(a)(2)(III).)
and to process the generated first time series data group and the second time series data group (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “process” groups of data. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
using the machine-learned version of the plurality of networks sections. (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
using the machine-learned version of the plurality of networks sections. (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Claim 4
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[The inference device according to claim 1, wherein the processor is configured to] divide the acquired time series data group into groups according to a data type or a time range (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can divide data into groups. See MPEP 2106.04(a)(2)(III).)
and to process the respective divided groups (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
using the machine-learned version of the plurality of network sections. (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
using the machine-learned version of the plurality of network sections. (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Claim 5
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[the plurality of network sections, each of which includes a normalization section that] performs a normalization process using a different method. (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can normalize different groups of data in different ways. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
[The inference device according to claim 1, wherein the processor is configured to process the acquired time series data group using the machine-learned version of the plurality of network sections,] each of which includes a normalization section that (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
[The inference device according to claim 1, wherein the processor is configured to process the acquired time series data group using the machine-learned version of the plurality of network sections,] each of which includes a normalization section that (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Claim 6
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[The inference device according to claim 1, wherein the processor is configured to:] divide the acquired time series data group into a first time series data group… and a second time series data group (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
and process the first time series data group and the second time series data group (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
[a first time series data group] that is measured in accordance with the processing of the target object in a first processing space of the predetermined processing unit, [and a second time series data group] that is measured in accordance with the processing of the target object in a second processing space; (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
using the machine-learned version of the plurality of network sections. (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
[a first time series data group] that is measured in accordance with the processing of the target object in a first processing space of the predetermined processing unit, [and a second time series data group] that is measured in accordance with the processing of the target object in a second processing space; (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
using the machine-learned version of the plurality of network sections. (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Claim 7
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
[The inference device according to claim 1, wherein] the time series data group is data measured in accordance with processing in a substrate processing device. (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
[The inference device according to claim 1, wherein] the time series data group is data measured in accordance with processing in a substrate processing device. (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
Claim 10
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[The inference device according to claim 1, wherein the processor is further configured to] select a network section from among the plurality of network sections based on a processing space or a criterion corresponding to identification information included in the acquired time series data group. (This recites a mental process that can be performed in the human mind or by a human using pen and paper. A human being can make a selection based on a criterion. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
Claim 11
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[The inference device according to claim 1, wherein the processor is further configured to] divide the acquired time series data group into a plurality of data sets (This recites a mental process that can be performed in the human mind or by a human using pen and paper. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
and to cause at least one of the plurality of data sets to be processed by a plurality of different network sections among the plurality of network sections. (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
and to cause at least one of the plurality of data sets to be processed by a plurality of different network sections among the plurality of network sections. (This recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
Claim 12
Step 2A Prong 1. The claim recites the following abstract ideas:
The abstract idea(s) in the parent claim(s).
[The inference device according to claim 11, wherein the processor is further configured to] cause each of the at least one of the plurality of data sets to be processed by a plurality of normalization processes that use different methods. (This recites a mathematical concept and a mental process that can be performed in the human mind or by a human using pen and paper. A human being can perform normalization processes. See MPEP 2106.04(a)(2)(I, III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
The additional element(s) in the parent claim(s).
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
The additional element(s) in the parent claim(s).
Claim 8
Step 1. The claim falls under the statutory category of methods.
Step 2A Prong 1. The claim recites the following abstract ideas:
An inference method comprising: processing the acquired time series data group (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “process” time series data. See MPEP 2106.04(a)(2)(III).)
process the acquired time series data group to output respective output data, (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “process” time series data. See MPEP 2106.04(a)(2)(III).)
combine the respective output data to output a combined result, (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “combine” data. See MPEP 2106.04(a)(2)(III).)
tuning the respective output data… wherein the respective output data are tuned using a correction parameter corresponding to an error included in the inference result. (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can adjust/tune data based on an analysis of certain parameters. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
acquiring a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
by using a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to… the coupling section being configured to (This merely recites using “machine-learned versions of sections” to perform mental processes identified above. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data that is obtained from a resultant object obtained by processing the target object; (This recites a generic property of training a supervised machine learning model, since training results in the model’s predictions converging to (“approaching”) expected output. In other this, this merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[the respective output data] that is output by processing the acquired time series data group using the machine-learned version of the plurality of network sections and that is not combined by the machine-learned version of the coupling section, (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
and outputting an inference result by combining the respective tuned output data; (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
acquiring a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; (This insignificant extra-solution activity is well-understood, routine, conventional as it is mere data transfer. See MPEP 2106.05(d)(II), “Receiving or transmitting data over a network” and/or “Storing and retrieving information in memory”.)
by using a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to… the coupling section being configured to (This merely recites using “machine-learned versions of sections” to perform mental processes identified above. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data that is obtained from a resultant object obtained by processing the target object; (This recites a generic property of training a supervised machine learning model, since training results in the model’s predictions converging to (“approaching”) expected output. In other this, this merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[the respective output data] that is output by processing the acquired time series data group using the machine-learned version of the plurality of network sections and that is not combined by the machine-learned version of the coupling section, (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
and outputting an inference result by combining the respective tuned output data; (The insignificant extra-solution activity is well-understood, routine, conventional as it is merely presenting output. See MPEP 2106.05(d)(II), “Presenting offers”.)
Claim 9
Step 1. The claim falls under the statutory category of machines.
Step 2A Prong 1. The claim recites the following abstract ideas:
processing the acquired time series data group (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “process” time series data. See MPEP 2106.04(a)(2)(III).)
process the acquired time series data group to output respective output data, (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “process” time series data. See MPEP 2106.04(a)(2)(III).)
combine the respective output data to output a combined result, (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can “combine” data. See MPEP 2106.04(a)(2)(III).)
tuning the respective output data… wherein the respective output data are tuned using a correction parameter corresponding to an error included in the inference result. (This recites a mental process that can be performed in the human mind or by a human using pen and paper, since a human being can adjust/tune data based on an analysis of certain parameters. See MPEP 2106.04(a)(2)(III).)
Step 2A Prong 2. The claim recites the following additional elements which, considered individually and as an ordered combination, do not integrate the abstract idea into a practical application:
A non-transitory recording computer readable medium storing an inference program that causes a computer to execute: (This recites generic computing components for performing an abstract idea. See MPEP 2106.05(f)(2).)
acquiring a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
by using a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to… the coupling section being configured to (This merely recites using “machine-learned versions of network sections” to perform mental processes identified above. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data that is obtained from a resultant object obtained by processing the target object; (This recites a generic property of training a supervised machine learning model, since training results in the model’s predictions converging to (“approaching”) expected output. In other this, this merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[the respective output data] that is output by processing the acquired time series data group using the machine-learned version of the plurality of network sections and that is not combined by the machine-learned version of the coupling section, (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
and outputting an inference result by combining the respective tuned output data; (This recites insignificant extra-solution activity. See MPEP 2106.05(g).)
Step 2B. The claim recites the following additional elements which, considered individually and as an ordered combination, do not amount to significantly more than the abstract idea:
A non-transitory recording computer readable medium storing an inference program that causes a computer to execute: (This recites generic computing components for performing an abstract idea. See MPEP 2106.05(f)(2).)
acquiring a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; (This insignificant extra-solution activity is well-understood, routine, conventional as it is mere data transfer. See MPEP 2106.05(d)(II), “Receiving or transmitting data over a network” and/or “Storing and retrieving information in memory”.)
by using a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to… the coupling section being configured to (This merely recites using “machine-learned versions of network sections” to perform mental processes identified above. In other words, this recites merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data that is obtained from a resultant object obtained by processing the target object; (This recites a generic property of training a supervised machine learning model, since training results in the model’s predictions converging to (“approaching”) expected output. In other this, this merely applying (or equivalent) an abstract idea, or implementing an abstract idea on a computer, or using a computer as a tool to perform an abstract idea. See MPEP 2106.05(f).)
[the respective output data] that is output by processing the acquired time series data group using the machine-learned version of the plurality of network sections and that is not combined by the machine-learned version of the coupling section, (This recites data of a particular type or source, merely linking an abstract idea to a particular field of use. See MPEP 2106.05(h).)
and outputting an inference result by combining the respective tuned output data; (The insignificant extra-solution activity is well-understood, routine, conventional as it is merely presenting output. See MPEP 2106.05(d)(II), “Presenting offers”.)
Claim Rejections - 35 USC 103
The following is a quotation of 35 USC 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1, 3-4, 6, and 8-11 is/are rejected under 35 USC 103 as being unpatentable over Forest AGOSTINELLI et al. (Adaptive Multi-Column Deep Neural Networks with Application to Robust Image Denoising, published 2013; hereafter, “Agostinelli”) in view of E. M. MAVROVOUNIOTIS et al. (Hierarchical Neural Networks, published 1992; hereafter, “Mavrovouniotis”).
Claim 1
Agostinelli discloses:
acquiring [a time series data group] ([Agostinelli, figure 1]: In the methods disclosed by Agostinelli, the input that is acquired is image data rather than a time series data group. The combination with Mavrovouniotis below discloses the time series data.)
a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to process the acquired [time series data group] to output respective output data, the coupling section being configured to combine the respective output data to output a combined result, ([Agostinelli, section 3.2 and figure 1]: Agostinelli discloses a neural network architecture consisting of multiple columns, where each column produces an output, and where each column output is weighted using a weight s_c computed by a weight prediction module [Agostinelli, figure 1(b) and section 3.2]. The columns are the “plurality of network sections” of the claim, the column outputs are the “respective output data” of the claim, the final output (i.e., the weighted average of the column outputs) is the “combined result” of the claim, and the component which computes the weighted average is the “coupling section” of the claim.)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data ([Agostinelli, section 3.2.1]: Agostinelli discloses that the architecture therein “has three training phases: training the [columns], determining optimal weights for a [training set], and then training the weight prediction module” [Agostinelli, section 3.2.1]. This training process ensures that the architecture therein is “machine-learned in an integrated manner such that” the limitation recited by the claim is satisfied: the training procedure is designed to minimize error [Agostinelli, section 3.2.2 equation (6)], i.e., which means that the final output of the network (denoted “hat{Y}s”) converges to (“approaches”) the expected/true output (denoted “y”). The expected/true output maps to the “inspection data” of the claim.)
tune the respective output data ([Agostinelli, section 3.2 and figure 1]: As noted above, Agostinelli discloses weighting column outputs. Weighting column outputs is a form of adjusting the column outputs (i.e., the “respective output data” of the claim) and maps to the “tun[ing] the respective output data” step of the claim.) that is output by processing the acquired [time series data group] using the machine-learned version of the plurality of network sections ([Agostinelli, section 3.2 and figure 1]: The column outputs are output by the columns (i.e., “the plurality of network sections” of the claim), so the mappings given have the property that the “respective output data… is output by… using the machine-learned version of the plurality of network sections” as recited by this limitation.) and that is not combined by the machine-learned version of the coupling section, ([Agostinelli, section 3.2 and figure 1]: When the column outputs output by the columns and are being weighted/tuned, these column outputs are “not combined by the machine-learned version of the coupling section” as recited by this limitation.)
and output an inference result by combining the respective tuned output data; ([Agostinelli, section 3.2 and figure 1]: Based on the mappings given above, the column outputs multiplied by their respective weights are the “respective tuned output data” of the claim. As noted above, the final output, which is obtained by using the weights to “linearly combine the column outputs into a weighted average” [Agostinelli, section 3.2 first paragraph], i.e., by summing the weighted/tuned column outputs. The final output maps to the “inference result” of the claim.)
tune the respective output data using a correction parameter corresponding to an error included in the inference result. ([Agostinelli, section 3.2 and figure 1]: Agostinelli explains that the weights are chosen to minimize a prediction error [Agostinelli, section 3.2 equations (6-8)]. In other words, the quantity being minimzed in [Agostinelli, section 3.2 equation (6)] maps to the “error included in the inference result” of the claim and the weights map to the “correction parameter” of the claim. These mappings ensure that the tuning step as mapped above “us[es the] correction parameter” as recited by the claim. Moreover, the fact that weights are chosen to minimize error means that the “correction parameter” as mapped above falls under the broadest reasonable interpretation of “corresponding to [the] error” as recited by the claim.)
While it is clear from context that the methods of Agostinelli are intended to be implemented on computers, it may be argued that this is not explicitly disclosed therein. Furthermore, the multi-column architecture of Agostinelli is applied on image data rather than on time-series data related to a manufacturing process. In other words, Agostinelli does not distinctly disclose:
An inference device comprising: a memory and a processor that is coupled to the memory and that is configured to: … function as… wherein the processor is configured to
[acquire] a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; … [to process the acquired] time series data group… [by processing the acquired] time series data group
[inspection data] that is obtained from a resultant object obtained by processing the target object;
Mavrovouniotis is in the field of machine learning. It discloses a hierarchical architecture for neural networks, which overlaps with the multi-column architecture of Agostinelli. For example, [Mavrovouniotis, figure 10] can be viewed as a multi-column architecture with four columns, where the value at node 20 is the final output, nodes 16-19 each belong to different columns (so, for instance, the nodes 0, 1, 2, 3, 4, and 16 form one of the four columns), and the weights used for averaging in Agostinelli correspond to the connection weights for the connections coming into node 20. Moreover, Agostinelli in view of Mavrovouniotis also discloses:
An inference device comprising: a memory and a processor that is coupled to the memory and that is configured to: … function as… wherein the processor is configured to ([Mavrovouniotis, section 3]: Mavrovouniotis discloses implementing the methods disclosed therein on “Macintosh II computers” [Mavrovouniotis, section 3 paragraph beginning “The software”]. A Macintosh II computer maps to the “inference device” of the claim, and the memory and processor contained therein to the “memory” and “processor” of the claim.)
[acquire] a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; … [to process the acquired] time series data group… [by processing the acquired] time series data group ([Mavrovouniotis, sections 2.2 and section 3]: Mavrovouniotis discloses data associated to a “process fluid” in “a distillation column as a typical chemical engineering system” [Mavrovouniotis, section 2.2 first paragraph]. There are multiple streams indexed by the variable i, and the variables in the dataset are a flow rate F_i for each stream i, a temperature T_i for each stream i, and three concentrations A_i, B_i, and C_i for each stream i (one concentration for each of three system components A, B, and C) [Mavrovouniotis, section 2.2 first paragraph], with “[t]he value of a variable V_i at time t-kΔt [being] denoted V_i(k)” [Mavrovouniotis, section 2.2 second paragraph]. The dataset is the “time series data group” of the claim, the chemical engineering system is the “manufacturing process” of the claim, the distillation column is the “predetermined processing unit” of the claim, and the process fluid passing through the column is the “target object” of the claim; in the combination, this data is used in place of the image data of Agostinelli. The examiner notes that [Mavrovouniotis, section 3] also describes a simplified dataset with four variables F_1, F_2, T_1, and T_2 (i.e., with two streams and without considering the concentration variables) [Mavrovouniotis, section 3 paragraph beginning “The main example”; see also, Mavrovouniotis, figure 10], and this simplified dataset also falls under the broadest reasonable interpretation of the claim. The examiner notes further that Mavrovouniotis also discloses clustering input variables [Mavrovouniotis, section 2.3], and in the combination, each cluster of variables is passed into the one of the columns of Agostinelli. This combination is elaborated upon in mappings for the dependent claims.)
[inspection data] that is obtained from a resultant object obtained by processing the target object; ([Agostinelli, section 3.2.1; Mavrovouniotis, sections 2.2 and section 3]: As noted above, the true/expected values in the training data map to the “inspection data” of the claim. Since the process fluid passing through the chemical engineering system is mapped above to the “target object” of the claim, the process fluid coming out of the chemical engineering system maps to the “resultant object obtained by processing the target object” of the claim. The true/expected values then fall under the broadest reasonable interpretation of being “obtained from [the] resultant object” as recited by the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the architecture of Agostinelli with clustered time-series data as in Mavrovouniotis because the methods of Mavrovouniotis build “prior knowledge we have about the structure of the physical system, the task at hand, and the kinds of patterns that are most important or most likely” into the architecture of the model and “give[s] the network hints about what kinds of patterns and relationships to look for” [Mavrovouniotis, section 2.1 first paragraph], thereby resulting in a more effective and efficient system.
Claim 3
Agostinelli in view of Mavrovouniotis discloses the elements of the parent claim(s). It also discloses:
[The inference device according to claim 1, wherein the processor is configured to] generate a first time series data group and a second time series data group by processing the acquired time series data group according to a first criterion and a second criterion, respectively, and to process the generated first time series data group and the second time series data group using the machine-learned version of the plurality of networks sections. ([Mavrovouniotis, section 2.3 and figure 10]: As noted under the parent claims, Mavrovouniotis discloses clustering input variables, and in the combination, each cluster of variables is passed to (i.e., “process[ed]… using”) one of the columns of Agostinelli. Any cluster can map to the “first time series data group” of the claim and also to the “second time series data group” of the claim, and the criteria by which the clusters are defined map to the corresponding “criteri[a]” of the claim. For example, Mavrovouniotis discloses a clustering strategy where “[f]or any given process variable, its measurements at different time points form an important cluster” [Mavrovouniotis, section 2.3 first bullet; see also, Mavrovouniotis, figure 10], which means that the data corresponding to one process variable (say F_1) forms the “first time series data group” of the claim, and the property of being a measurement of that process variable is the “first criterion” of the claim. The examiner notes that the broadest reasonable interpretation of the claim does not require the “second time series data group” to be distinct from the “first time series data group”, but such a redundant mapping is not necessary since the measurements of one of the other process variables (say F_2) can also be mapped to the “second time series data group” of the claim.)
The same motivation to combine applies.
Claim 4
Agostinelli in view of Mavrovouniotis discloses the elements of the parent claim(s). It also discloses:
[The inference device according to claim 1, wherein the processor is configured to] divide the acquired time series data group into groups according to a data type or a time range and to process the respective divided groups using the machine-learned version of the plurality of network sections. ([Mavrovouniotis, section 2.3]: As noted above, Mavrovouniotis discloses clustering input variables, and in the combination, each cluster of variables is passed to (i.e., “process[ed]… using”) one of the columns in the architecture of Agostinelli. Clustering input variables falls under the broadest reasonable interpretation of “dividing the acquired time series data group into groups” with the clusters of Mavrovouniotis mapping to the “groups” of the claim. Mavrovouniotis discloses several strategies for clustering, any of which fall under the broadest reasonable interpretation of being “according to a data type of a time range” as recited by the claim. For example, it describes a strategy where “[f]or any given process stream and any fixed time point, the measurements for all the variables of the stream are related” [Mavrovouniotis, section 2.3 second bullet], which falls under the broadest reasonable interpretation of being both “according to a data type” (the stream being the “data type” of the claim) and “according to… a time range” (since each time point represents “time range” of duration Δt [Mavrovouniotis, section 2.2 and figure 4]) as recited by the claim.)
The same motivation to combine applies.
Claim 6
Agostinelli in view of Mavrovouniotis discloses the elements of the parent claim(s). It also discloses:
[The inference device according to claim 1, wherein the processor is configured to:] divide the acquired time series data group into a first time series data group that is measured in accordance with the processing of the target object in a first processing space of the predetermined processing unit, and a second time series data group that is measured in accordance with the processing of the target object in a second processing space; and process the first time series data group and the second time series data group using the machine-learned version of the plurality of network sections. ([Mavrovouniotis, section 2.3]: As noted under the parent claims, Mavrovouniotis discloses clustering input variables, and in the combination, each cluster of variables is passed to (i.e., “process[ed]… using”) one of the columns in the architecture of Agostinelli. In particular, it discloses a clustering strategy where “[f]or any given process variable, its measurements at different time points form an important cluster” [Mavrovouniotis, section 2.3 first bullet]. As noted above, Mavrovouniotis also discloses concentration variables A_i, B_i, and C_i associated with three components A, B, and C in the distillation column. One of the components (say A) maps to the “first processing space” of the claim, and the corresponding cluster of input variables (i.e., the cluster of measurements of A_i) maps to the “first time series data group” of the claim. The examiner notes that the broadest reasonable interpretation of the claim does not require the “second processing space” or the “second time series data group” to be distinct from the first, but such a redundant mapping is not necessary since these can also be mapped, respectively, to a different component (say B) and to the corresponding cluster of input variables (i.e., B_i).)
The same motivation to combine applies.
Claim 10
Agostinelli in view of Mavrovouniotis discloses the elements of the parent claim(s).
[The inference device according to claim 1, wherein the processor is further configured to] select a network section from among the plurality of network sections based on a processing space or a criterion corresponding to identification information included in the acquired time series data group. ([Agostinelli, figure 1; Mavrovouniotis, sections 2.2-2.3]: As noted under the parent claims, Mavrovouniotis discloses clustering input variables, and in the combination, each cluster of variables is passed to one of the columns in the architecture of Agostinelli. Selecting a column to which to pass any particular cluster falls under the broadest reasonable interpretation of “select[ing] a network section” as recited by the claim. The criterion used for clustering maps to the “criterion corresponding to identification information” of the claim, so that the selection of a column is done “based on… [the] criterion” as recited by the claim. Alternatively, the distillation column of Mavrovouniotis has three components A, B, and C, [Mavrovouniotis, section 2.2 first paragraph], and if input variables are clustered based on component, then the selection of network section would also be “based on a processing space” as recited by the claim (the component A, B, or C mapping to the “processing space” of the claim).)
The same motivation to combine applies.
Claim 11
Agostinelli in view of Mavrovouniotis discloses the elements of the parent claim(s). It also discloses:
[The inference device according to claim 1, wherein the processor is further configured to] divide the acquired time series data group into a plurality of data sets and to cause at least one of the plurality of data sets to be processed by a plurality of different network sections among the plurality of network sections. ([Mavrovouniotis, section 2.3]: As noted above, Mavrovouniotis discloses clustering input variables, and the clusters of Mavrovouniotis map to the “plurality of data sets” of the claim. Moreover, the architectures disclosed in Mavrovouniotis include architectures where a single cluster is “processed by a plurality of different network sections” as recited by the claim. For example, the architecture of [Mavrovouniotis, figure 11] includes a number of bifurcations such as the one indicating that the input labeled 0 is processed by both network sections 16 and 17, which is substantially similar to the type of bifurcation described by the claim (and, for example, in [specification, figure 9].)
The same motivation to combine applies.
Claim 8
Agostinelli discloses:
An inference method comprising: acquiring [a time series data group] ([Agostinelli, figure 1]: In the methods disclosed by Agostinelli, the input that is acquired is image data rather than a time series data group. The combination with Mavrovouniotis below discloses the time series data.)
processing the acquired [time series data group] by using a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to process the acquired [time series data group] to output respective output data, the coupling section being configured to combine the respective output data to output a combined result, ([Agostinelli, section 3.2 and figure 1]: Agostinelli discloses a neural network architecture consisting of multiple columns, where each column produces an output, and where each column output is weighted using a weight s_c computed by a weight prediction module [Agostinelli, figure 1(b) and section 3.2]. The columns are the “plurality of network sections” of the claim, the column outputs are the “respective output data” of the claim, the final output (i.e., the weighted average of the column outputs) is the “combined result” of the claim, and the component which computes the weighted average is the “coupling section” of the claim.)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data ([Agostinelli, section 3.2.1]: Agostinelli discloses that the architecture therein “has three training phases: training the [columns], determining optimal weights for a [training set], and then training the weight prediction module” [Agostinelli, section 3.2.1]. This training process ensures that the architecture therein is “machine-learned in an integrated manner such that” the limitation recited by the claim is satisfied. Indeed, the training procedure is designed to minimize error [Agostinelli, section 3.2.2 equation (6)], i.e., which means that the final output of the network (denoted “hat{Y}s”) converges to (“approaches”) the expected/true output (denoted “y”). The expected/true output maps to the “inspection data” of the claim.)
and tuning the respective output data ([Agostinelli, section 3.2 and figure 1]: As noted above, Agostinelli discloses weighting column outputs. Weighting column outputs is a form of adjusting the column outputs (i.e., the “respective output data” of the claim) and maps to the “tun[ing] the respective output data” step of the claim.) that is output by processing the acquired [time series data group] using the machine-learned version of the plurality of network sections ([Agostinelli, section 3.2 and figure 1]: The column outputs are output by the columns (i.e., “the plurality of network sections” of the claim), so the mappings given have the property that the “respective output data… is output by… using the machine-learned version of the plurality of network sections” as recited by this limitation.) and that is not combined by the machine-learned version of the coupling section, ([Agostinelli, section 3.2 and figure 1]: When the column outputs output by the columns and are being weighted/tuned, these column outputs are “not combined by the machine-learned version of the coupling section” as recited by this limitation.)
and outputting an inference result by combining the respective tuned output data; ([Agostinelli, section 3.2 and figure 1]: Based on the mappings given above, the column outputs multiplied by their respective weights are the “respective tuned output data” of the claim. As noted above, the final output, which is obtained by using the weights to “linearly combine the column outputs into a weighted average” [Agostinelli, section 3.2 first paragraph], i.e., by summing the weighted/tuned column outputs. The final output maps to the “inference result” of the claim.)
wherein the respective output data are tuned using a correction parameter corresponding to an error included in the inference result. ([Agostinelli, section 3.2 and figure 1]: Agostinelli explains that the weights are chosen to minimize a prediction error [Agostinelli, section 3.2 equations (6-8)]. In other words, the quantity being minimzed in [Agostinelli, section 3.2 equation (6)] maps to the “error included in the inference result” of the claim and the weights map to the “correction parameter” of the claim. These mappings ensure that the tuning step as mapped above “us[es the] correction parameter” as recited by the claim. Moreover, the fact that weights are chosen to minimize error means that the “correction parameter” as mapped above falls under the broadest reasonable interpretation of “corresponding to [the] error” as recited by the claim.)
The multi-column architecture of Agostinelli is applied on image data rather than on time-series data related to a manufacturing process. In other words, Agostinelli does not distinctly disclose:
[acquiring] a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; [processing the acquired] time series data group… [to process the acquired] time series data group… [by processing the acquired] time series data group
[inspection data] that is obtained from a resultant object obtained by processing the target object;
Mavrovouniotis is in the field of machine learning. It discloses a hierarchical architecture for neural networks, which overlaps with the multi-column architecture of Agostinelli. For example, [Mavrovouniotis, figure 10] can be viewed as a multi-column architecture with four columns, where the value at node 20 is the final output, nodes 16-19 each belong to different columns (so, for instance, the nodes 0, 1, 2, 3, 4, and 16 form one of the four columns), and the weights used for averaging in Agostinelli correspond to the connection weights for the connections coming into node 20. Moreover, Agostinelli in view of Mavrovouniotis also discloses:
[acquiring] a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; … [to process the acquired] time series data group… [by processing the acquired] time series data group ([Mavrovouniotis, sections 2.2 and section 3]: Mavrovouniotis discloses data associated to a “process fluid” in “a distillation column as a typical chemical engineering system” [Mavrovouniotis, section 2.2 first paragraph]. There are multiple streams indexed by the variable i, and the variables in the dataset are a flow rate F_i for each stream i, a temperature T_i for each stream i, and three concentrations A_i, B_i, and C_i for each stream i (one concentration for each of three system components A, B, and C) [Mavrovouniotis, section 2.2 first paragraph], with “[t]he value of a variable V_i at time t-kΔt [being] denoted V_i(k)” [Mavrovouniotis, section 2.2 second paragraph]. The dataset is the “time series data group” of the claim, the chemical engineering system is the “manufacturing process” of the claim, the distillation column is the “predetermined processing unit” of the claim, and the process fluid passing through the column is the “target object” of the claim; in the combination, this data is used in place of the image data of Agostinelli. The examiner notes that [Mavrovouniotis, section 3] also describes a simplified dataset with four variables F_1, F_2, T_1, and T_2 (i.e., with two streams and without considering the concentration variables) [Mavrovouniotis, section 3 paragraph beginning “The main example”; see also, Mavrovouniotis, figure 10], and this simplified dataset also falls under the broadest reasonable interpretation of the claim. The examiner notes further that Mavrovouniotis also discloses clustering input variables [Mavrovouniotis, section 2.3], and in the combination, each cluster of variables is passed into the one of the columns of Agostinelli. This combination is elaborated upon in mappings for the dependent claims.)
[inspection data] that is obtained from a resultant object obtained by processing the target object; ([Agostinelli, section 3.2.1; Mavrovouniotis, sections 2.2 and section 3]: As noted above, the true/expected values in the training data map to the “inspection data” of the claim. Since the process fluid passing through the chemical engineering system is mapped above to the “target object” of the claim, the process fluid coming out of the chemical engineering system maps to the “resultant object obtained by processing the target object” of the claim. The true/expected values then fall under the broadest reasonable interpretation of being “obtained from [the] resultant object” as recited by the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the architecture of Agostinelli with clustered time-series data as in Mavrovouniotis because the methods of Mavrovouniotis build “prior knowledge we have about the structure of the physical system, the task at hand, and the kinds of patterns that are most important or most likely” into the architecture of the model and “give[s] the network hints about what kinds of patterns and relationships to look for” [Mavrovouniotis, section 2.1 first paragraph], thereby resulting in a more effective and efficient system.
Claim 9
Agostinelli discloses:
acquiring [a time series data group] ([Agostinelli, figure 1]: In the methods disclosed by Agostinelli, the input that is acquired is image data rather than a time series data group. The combination with Mavrovouniotis below discloses the time series data.)
processing the acquired [time series data group] by using a machine-learned version of a plurality of network sections and a machine-learned version of a coupling section, the plurality of network sections being configured to process the acquired [time series data group] to output respective output data, the coupling section being configured to combine the respective output data to output a combined result, ([Agostinelli, section 3.2 and figure 1]: Agostinelli discloses a neural network architecture consisting of multiple columns, where each column produces an output, and where each column output is weighted using a weight s_c computed by a weight prediction module [Agostinelli, figure 1(b) and section 3.2]. The columns are the “plurality of network sections” of the claim, the column outputs are the “respective output data” of the claim, the final output (i.e., the weighted average of the column outputs) is the “combined result” of the claim, and the component which computes the weighted average is the “coupling section” of the claim.)
and the plurality of network sections and the coupling section being machine-learned in an integrated manner such that the combined result output from the coupling section approaches inspection data ([Agostinelli, section 3.2.1]: Agostinelli discloses that the architecture therein “has three training phases: training the [columns], determining optimal weights for a [training set], and then training the weight prediction module” [Agostinelli, section 3.2.1]. This training process ensures that the architecture therein is “machine-learned in an integrated manner such that” the limitation recited by the claim is satisfied. Indeed, the training procedure is designed to minimize error [Agostinelli, section 3.2.2 equation (6)], i.e., which means that the final output of the network (denoted “hat{Y}s”) converges to (“approaches”) the expected/true output (denoted “y”). The expected/true output maps to the “inspection data” of the claim.)
and tuning the respective output data ([Agostinelli, section 3.2 and figure 1]: As noted above, Agostinelli discloses weighting column outputs. Weighting column outputs is a form of adjusting the column outputs (i.e., the “respective output data” of the claim) and maps to the “tun[ing] the respective output data” step of the claim.) that is output by processing the acquired time series data group using the machine-learned version of the plurality of network sections ([Agostinelli, section 3.2 and figure 1]: The column outputs are output by the columns (i.e., “the plurality of network sections” of the claim), so the mappings given have the property that the “respective output data… is output by… using the machine-learned version of the plurality of network sections” as recited by this limitation.) and that is not combined by the machine-learned version of the coupling section, ([Agostinelli, section 3.2 and figure 1]: When the column outputs output by the columns and are being weighted/tuned, these column outputs are “not combined by the machine-learned version of the coupling section” as recited by this limitation.)
and outputting an inference result by combining the respective tuned output data; ([Agostinelli, section 3.2 and figure 1]: Based on the mappings given above, the column outputs multiplied by their respective weights are the “respective tuned output data” of the claim. As noted above, the final output, which is obtained by using the weights to “linearly combine the column outputs into a weighted average” [Agostinelli, section 3.2 first paragraph], i.e., by summing the weighted/tuned column outputs. The final output maps to the “inference result” of the claim.)
wherein the respective output data are tuned using a correction parameter corresponding to an error included in the inference result. ([Agostinelli, section 3.2 and figure 1]: Agostinelli explains that the weights are chosen to minimize a prediction error [Agostinelli, section 3.2 equations (6-8)]. In other words, the quantity being minimzed in [Agostinelli, section 3.2 equation (6)] maps to the “error included in the inference result” of the claim and the weights map to the “correction parameter” of the claim. These mappings ensure that the tuning step as mapped above “us[es the] correction parameter” as recited by the claim. Moreover, the fact that weights are chosen to minimize error means that the “correction parameter” as mapped above falls under the broadest reasonable interpretation of “corresponding to [the] error” as recited by the claim.)
While it is clear from context that the methods of Agostinelli are intended to be implemented on computers, it may be argued that this is not explicitly disclosed therein. Furthermore, the multi-column architecture of Agostinelli is applied on image data rather than on time-series data related to a manufacturing process. In other words, Agostinelli does not distinctly disclose:
A non-transitory recording computer readable medium storing an inference program that causes a computer to execute:
[acquiring] a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; [processing the acquired] time series data group… [to process the acquired] time series data group… [by processing the acquired] time series data group
[inspection data] that is obtained from a resultant object obtained by processing the target object;
Mavrovouniotis is in the field of machine learning. It discloses a hierarchical architecture for neural networks, which overlaps with the multi-column architecture of Agostinelli. For example, [Mavrovouniotis, figure 10] can be viewed as a multi-column architecture with four columns, where the value at node 20 is the final output, nodes 16-19 each belong to different columns (so, for instance, the nodes 0, 1, 2, 3, 4, and 16 form one of the four columns), and the weights used for averaging in Agostinelli correspond to the connection weights for the connections coming into node 20. Moreover, Agostinelli in view of Mavrovouniotis also discloses:
A non-transitory recording computer readable medium storing an inference program that causes a computer to execute: ([Mavrovouniotis, section 3]: Mavrovouniotis discloses implementing the methods disclosed therein as “software… running on Macintosh II computers” [Mavrovouniotis, section 3 paragraph beginning “The software”]. The software is the “inference program” of the claim, the Macintosh II computer is the “computer” of the claim, and its hard drive is the “non-transitory recording computer readably medium” of the claim.)
[acquiring] a time series data group measured in accordance with processing of a target object in a predetermined processing unit of a manufacturing process; … [to process the acquired] time series data group… [by processing the acquired] time series data group ([Mavrovouniotis, sections 2.2 and section 3]: Mavrovouniotis discloses data associated to a “process fluid” in “a distillation column as a typical chemical engineering system” [Mavrovouniotis, section 2.2 first paragraph]. There are multiple streams indexed by the variable i, and the variables in the dataset are a flow rate F_i for each stream i, a temperature T_i for each stream i, and three concentrations A_i, B_i, and C_i for each stream i (one concentration for each of three system components A, B, and C) [Mavrovouniotis, section 2.2 first paragraph], with “[t]he value of a variable V_i at time t-kΔt [being] denoted V_i(k)” [Mavrovouniotis, section 2.2 second paragraph]. The dataset is the “time series data group” of the claim, the chemical engineering system is the “manufacturing process” of the claim, the distillation column is the “predetermined processing unit” of the claim, and the process fluid passing through the column is the “target object” of the claim; in the combination, this data is used in place of the image data of Agostinelli. The examiner notes that [Mavrovouniotis, section 3] also describes a simplified dataset with four variables F_1, F_2, T_1, and T_2 (i.e., with two streams and without considering the concentration variables) [Mavrovouniotis, section 3 paragraph beginning “The main example”; see also, Mavrovouniotis, figure 10], and this simplified dataset also falls under the broadest reasonable interpretation of the claim. The examiner notes further that Mavrovouniotis also discloses clustering input variables [Mavrovouniotis, section 2.3], and in the combination, each cluster of variables is passed into the one of the columns of Agostinelli. This combination is elaborated upon in mappings for the dependent claims.)
[inspection data] that is obtained from a resultant object obtained by processing the target object; ([Agostinelli, section 3.2.1; Mavrovouniotis, sections 2.2 and section 3]: As noted above, the true/expected values in the training data map to the “inspection data” of the claim. Since the process fluid passing through the chemical engineering system is mapped above to the “target object” of the claim, the process fluid coming out of the chemical engineering system maps to the “resultant object obtained by processing the target object” of the claim. The true/expected values then fall under the broadest reasonable interpretation of being “obtained from [the] resultant object” as recited by the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine the architecture of Agostinelli with clustered time-series data as in Mavrovouniotis because the methods of Mavrovouniotis build “prior knowledge we have about the structure of the physical system, the task at hand, and the kinds of patterns that are most important or most likely” into the architecture of the model and “give[s] the network hints about what kinds of patterns and relationships to look for” [Mavrovouniotis, section 2.1 first paragraph], thereby resulting in a more effective and efficient system.
Claim(s) 5 and 12 is/are rejected under 35 USC 103 as being unpatentable over Agostinelli in view of Mavrovouniotis, further in view of Dan CIREȘAN et al. (Multi-column Deep Neural Networks for Image Classification, published 2012; hereafter, “Cireșan”).
Claim 5
Agostinelli in view of Mavrovouniotis discloses the elements of the parent claim(s). It does not distinctly disclose:
[The inference device according to claim 1, wherein the processor is configured to] process the acquired time series data group using the machine-learned version of the plurality of network sections, each of which includes a normalization section that performs a normalization process using a different method.
Cireșan is in the field of machine learning and discloses a multi-column architecture. Moreover, Agostinelli in view of Mavrovouniotis and Cireșan discloses:
[The inference device according to claim 1, wherein the processor is configured to] process the acquired time series data group using the machine-learned version of the plurality of network sections, each of which includes a normalization section that performs a normalization process using a different method. ([Cireșan, section 2 and figure 1(b)]: In the multi-column architecture of Cireșan, the columns “can be trained… on inputs preprocessed in different ways” [Cireșan, section 2 last paragraph]. More precisely, there are preprocessing blocks P_0, …, P_{n-1} and “[a]n arbitrary number of columns can be trained on inputs preprocessed in different ways” [Cireșan, figure 1(b) and caption]. The preprocessing block maps to the “normalization section” of the claim, and the preprocessing itself maps to the “normalization process” of the claim; the examiner notes that Cireșan discloses an exemplary embodiment where the preprocessing is described as a “normalization” [Cireșan, section 3.1]. In the combination, the multi-column architectures of Agostinelli and Cireșan are combined so that each column has different preprocessing blocks as described in Cireșan, and so that column outputs undergo weighted averaging as described in Agostinelli.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine weighted averaging of column outputs as in Agostinelli in view of Mavrovouniotis with the use of different preprocessing blocks for columns as in Cireșan because preprocessing inputs in different ways “helps to reduce both error rate and number of columns required to reach a given accuracy” [Cireșan, section 2 last paragraph], so the combination would achieve the advantages of both architectures and be more effective and efficient overall.
Claim 12
Agostinelli in view of Mavrovouniotis discloses the elements of the parent claim(s). It does not distinctly disclose:
[The inference device according to claim 11, wherein the processor is further configured to] cause each of the at least one of the plurality of data sets to be processed by a plurality of normalization processes that use different methods.
Cireșan is in the field of machine learning and discloses a multi-column architecture. Moreover, Agostinelli in view of Mavrovouniotis and Cireșan discloses:
[The inference device according to claim 11, wherein the processor is further configured to] cause each of the at least one of the plurality of data sets to be processed by a plurality of normalization processes that use different methods. ([Cireșan, section 2 and figure 1(b)]: In the multi-column architecture of Cireșan, the columns “can be trained… on inputs preprocessed in different ways” [Cireșan, section 2 last paragraph]. More precisely, there are preprocessing blocks P_0, …, P_{n-1} and “[a]n arbitrary number of columns can be trained on inputs preprocessed in different ways” [Cireșan, figure 1(b) and caption]. The preprocessing block maps to the “normalization section” of the claim, and the preprocessing itself maps to the “normalization process” of the claim; the examiner notes that Cireșan discloses an exemplary embodiment where the preprocessing is described as a “normalization” [Cireșan, section 3.1]. In the combination, the multi-column architectures of Agostinelli and Cireșan are combined so that each column has different preprocessing blocks as described in Cireșan, and so that column outputs undergo weighted averaging as described in Agostinelli. As noted under the parent claim, Agostinelli in view of Mavrovouniotis already discloses a single group being processed by a plurality of network sections. This means that, in the combination with Cireșan, a single group is in fact “processed by a plurality of normalization processes that use different methods” as recited by the claim, since it is processed by each of the normalization processes in each of the network sections by which it is processed.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to combine weighted averaging of column outputs as in Agostinelli in view of Mavrovouniotis with the use of different preprocessing blocks for columns as in Cireșan because preprocessing inputs in different ways “helps to reduce both error rate and number of columns required to reach a given accuracy” [Cireșan, section 2 last paragraph], so the combination would achieve the advantages of both architectures and be more effective and efficient overall.
Claim(s) 7 is/are rejected under 35 USC 103 as being unpatentable over Agostinelli in view of Mavrovouniotis, further in view of Raman NURANI et al. (US20190121237A1, published 2019-04-25; hereafter, “Nurani”).
Claim 7
Agostinelli in view of Mavrovouniotis discloses the elements of the parent claim(s). In view of the fact that “substrate”, in the context of the present application, refers to silicon wafers, it does not distinctly disclose:
[The inference device according to claim 1, wherein] the time series data group is data measured in accordance with processing in a substrate processing device.
Nurani is in the field of machine learning. Moreover, Agostinelli in view of Mavrouvniotis and Nurani discloses:
[The inference device according to claim 1, wherein] the time series data group is data measured in accordance with processing in a substrate processing device. ([Nurani, 0019]: Nurani discloses “gathering data about various processes and parameters of wafers during the manufacturing process” where the “data may include, for example, time series data about the behavior of various wafer parameters… and tool parameters throughout the manufacturing process” [Nurani, 0019]. In the combination, the time-series data related to distillation as disclosed in Mavrovouniotis is replaced with the time-series data related to wafer manufacturing as in Nurani. The wafer manufacturing tools are the “substrate processing device” of the claim, and manufacturing the wafers is the “processing in a substrate processing device” of the claim.)
Before the effective filing date of the invention, it would have been obvious to a person of ordinary skill in the art to employ the neural network architectures of Agostinelli in view of Mavrovouniotis with an application directed towards manufacturing wafers because “there is a need for a more intelligent and efficient way to determine the quality of a large number of manufactured wafers” [Nurani, 0018] and the combination would provide an intelligent and efficient way of performing this task.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Shishir AGRAWAL whose telephone number is +1 703-756-1183. The examiner can normally be reached Monday through Thursday, 08:30-14:30 Pacific Time.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Alexey SHMATOV can be reached on +1 571-270-3428. The fax phone number for the organization where this application or proceeding is assigned is +1 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at +1 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call +1 800-786-9199 (IN USA OR CANADA) or +1 571-272-1000.
/S.A./Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123