DETAILED ACTION
This non-final action is responsive to application 18/760,084 as submitted on July 01, 2024.
Claim status is currently pending and under examination for claims 1-11 of which independent claims are 1, 6, 9 and 11.
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 .
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) are:
In Claim 9:
“a module for determining …”
“a module for generating …”
“a module for grouping …”
“a module for applying …”
In Claim 10:
“a module for applying …”
In Claim 11:
“a module for determining …”
“a module for generating …”
“a module for grouping …”
“a module for applying …”
“a module for learning …”
Because these claim limitation(s) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, they are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-11 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 1 recites the limitation “said subset of data” in lines 13-14. There is a lack of antecedent basis for “said subset of data”. It is not clear whether “said subset of data” refers to the first initial set of data or the second initial set of data, therefore the claim is rendered indefinite. For examination, the Examiner interprets “said subset of data” to refer to a subset of training data.
Dependent Claims 2-5 and 7 inherit issues of indefiniteness, therefore, the same rationale of rejection applies to Claims 2-5 and 7.
Claim 6 recites the limitation “the values of the prediction model”. There is a lack of antecedent basis for “the prediction model”, therefore the claim is rendered indefinite. For examination, the Examiner interprets “the prediction model” to refer to the “model for predicting a risk of occurrence of a feared rare event.” Furthermore, there is a lack of antecedent basis for “the values”, therefore the claim is rendered indefinite. For examination, “the values of the prediction model” is interpreted to refer to hyperparameters.
Dependent Claim 8 is dependent on Claim 6 and inherits issues of indefiniteness, therefore, the same rationale of rejection applies to Claim 8.
Claim 9 recites similar limitations corresponding to Claim 1, therefore the same rationale of rejection applies to Claim 9.
Dependent Claim 10 is dependent on Claim 9 and inherits issues of indefiniteness, therefore, the same rationale of rejection applies to Claim 10.
Claim 11 recites similar limitations corresponding to Claim 6, therefore the same rationale of rejection applies to Claim 11.
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 1-11 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Independent Claims 1, 6, 9 and 11
Step 2A Prong One: Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes, independent claim 1, under the broadest reasonable interpretation, recites the following limitations that are abstract ideas:
A) determination, from said first and second initial datasets, by a machine learning method, of at least one subset of variables associated with a risk of occurrence of said rare event higher than a risk threshold, said risk being calculated from the values of said variables, (mental process)
B) generation of a third set of data by pseudo-random variations on said subset of data, and calculation of an associated risk, (mental process)
C) grouping said first, second and third datasets into an augmented training database, (mental process)
The “determination” step involves identifying which variables are associated with a high risk of occurrence of a rare event which amounts to no more than observations, evaluations, and judgments that can be performed in the human mind or with the use of a physical aid (e.g., pen and paper). The claim recites the step of determining a subset of variables at a high degree of generality, thus the step is not required to have any specific level of complexity that would preclude the step from being mental processes. Therefore, the “determination” step is considered to be mental processes, see MPEP § 2106.04(a)(2)(III).
The “generation” step involves determining new data samples by changing values of a subset of data and using a calculated risk which amounts to no more than observations, evaluations, and judgments that can be performed in the human mind or with the use of a physical aid (e.g., pen and paper). The claim recites the step of generating a third set of data at a high degree of generality, thus the step is not required to have any specific level of complexity that would preclude the step from being mental processes. Therefore, the “generation” step is considered to be mental processes, see MPEP § 2106.04(a)(2)(III).
The “grouping” step involves identifying a collection of augmented training data using data from three datasets which amounts to no more than observations, evaluations, and judgments that can be performed in the human mind or with the use of a physical aid (e.g., pen and paper). The claim recites the step of grouping datasets at a high degree of generality, thus the step is not required to have any specific level of complexity that would preclude the step from being mental processes. Therefore, the “grouping” step is considered to be mental processes, see MPEP § 2106.04(a)(2)(III).
Therefore, the independent claim recites a judicial exception. Independent claims 6, 9 and 11 recite similar limitations corresponding to claim 1, therefore the same subject matter eligibility analysis is applied.
Step 2A Prong Two: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, the judicial exception recited above is not integrated into a practical application. The claims recite the following additional elements, but these additional elements are not sufficient to integrate the judicial exception into a practical application:
said training data being formed by vectors of variables representative of a state of at least one system monitored before the occurrence of said feared rare event, (MPEP § 2106.05(f) mere instructions to implement an abstract idea on a computer, or generally links exception to a technological environment)
from input training data including a first initial set of data associated with an absence of occurrence of said rare event, and a second initial set of data associated with a presence of occurrence of said rare event, (MPEP § 2106.05(f) mere instructions to implement an abstract idea on a computer, or generally links exception to a technological environment)
the method including steps, implemented by at least one computer processor, of: (MPEP § 2106.05(f) mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea)
A) determination, from said first and second initial datasets, by a machine learning method, of at least one subset of variables associated with a risk of occurrence of said rare event higher than a risk threshold, said risk being calculated from the values of said variables, (MPEP § 2106.05(f) mere instructions to implement an abstract idea on a computer, or generally links exception to a technological environment)
D) application of a first classification model by unsupervised learning based on augmented training data, resulting in at least three groups of training data, comprising a first group of training data associated with a risk of occurrence of said rare event lower than a first threshold; a second group of training data associated with a risk of occurrence of said rare event higher than a second threshold; a third group of training data associated with a third risk, lower than the second threshold and higher than the first threshold. (MPEP § 2106.05(f) mere instructions to implement an abstract idea on a computer, or generally links exception to a technological environment)
and further comprising learning the values of the prediction model using the training data obtained by the method of generating training data. (claim 6, claim 11) (MPEP § 2106.05(f) mere instructions to implement an abstract idea on a computer, or generally links exception to a technological environment)
the device including at least one calculation processor configured to implement: (claim 9, claim 11) (MPEP § 2106.05(f) mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea)
The “said training data being formed …” step is recited at a high-level of generality such that the limitation amounts to no more than mere instructions to “apply” the judicial exception on a computer. It can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
The “from input training data including …” step is recited at a high-level of generality such that the limitation amounts to no more than mere instructions to “apply” the judicial exception on a computer. It can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
The “determination” step requires a “machine learning method” to determine a subset of variables. The machine learning method is used to apply the recited judicial exception without placing any limitation on how the machine learning method operates. The limitation amounts to mere instructions to “apply” the judicial exception on a computer. It can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
The “application” step is recited at a high-level of generality such that the limitation amounts to no more than mere instructions to “apply” the judicial exception on a computer. It can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
The “learning” step is recited at a high-level of generality such that the limitation amounts to no more than mere instructions to “apply” the judicial exception on a computer. It can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
The remaining additional elements are recited at a high-level of generality such that they amount to no more than mere instructions to “apply” an exception using a generic component. 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, see MPEP § 2106.05(f).
Therefore, the above limitations do not integrate the judicial exception into a practical application.
Step 2B: Does the claim recite additional elements that amount to significantly more than the judicial exception?
No. The claims do not include additional elements that are sufficient for the claims to amount to significantly more than the judicial exception.
In regards to the “machine learning method” in the “determination” step, the limitations are recited so generically such that they amount to no more than mere instructions to “apply” the judicial exception on a computer using generic computer components. Mere instructions to apply a judicial exception cannot provide an inventive concept. See MPEP § 2106.05(f).
In regards to the “said training data being formed …”, “from input training data including …”, “application”and “learning” steps and the remaining additional elements, the limitations are recited so generically such that they amount to no more than mere instructions to “apply” the judicial exception on a computer using generic computer components. Mere instructions to apply a judicial exception cannot provide an inventive concept. See MPEP § 2106.05(f).
Therefore, independent claims 1, 6, 9 and 11 are not patent eligible.
Dependent Claims 2-5, 7-8 and 10
The remaining dependent claims being rejected do not recite additional elements, whether considered individually or in combination, that are sufficient to integrate the judicial exception into a practical application or amount to significantly more than a judicial exception.
Claim limitation
Examiner analysis
2 and 10. The method according claim 1, further including E) an application of at least one second machine learning classification model to reclassify at least the training data of the third group into the first group or into the second group,
and estimation of an associated reclassification error probability,
the first group of training data thereby obtained forming a first output set of training data and the second group of training data thereby obtained forming a second output set of training data.
The “estimation” step is a mental process akin to a human evaluation/judgment/observation.
The “application” step is recited at a high-level of generality such that the limitations amount to no more than mere instructions to “apply” the judicial exception on a computer. They can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
3. The method according to claim 2, wherein step E) includes the application of a plurality of second classifier models:
application of a semi-supervised learning classifier to the training data of the first group and the training data of the third group,
application of a supervised learning classifier to the training data of the first group and the training data of the second group
and application of an unsupervised learning classifier to the training data of the second group and the training data of the third group.
The step is recited at a high-level of generality such that the limitations amount to no more than mere instructions to “apply” the judicial exception on a computer. They can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
4. The method according to claim 2, wherein if said reclassification error probability is higher than an error threshold, steps A) to D) are iterated,
said first group of training data forming the first initial set of data and said second group of training data forming said second initial set of data for a subsequent iteration.
The step is recited at a high-level of generality such that the limitations amount to no more than mere instructions to “apply” the judicial exception on a computer. They can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
5. The method according to claim 1, in which a random forest machine learning method is applied in the determining step A).
The step is recited at a high-level of generality such that the limitations amount to no more than mere instructions to “apply” the judicial exception on a computer. They can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
7. A computer program including software instructions which, when executed by a programmable device, implement a method of generating training data according to claim 1.
The step is recited at a high-level of generality such that the limitations amount to no more than mere instructions to “apply” the judicial exception on a computer. They can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
8. A computer program including software instructions which, when executed by a programmable device, implement a machine learning method according to claim 6.
The step is recited at a high-level of generality such that the limitations amount to no more than mere instructions to “apply” the judicial exception on a computer. They can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of computers, see MPEP § 2106.05(f).
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The following are the references relied upon in the rejections below:
Perboli (EP 4033421 A1)
Mugan (US 11977991 B1)
Yang, Yiben, et al. "Generative data augmentation for commonsense reasoning." Findings of the Association for Computational Linguistics: EMNLP 2020. 2020.
Claims 1-2 and 5-11 are rejected under 35 U.S.C. 103 as being unpatentable over Perboli / Mugan / Yang.
Regarding Claim 1, Perboli teaches:
A method for generating training data for machine learning of a model for predicting a risk of occurrence of a feared rare event (The Examiner interprets “feared rare event” according to its broadest reasonable interpretation (BRI) in view of the Applicant’s specification as encompassing a malfunction. This interpretation is consistent with the illustrative descriptions in the Applicant’s specification at (P. 6, Lines 27-29) and (P. 5, Line 34 to P. 6, Line 2), (see excerpt below).
Applicant’s written description at (P. 6, Lines 27-29): “The training data are recorded before the occurrence of a feared rare event, but for at least part of the training data, the occurrence of the rare event has been noted (i.e. feared malfunction or reported rare disease).”
Applicant’s written description at (P. 5, Line 34 to P. 6, Line 2): “a system 2 for generating training data for machine learning of a model for predicting a risk of occurrence of a feared rare event. System 2 is presented generically. There are many applications of such a system, e.g. for predicting the risk of exceptional malfunction of a monitored industrial system”
[0062-0063] “train the AI artificial intelligence algorithm in order to make reliable malfunction predictions. In detail, the learning procedure 100 comprises acquiring a set of entity data DB referred to training entities from the databases”
[0043] “the step of predicting a malfunction of the entity by means of an artificial intelligence system comprises receiving a data group referred to the entity at the input of the artificial intelligence system and calculating a plurality of probability of malfunction of the entity”
[0054] “the monitored entity 4 is an industrial plant comprising a plurality of automated processing stations.”
Entity data (‘training data’) is acquired (‘generated’) for training an artificial intelligence algorithm (‘model’) to predict a malfunction of a monitored industrial plant (entity). A malfunction is predicted by calculating a probability of a malfunction for the industrial plant, therefore the probability of a malfunction is a risk of a malfunction occurring and therefore a malfunction is a feared rare event.),
said training data being formed by vectors of variables representative of a state of at least one system monitored before the occurrence of said feared rare event ([0065] “each entity data group comprises values referred to characteristic parameters of the respective training entity, both time variants and time invariants. … the time-dependent parameters comprise operating parameters of the training entity (instantaneous power absorbed, revolutions per minute of a mechanical/electromechanical component, quantity of input raw material, quantity of output finished or semi-finished product, etc.). In particular, the time-dependent parameters may comprise a plurality of values acquired at different instants of time, a trend in the value of the parameters over said time interval,”
[0066] “It will be noted that the data groups are then identified, or labelled, based on whether the corresponding training entity is subject to a malfunction or not.”
Each training entity (‘systems monitored’) that is monitored has an entity data group comprising time-dependent parameters. Time-dependent parameters represent values of operating parameters of a training entity at different time instances, therefore each time instance represents a different state of a monitored training entity and therefore time-dependent parameters are vectors of variables representative of a state. The data groups comprising time-dependent parameters are labelled based on whether a training entity is subject to a malfunction, therefore a malfunction has not yet occurred, and time-dependent parameters therefore represent a state monitored before the occurrence of a feared rare event (malfunction).),
from input training data including a first initial set of data associated with an absence of occurrence of said rare event, and a second initial set of data associated with a presence of occurrence of said rare event ([0023] “by means of the untrained artificial intelligence system, identifying data groups associated with training entities subject to a malfunction, and data groups associated with entities not subject to a malfunction; c. by means of the untrained artificial intelligence system, composing a training subset, wherein the subset comprises a first quantity of data groups associated with the entities subject to malfunction, the second quantity being greater than or equal to the first quantity”
Data groups (‘training data’) are organized into two subsets of training sets. One subset comprises a first quantity of data groups associated with entities subject to malfunction, therefore this subset (‘second initial set of data’) is associated with a presence of a malfunction (occurrence of rare events). Another subset comprises a second quantity of data groups associated with entities not subject to malfunction, therefore this subset (‘first initial set of data’) is associated with an absence of a malfunction (absence of occurrence of rare events).),
the method including steps, implemented by at least one computer processor, of ([0057] processors):
A) determination, from said first and second initial datasets, by a machine learning method, of at least one subset of variables associated with a risk of occurrence of said rare event higher than a risk threshold, said risk being calculated from the values of said variables ([0137] “In a preferred embodiment, the failure predictions and the input data are further processed in order to identify the parameters which have the greatest impact on the failure probability of the monitored company.”
[0107] “Furthermore, the artificial intelligence algorithm generates an indication of one or more characteristic parameters of the monitored entity 4 that are the most relevant results in the calculation of the probability of malfunction for each time interval considered.”
[0089] “a threshold malfunction probability (for example 60%) is selected from those calculated in block 112 based on which the AI artificial intelligence algorithm categorises a training entity as malfunctioning or non-malfunctioning by comparing the failure probability calculated for the entity with the threshold malfunction probability.”
An artificial intelligence algorithm (‘machine learning method’) identifies the characteristic parameters that have the greatest impact on malfunction (failure) probability, therefore the probability of a malfunction (risk of occurrence of a rare event) is calculated using values of characteristic parameters (‘values of variables’). The characteristic parameters that are most impactful (‘subset of variables’) are used to calculate a probability of a malfunction and the probability is compared to a threshold malfunction probability (‘risk threshold’) to categorize entities as malfunctions, therefore the most impactful characteristic parameters are associated with a risk of occurrence (probability of a malfunction) higher than a threshold malfunction probability.),
D) application of a first classification model by [supervised] learning based on … training data, resulting in at least three groups of training data ([0031-0033] “step g. of composing the training dataset by selecting a plurality of data groups of training entities with a failure probability lower than the threshold value, comprises: selecting a further threshold value of the failure probability, said further threshold value being lower than the threshold value, selecting a first quantity of data groups associated with entities having a failure probability lower than the further threshold value, and selecting a second quantity of data groups associated with entities having a failure probability comprised between the further threshold value and the threshold value. … This makes it possible to obtain balanced training data starting from a strongly unbalanced training data”
[0089] “a threshold malfunction probability (for example 60%) is selected from those calculated in block 112 based on which the AI artificial intelligence algorithm categorises a training entity as malfunctioning or non-malfunctioning”
[0041] “the step of predicting a malfunction of the entity by means of an artificial intelligence system comprises using a machine learning algorithm, selected from: Random Forest”
A training dataset is composed of training entities that are categorized according to a threshold value by using Random Forest. Entities with a malfunction probability greater than a threshold value (60%) are classified as malfunctioning, therefore resulting in a group of training data. A second threshold, further threshold, is selected that is less than the threshold value (60%). Entities with a malfunction probability less than the further threshold are selected, therefore resulting in a group of training data. Entities with a malfunction probability between the further threshold and threshold value (60%) are selected, therefore resulting in a group of training data.),
comprising a first group of training data associated with a risk of occurrence of said rare event lower than a first threshold ([0031] “selecting a further threshold value of the failure probability, said further threshold value being lower than the threshold value, selecting a first quantity of data groups associated with entities having a failure probability lower than the further threshold value,
A further threshold value (‘first threshold’) is selected and used to group (select) entities. Entities with a malfunction (failure) probability lower than the further threshold value are selected, resulting in a first group of training data associated with a risk of occurrence of rare event (malfunction probability) lower than a first threshold.);
a second group of training data associated with a risk of occurrence of said rare event higher than a second threshold ([0089] “a threshold malfunction probability (for example 60%) is selected from those calculated in block 112 based on which the AI artificial intelligence algorithm categorises a training entity as malfunctioning or non-malfunctioning”
Entities with a malfunction probability greater than a 60% threshold value (‘second threshold’) are classified as malfunctioning, therefore resulting in a second group of training data associated with a risk of occurrence of rare event (malfunction probability) higher than a second threshold.);
a third group of training data associated with a third risk, lower than the second threshold and higher than the first threshold ([0031] “and selecting a second quantity of data groups associated with entities having a failure probability comprised between the further threshold value and the threshold value”
Entities with a malfunction probability between the further threshold value (‘first threshold’) and the threshold value (‘second threshold’) are selected, resulting in a third group of training data. Since the malfunction probability is between the further threshold value and the threshold value, the entities selected have a malfunction probability (‘third risk’) that is lower than the threshold value (‘second threshold’) and higher than the further threshold value (‘first threshold’).).
However, Perboli does not teach generating a third set of data by pseudo-random variations, which is taught by Mugan:
B) generation of a third set of data by pseudo-random variations on said subset of data, and calculation of an associated risk (The Examiner interprets “feared rare event” according to its BRI in view of the Applicant’s specification as encompassing a low-incidence event. This interpretation is consistent with the illustrative descriptions in the Applicant’s specification at (P. 6, Lines 27-29) and (P. 5, Line 34 to P. 6, Line 2).
(Abstract) “a generative ML model that produces synthetic minority-class records to form an augmented training data set, … A method for producing actionable predictions of a low-incidence event by applying ML models to imbalanced class data by producing a prediction by a predictive ML model that has been trained on a data set comprising synthetic minority-class data records produced by a generative ML model, … The low-incidence event may comprise risk of opioid use disorder.”
(Col. 7, Lines 61-65) “When the generator neural network 530 creates a synthetic record 540 from a random number 520, the discriminator neural network 550 uses what it has learned from the real records 510 to determine whether the synthetic record is real or not”
(Col. 5, Lines 35-40) “a generative model is a feature predictor that generates records by changing existing records such they could still be the same minority class. An exemplary embodiment of this approach augments minority class data by learning to generate variations of the minority class records that are still likely to be in that minority class.”
A generative ML model is used to produce synthetic records of a minority class that is a low-incidence event (‘feared rare event’). People at risk of opioid use disorder is a low-incidence event that is predicted, therefore a low-incidence event is a feared rare event. The generative ML model uses a random number (see Figure 5 depicting the random number is a random seed) to generate synthetic records. The random seed is used to generate variations of minority class records (low-incidence events) to create synthetic records, therefore the synthetic records (‘third set of data’) are generated by pseudo-random variations on minority class records (‘subset of data’). When creating synthetic records, the likelihood of a created synthetic record remaining in the minority class is considered to keep the synthetic records as low-incidence events, therefore generating synthetic records based on a calculation of an associated risk (the likelihood of a synthetic record still being a low-incidence event).),
C) grouping said first, second and third datasets into an augmented training database ((Col. 4, Lines 45-50) “The synthetic minority-class data records 325 are combined with actual minority class data records 310 (e.g., records of patients who overdosed) and actual majority class data records 305 (i.e., patients who did not overdose) to make an augmented data set of training data 335.”),
D) application of a first classification model by [supervised] learning based on augmented training data, ((Col. 4, Lines 49-54) “an augmented data set of training data 335. The training data 335 can be used as training data (330) to train predictive model 360. After it has been trained, the predictive model 360 can be used to make predictions regarding live data 375 from live or production data set”),
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the method of Perboli with the synthetic training data generation technique disclosed by Mugan to generate an augmented training set. By generating an augmented training set, the minority class in a training set can be augmented by generating synthetic samples that are similar to the samples of the minority class, thereby resulting in a more balanced training dataset that can help reduce a model’s bias towards the majority class.
Furthermore, the combined method of Perboli / Mugan does not teach applying a first classification model by unsupervised learning based on augmented training data, which is taught by Yang:
D) application of a first classification model by unsupervised learning based on augmented training data, resulting in at least three groups of training data ((P. 1010, Sec. 2.1, Last Paragraph) “we consider using a supervised task model trained on the original training set, to re-label a candidate pool of synthetic answers and distractors. This is similar to treating the synthetic questions as unlabeled data and applying self-training. The utility of this self-training can be task-dependent;”
(P. 1015, Sec. 5.2, ¶1) “A related question is whether G-DAUGc’s optional self-supervised relabeling improves performance.”
(P. 1009, Figure 2 Caption) “Illustration of the G-DAUGc process: (1) generate synthetic data and train a task model, (2) relabel the generated data using the task model,”
(P. 1009, Sec. 2.1, ¶1) “will use multiple choice question answering as a running example to describe synthetic data generation … and a ground truth label yi ∈ {1,2,...,K}.”
A task model applies self-training (self-supervised learning) to relabel generated synthetic data (‘augmented training data’), therefore using unsupervised learning on synthetic data to result in groups of training data (assigning labels (groups) to synthetic data).),
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined method of Perboli / Mugan with the labeling technique disclosed by Yang to use unsupervised learning to classify augmented training data. By using unsupervised learning to classify augmented training data, labels can be applied to newly generated synthetic samples without human intervention, thereby reducing the time it takes to perform data labeling.
Regarding Claims 2 and 10, the combined method of Perboli / Mugan / Yang teaches:
The method according claim 1, further including E) an application of at least one second machine learning classification model to reclassify at least the training data of the third group into the first group or into the second group (Yang discloses “Illustration of the G-DAUGc process: (1) generate synthetic data and train a task model, (2) relabel the generated data using the task model,” (P. 1009, Figure 2 Caption)
Yang discloses “will use multiple choice question answering as a running example to describe synthetic data generation … and a ground truth label yi ∈ {1,2,...,K}.” (P. 1009, Sec. 2.1, ¶1)
A task model (‘second machine learning classification model’) is used to relabel (reclassify) generated synthetic data. Each synthetic data sample has a label from a set of ground truth labels yi ∈ {1,2,...,K}, therefore each ground truth label is a group (y1 is first group, y2 is second group, yK is third group). Synthetic data samples labeled with yK (‘training data of the third group’) are then relabeled with another ground truth label, therefore reclassifying synthetic data samples of a third group yK into a first group (ground truth label y1) or second group (ground truth label y2).),
and estimation of an associated reclassification error probability (Yang discloses “we used validation performance to determine whether or not to relabel our synthetic training data” (P. 1010, Sec. 2.1, Last Paragraph)
Yang discloses “In practice, since the generalization error is usually approximated by validation loss, a training example xi is considered detrimental if it increases validation loss” (P. 1022, Sec. D, ¶1)
(P. 1011, Sec. 3.1, ¶3) “detrimental synthetic data (i.e., the examples that have a positive estimated influence on the validation loss).”),
the first group of training data thereby obtained forming a first output set of training data (Synthetic training data is relabeled by changing each synthetic sample’s ground truth label (see P. 1009, Figure 2 Caption). The synthetic training samples that are relabeled with a ground truth label of y1 form a first output set of training data.)
and the second group of training data thereby obtained forming a second output set of training data (The synthetic training samples that are relabeled with a ground truth label of y2 form a second output set of training data.).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined method of Perboli / Mugan / Yang with the relabeling technique disclosed by Yang to relabel augmented training data. By relabeling augmented training data, an augmented sample can be relabeled to ensure that its label is correct, thereby ensuring that a machine learning model is only trained with correct labels thus improving classification accuracy.
Regarding Claim 5, the combined method of Perboli / Mugan / Yang teaches:
The method according to claim 1, in which a random forest machine learning method is applied in the determining step A) (Perboli discloses [0137] “the failure predictions and the input data are further processed in order to identify the parameters which have the greatest impact on the failure probability of the monitored company.”
[0041] “the step of predicting a malfunction of the entity by means of an artificial intelligence system comprises using a machine learning algorithm, selected from: Random Forest”
The parameters that have the greatest impact on malfunction (failure) probability are identified by using malfunction probability calculated by a Random Forest model.).
Regarding Claim 6, the rejection of claim 1 is incorporated. The difference in scope being:
A method of machine learning of a model for predicting a risk of occurrence of a feared rare event (Perboli discloses [0043] “the step of predicting a malfunction of the entity by means of an artificial intelligence system comprises receiving a data group referred to the entity at the input of the artificial intelligence system and calculating a plurality of probability of malfunction of the entity”),
and further comprising learning the values of the prediction model using the training data obtained by the method of generating training data (Perboli discloses [0079-0080] “The training data associated with the selected training entity subset is used as the first subset of training data DT1 (block 108) and is used to train the AI artificial intelligence algorithm (block 109). The AI artificial intelligence algorithm thus trained is subjected to evaluation (decision block 110)”
[0081] “In case the validation detects the production of sub-optimal results (output branch N of block 110), the hyperparameters of the artificial intelligence algorithm are adjusted, if necessary,”).
Regarding Claim 7, the combined method of Perboli / Mugan / Yang teaches:
A computer program including software instructions which, when executed by a programmable device, implement a method of generating training data according to claim 1 (Perboli discloses [0051] “malfunction prediction system based on artificial intelligence according to an embodiment of the present invention - referred to as 'system 1' in the following, for the sake of brevity - comprises an interface module 11,”
[0053] “the interface module 11 comprises necessary hardware resources, such as one or more processors, volatile memories, non-volatile memories, modems, etc.”
A computer (‘programmable device’) is implied by using artificial intelligence and processors to predict malfunctions, which further implies a computer program storing executable software instructions.).
Regarding Claim 8, the combined method of Perboli / Mugan / Yang teaches:
A computer program including software instructions which, when executed by a programmable device, implement a machine learning method according to claim 6 (Perboli discloses [0051] “malfunction prediction system based on artificial intelligence according to an embodiment of the present invention - referred to as 'system 1' in the following, for the sake of brevity - comprises an interface module 11,”
[0053] “the interface module 11 comprises necessary hardware resources, such as one or more processors, volatile memories, non-volatile memories, modems, etc.”
A computer (‘programmable device’) is implied by using artificial intelligence and processors to predict malfunctions, which further implies a computer program storing executable software instructions.).
Regarding Claim 9, the rejection of claim 1 is incorporated. The difference in scope being:
A device for generating training data (Perboli discloses [0016] “a method and system for predicting a malfunction capable of automatically creating an optimal set of training data”
([0051] “malfunction prediction system based on artificial intelligence … comprises an interface module 11,”
A computer (‘device’) is implied by using artificial intelligence to predict malfunctions.),
the device including at least one calculation processor configured to implement ([0053] “the interface module 11 comprises necessary hardware resources, such as one or more processors.”).
Regarding Claim 11, the rejection of claim 1 is incorporated. The difference in scope being:
A device for automatic learning of a model for predicting a risk of occurrence of a feared rare event (Perboli discloses [0043] “the step of predicting a malfunction of the entity by means of an artificial intelligence system comprises receiving a data group referred to the entity at the input of the artificial intelligence system and calculating a plurality of probability of malfunction of the entity”
A computer (‘device’) is implied by using an artificial intelligence system to predict a probability of a malfunction.),
including a device for generating learning data ([0016] “a method and system for predicting a malfunction capable of automatically creating an optimal set of training data”
A computer (‘device’) is implied by creating an optimal set of training data for predicting malfunctions.),
the device including at least one calculation processor configured to implement ([0053] “the interface module 11 comprises necessary hardware resources, such as one or more processors.”):
and further comprising a module for learning the values of the prediction model implementing the training data obtained by said device for generating training data (Perboli discloses [0079-0080] “The training data associated with the selected training entity subset is used as the first subset of training data DT1 (block 108) and is used to train the AI artificial intelligence algorithm (block 109). The AI artificial intelligence algorithm thus trained is subjected to evaluation (decision block 110)”
[0081] “In case the validation detects the production of sub-optimal results (output branch N of block 110), the hyperparameters of the artificial intelligence algorithm are adjusted, if necessary,”).
The following are the references relied upon in the rejections below:
Mo, Jianwen, Yuwan Gan, and Hua Yuan. "Weighted pseudo labeled data and mutual learning for semi-supervised classification." IEEE Access 9 (2021): 36522-36534.
Claims 3-4 are rejected under 35 U.S.C. 103 as being unpatentable over Perboli / Mugan / Yang / Mo.
Regarding Claim 3, the combined method of Perboli / Mugan / Yang teaches:
The method according to claim 2, wherein step E) includes the application of a plurality of second classifier models: … application of a supervised learning classifier to the training data of the first group and the training data of the second group (Yang discloses “we consider using a supervised task model trained on the original training set, to re-label a candidate pool of synthetic answers and distractors. This is similar to treating the synthetic questions as unlabeled data and applying self-training. The utility of this self-training can be task-dependent;” (P. 1010, Sec. 2.1, Last Paragraph)
(P. 1009, Sec. 2.1, ¶1) “will use multiple choice question answering as a running example to describe synthetic data generation … and a ground truth label yi ∈ {1,2,...,K}.”
A task model is trained by supervised learning and then performs self-supervised (unsupervised) learning to relabel generated synthetic training data. Each synthetic data sample has a label from a set of ground truth labels yi ∈ {1,2,...,K}, therefore each ground truth label is a group (y1 is first group, y2 is second group, yK is third group). Synthetic data samples labeled with y1 (‘training data of the first group’) and samples labeled with y2 (‘training data of the second group’) are then relabeled with another ground truth label, therefore reclassifying synthetic data samples of a first group y1 and second group y2.)
and application of an unsupervised learning classifier to the training data of the second group and the training data of the third group (Yang discloses “A related question is whether G-DAUGc’s optional self-supervised relabeling improves performance.” (P. 1015, Sec. 5.2, ¶1)
A task model is trained by supervised learning and then performs self-supervised (unsupervised) learning to relabel generated synthetic training data (see P. 1010, Sec. 2.1, Last Paragraph). Synthetic training data samples labeled with yK (‘training data of the third group’) and samples labeled with y2 (‘training data of the second group’) are then relabeled with another ground truth label, therefore reclassifying synthetic data samples of a third group yK and second group y2.).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined method of Perboli / Mugan / Yang with the relabeling technique disclosed by Yang to relabel augmented training data. By relabeling augmented training data, an augmented sample can be relabeled to ensure that its label is correct, thereby ensuring that a machine learning model is only trained with correct labels thus improving classification accuracy.
However, the combination does not teach using a semi-supervised learning classifier to reclassify training data, which is taught by Mo:
application of a semi-supervised learning classifier to the training data of the first group and the training data of the third group ((P. 36522, Abstract) “The purpose of our method is to improve the classification performance of semi-supervised learning models and rectify the incorrect pseudo labels in the process of training. … The pseudo labels of unlabeled data are inferred by network predictions.”
(P. 36526, Sec. III-B, Last Paragraph) “The bootstrap sampling principle is employed to remove the new pseudo labeled samples from the labeled training set and add them to the unlabeled dataset after each training iteration. Especially, employing the bootstrap sampling principle gives incorrect pseudo labeled samples a chance to be rectified and relabeled in the next iteration. Afterwards, the updated model is employed to relabel the samples in the unlabeled dataset and obtain new pseudo labeled samples, which are combined with real labeled samples to retrain the model.”
(P. 36524, Sec. II-A, ¶1) “Before each iteration of training, three different labeled training sets
X
(
i
)
are generated … Each training set consists of two parts, namely,
X
(
i
)
=
D
a
(
i
)
∪
D
b
(
i
)
, where
D
a
(
i
)
denotes the initial real labeled data, and
D
b
(
i
)
denotes the new pseudo labeled data.”
A semi-supervised learning model is trained to relabel (rectify) newly generated pseudo labeled samples from three different labeled training sets (first, second, and third sets (groups) of training data).),
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined method of Perboli / Mugan / Yang with the relabeling technique disclosed by Mo to relabel augmented training data by using a semi-supervised model. By relabeling augmented training data by using a semi-supervised model, the semi-supervised model can use ground truth to ensure an augmented sample’s label is correct, thereby ensuring that a model is only trained with correct labels thus improving classification accuracy.
Regarding Claim 4, the combined method of Perboli / Mugan / Yang teaches: The method according to claim 2, however the combination does not teach iterating steps A) to D) when a reclassification error probability is higher than an error threshold, which is taught by Mo:
wherein if said reclassification error probability is higher than an error threshold, steps A) to D) are iterated ((P. 36525, Sec. III-B, ¶3) “According to information theory, information entropy can reflect the uncertainty of the pseudo labels. As the entropy of a pseudo label increases, the uncertainty increases. Therefore, weights are assigned to the pseudo labeled samples as follows: … where the numerator denotes the information entropy of pseudo label … The weights are considered an evaluation criterion to predict the confidence of the samples. … if the confidence is above the preset threshold, the sample will be considered a high confidence sample … if the confidence is below the preset threshold, the sample will be considered a low-confidence sample, added back to unlabeled dataset DU and relabeled in the next iteration.”
In each iteration, confidence is used to determine if pseudo labeled samples are incorrectly labeled so that they can be relabeled in a subsequent iteration, therefore confidence is the probability of a sample being correctly labeled and confidence is therefore reclassification error probability.
The Examiner interprets Step A according to its BRI to encompass feature extraction.
Mo discloses Figure 1 on P. 36524 (reproduced below) depicting feature extraction of an unlabeled dataset (incorrect pseudo labeled samples). Incorrect pseudo labeled samples have their features extracted so that their labels can be corrected in a subsequent iteration, therefore feature extraction is iterated.
PNG
media_image1.png
494
1221
media_image1.png
Greyscale
The Examiner interprets Step B according to its BRI to encompass generating pseudo-labeled samples.
(P. 36524, Sec. II-A, ¶1) “Before each iteration of training, three different labeled training sets
X
(
i
)
are generated … Each training set consists of two parts, namely,
X
(
i
)
=
D
a
(
i
)
∪
D
b
(
i
)
, where
D
a
(
i
)
denotes the initial real labeled data, and
D
b
(
i
)
denotes the new pseudo labeled data.”
The Examiner interprets Step C according to its BRI to encompass generating an augmented training data set.
(P. 36526, Sec. III-B, Last Paragraph) “The pseudo labels of unlabeled data are inferred by network predictions. Second, based on selection and weighting strategies for pseudo labeled data, pseudo labeled data with high confidence are selected and added to the real labeled training set.”
Pseudo labeled data that is determined to be correctly labeled based on confidence are added to a real labeled training set (augmented data set).
The Examiner interprets Step D according to its BRI to encompass classifying augmented training data.
(P. 36526, Sec. III-B, Last Paragraph) “The bootstrap sampling principle is employed to remove the new pseudo labeled samples from the labeled training set and add them to the unlabeled dataset after each training iteration. Especially, employing the bootstrap sampling principle gives incorrect pseudo labeled samples a chance to be rectified and relabeled in the next iteration.”
Therefore, feature extraction (step A), generating pseudo-labeled samples (step B), generating an augmented data set (step C), and classifying augmented training data (step D) are iterated based on predicted confidence (‘reclassification error probability’).),
said first group of training data forming the first initial set of data and said second group of training data forming said second initial set of data for a subsequent iteration (Pseudo labeled data that is determined to be correctly labeled based on a high confidence are added to a real labeled training set (‘first initial set of data’), see (P. 36526, Sec. III-B, Last Paragraph). Pseudo labeled data that is determined to be incorrectly labeled is added to an unlabeled dataset (‘second initial set of data’), see(P. 36526, Sec. III-B, Last Paragraph).).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to modify the combined method of Perboli / Mugan / Yang with the relabeling technique disclosed by Mo to iteratively relabel augmented training data. By iteratively relabeling augmented training data, samples with incorrect labels can be corrected in subsequent iterations, thereby ensuring that generated training data contains accurate labels to perform training with.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Ben-Itzhak et al. (US 20220012625 A1) teaches classifying an unlabeled training dataset for anomaly detection by using anomaly score thresholds to create labels.
Taheri et al. (“On defending against label flipping attacks on malware detection systems.”) teaches relabeling training samples by using K-nearest neighbors to generate training data for detecting malware.
Kong et al. (CN 113807455 A) teaches grouping augmented training samples into groups and then calculating a probability distribution for each group to determine the probability of each sample in a group belonging to a pre-determined class. The difference between probability distributions is then used to update an unsupervised clustering model.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PEDRO J MORALES whose telephone number is (571)272-6106. The examiner can normally be reached 8:30 AM - 6:00 PM.
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, MIRANDA M HUANG can be reached at (571)270-7092. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/PEDRO J MORALES/Examiner, Art Unit 2124
/MIRANDA M HUANG/Supervisory Patent Examiner, Art Unit 2124
Prosecution Insights