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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 12/10/2025, 2/5/2026, 4/9/2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Response to Arguments
Regarding the 101 Rejection
The applicant argues (page 12):
PNG
media_image1.png
522
751
media_image1.png
Greyscale
Examiner’s Response:
In response, The Examiner respectfully disagrees that the claims are eligible at Step 2A, Prong One and maintains the rejection. The examiner acknowledges that the high-level act of “training” a machine learning model may not constitute a mental process. However, the amended claims do not simply recite “training an instance of a tree based inference model using training data” and stop there. Rather, the claim goes on to define how that training is accomplished, through a splitting rule that partitions training data, a function that rewards predictability of labels and discourages predictability of bias labels, the function assigning a numerical value based on division of records, and optimization of that function. These specific limitations, which define the substance of the claimed training, recite mathematical concepts (e.g., computing and optimizing predictive functions, assigning numerical values, optimization using mathematical algorithms as supported by the instant specification at paragraph [0061]) and mental processes (e.g., partitioning data using a splitting rule, subsuming nodes to limit the total number of nodes). The abstract idea is not the training limitation itself; it is the particular steps the claims recite to carry out the training, and those steps fall within the mathematical concepts and mental processes groupings. As set forth in the updated rejection, the high-level recitation of "training an instance of a tree based inference model using training data" is treated as an additional element under Step 2A, Prong Two, where it amounts to no more than a mere instruction to implement the abstract idea (MPEP 2106.05(f)). Accordingly, the eligibility analysis does not end at Step 2A, Prong One.
With respect to Example 39 as the applicant stated, the Examiner notes that those examples involve claims where training was recited with more specificity and was tied to a particular technological improvement in how the model itself operates. In the instant application, the specific manner of training recited in the claims is defined by the mathematical relationships and mental processes that constitute the identified abstract idea. The mere addition of the word “training” as a wrapper around abstract mathematical and mental operations does not transform those operations into something other than an abstract idea.
The applicant argues (page 12-13):"Applicant respectfully reminds the Office that the Director of the USPTO (John A. Squires) issued a memorandum ('memo') on December 4, 2025, titled 'Subject Matter Eligibility Declarations' in view of a decision issued by Director Squires in Appeal No. 2024-000567 titled In re Desjardins ('the Desjardins memo'). … In the Desjardins memo, Director Squires emphasized that: (i) the Appeals Review Panel (ARP) 'warned against overbroad Section 101 rejections because '[c]ategorically excluding AI innovations from patent protection in the United States jeopardized America's leadership in [] critical emerging technolog[ies].'; and (ii) claims directed to improvements to the functions/functionings of machine learning models are patent eligible. … Finally, and most importantly, this decision (i.e., the Desjardins decision as referred to in the Desjardins memo) has been designated by Director Squires as precedential."
Examiner’s Response:In response, the Examiner is aware of the “In re Desjardins ('the Desjardins memo')” and the Director’s memoranda, and the Examiner does not categorically exclude AI innovations from patent eligibility. The present rejection is not an overboard categorical exclusion of AI-related claims. Rather it is a claim-specific analysis under the subject matter eligibility test as set forth in the MPEP section 2106. Which is applied to the particular limitations recited in this application. The existence of the Desjardins as precedent does not mean that every claim reciting machine learning training is automatically patent eligible. The Desjardins decision outlines that claims directed to specific improvements to how a machine learning model itself operates can be patent eligible. Each claim must be evaluated on its own merits based on the specific limitations recited.
Applicant’s Argument (Page 13-14):
"Applicant submits that the pending claims (like the Desjardin claims) are also patent eligible because they are directed to improvements to the functioning of a machine learning model. In particular, paragraphs [0061]-[0067] of the Original Specification describe a training process preformed to reduce latent bias in inference models thereby improving the quality of inferences provided by the inference models and 'address, among others, the technical problem of latent bias exhibited by inference models.' Original Specification at para. [0016]. Thus, it should be clear that the Original Specification describes a detailed technical solution for reducing latent bias inference models, which directly translates to an improvement in the functioning of the inference models for generating improved and better inferences."
Examiner's Response:
In response, the Examiner respectfully disagrees that the pending claims are analogous to the claims in Desjardins. In Desjardins, the claims were found eligible because they addressed a specific technical problem in how the machine learning model itself operates. Which was catastrophic forgetting in continual learning systems. And the claims recited specific technical steps that changed the model's architecture or learning mechanism to protect prior knowledge while acquiring new knowledge. That improvement was to the model's operational functionality itself, not merely to the quality of its output.In the instant application, the claimed “improvement” which is reducing latent bias relates to improving the quality of the inferences (i.e., the output) produced by the model. While the Examiner does not dispute that reducing bias is a desirable goal, the mechanism by which the claims achieve this goal is through mathematical operations: a splitting rule defined as a function that assigns numerical values, rewards and discourages predictability metrics, and is optimized using mathematical algorithms (specification at paragraph [0061] describes gradient descent and evolution-based algorithms). The claims recite the mathematical formulation of a bias-aware objective function and its optimization. These are the limitations identified as abstract ideas. Reducing output bias through a modified mathematical objective function is not the same type of improvement as Desjardins, where the model's ability to learn without catastrophic failure was structurally improved.
Applicant’s argues (Page 14):
The above-referenced steps and processes discussed in paragraphs [0061]-[0067] of the Original Specification are reflected in amended independent claim 1 (reproduced in part below) and result in the above-discussed improvements to the functioning of the inference models (i.e., the reduction and/or prevention of latent bias to generate better quality, bias-less/bias-reduced inferences).
Original independent claims 8 and 15 recite substantially similar limitations. As further shown above, the independent claims have been amended to highlight the prevention of the latent bias in the inferences generated by the claimed inference model. Thus, there should be no question that both the original and amended independent claims (like the claims in Desjardins) reflect the improvements (namely, the prevention/reduction of latent bias in inference models) described in the specification. See revised MPEP § 2106.04(d)(I), which now reads: (…) Second, if the specification sets forth an improvement in technology or a technical field, the claim must be evaluated to ensure that the claim itself reflects the disclosed improvement, i.e., that the claim includes the components or steps of the invention that provide the improvement described in the specification. The claim itself does not need to explicitly recite the improvement described in the specification (e.g., "thereby increasing the bandwidth of the channel") See, e.g., Ex Parte Desjardins, Appeal No. 2024- 000567 (PTAB September 26, 2025, Appeals Review Panel Decision) (precedential), inAppl. Seri
which the specification identified the improvement to machine learning technology by explaining how the machine learning model is trained to learn new tasks while protecting knowledge about 3 previous tasks to overcome the problem of "catastrophic forgetting,"
Examiner’s Response:
As discussed above, the specification describes reducing latent bias in the output of inference models. While this is beneficial from a fairness and accuracy standpoint, the Examiner does not find that the specification articulates an improvement to the technology itself. The specification does not describe, for example, reduced computational requirements, lower memory usage, faster processing, improved model convergence, prevention of a technical failure mode (like catastrophic forgetting in Desjardins), or any change to the computer's or model's underlying technical capabilities. Rather, the specification describes a mathematical training procedure that produces a model less likely to exhibit bias. This is an improvement in the quality of the result, not an improvement in how the underlying technology functions. Therefore, even under the revised MPEP guidance, the claims do not integrate the abstract idea into a practical application.
Applicant’s Argument (Page 15):
Thus, for at least the above reasons, regardless of whether the original and amended independent claims are directed to an allegedly abstract idea, the amended independent claims are, like the claims in Enfish, McRo, and newly decided Ex Parte Desjardins, directed to "improvements to technology or computer functionality" and integrate any allegedly recited judicial exception into a practical application. See MPEP § 2106.04(d) (Emphasis Added). Thus, like the claims in Desjardins, the pending independent claims are also eligible at Step 2A, prong two of the Office's Subject Matter Eligibility Test.
Examiner’s Response:
In response, the Examiner respectfully finds the comparison to Enfish, McRo, and Desjardins unpersuasive. The claims in these cases were directed to a concrete change/improvement to computer functionality. For example, in Desjardins, the claims addressed catastrophic forgetting, a specific technical failure mode in how the model learns which is through structural changes to the training regime. The present claims do not recite such analogous improvement. The claims recite mathematical operations (a function that assigns numerical values, optimization of that function) wrapped in the context of training a decision tree model. The "improvement" argued by Applicant which is that the resulting model will exhibit less latent bias is a desired outcome or intended result of performing the recited mathematical operations, not a technical improvement to the model's underlying architecture or to computer functionality. Accordingly, the claims do not integrate the judicial exception into a practical application under step 2A, Prong Two, and the rejection under 35 U.S.C. 101 is maintained for all pending claims.
Regarding the 103 Rejection
Examiner’s Response:
In response to the amended claims regarding the 103 rejections for claims 1, 2, 4, 5, 8, 9, 11, 12, 15, 16, 18, 19. The examiner acknowledges the amendments overcome the prior art. New grounds of rejection has been made in light of Kamiran in view Gauci further in view of Ignatov. Applicant’s arguments are moot in view of new grounds of rejection.
In response to the new claims 21-28, Applicant’s arguments are moot in view of new grounds for rejection. Accordingly, claims 21-28 are rejected in light of Kamiran in view of Gauci further in view of Ignatov.
Amendments to the SpecificationIn response to the amendments made regarding the specification objections presented in the previous office action, the examiner withdraws these objections.Claim Objections
In response to the amendments made regarding the claim objections presented in the previous office
action, the examiner withdraws the claim objections.
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, 2, 4, 5, 8, 9, 11, 12, 15, 16, 18, 19, 21-28 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1
Step 1: Claim 1 is directly to a method, corresponding to a process, which is one of the statutory categories.
Step 2A Prong 1: The claim recites:
A method for (This step is practically implementable in the human mind and is understood to be a recitation of a mental process. As shown in FIG. 2B, a person with a pen and paper can partition the training data using a splitting rule and, using their judgement to manually assign the records to each node so as to favor (incentivize) the labels and disfavor (disincentivize) the bias labels.)
subsuming nodes to limit a total number of nodes of the tree based inference model, (This step is practically implementable in the human mind and is understood to be a recitation of a mental process. Subsuming nodes to limit the total number of nodes amounts to combing or merging groups of nodes. A person can mentally or with the aid of a pen and paper, combine or merge groups of nodes of the drawn tree (FIG. 2B) using their judgement to reduce the number of nodes.)
wherein the splitting rule partitions records of the training data into two groups using a function that rewards predictability of the labels by the instance of the tree based inference model and discourages predictability of the bias labels by the instance of the tree based inference model, (This step describes performing mathematical calculations (e.g., computing and optimizing predictive functions to partition data). Therefore, this limitation recites a mathematical concept, which falls within the abstract idea group of mathematical relationships.; or a mental process)
wherein the function assigns a numerical value based on division of the records among the two groups, and the records are partitioned through optimization of the function, (This step describes performing mathematical calculations including a function that assigns numerical values to data. Additionally, the optimization step is also represented as an abstract idea of a mathematical calculation, as it is defined as an optimization technique that uses a mathematical algorithm, as supported by the instant application in paragraph 0061.)
identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services; (This step for identifying an occurrence of a condition that indicates an inference is necessary is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can identify whether an inference is needed by simply observing a condition and making a mental judgement.)
based on the occurrence: (This step is practically implementable in the human mind and is understood to be a recitation of a mental process. A person can identify an occurrence and take subsequent actions based on the identified occurrence.)
obtaining the tree based inference model from among the inference models (This step for obtaining a tree based inference model is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. The tree-based inference model is shown in FIG. 2B, a person with a pen and paper can draw this model given a list of training data.)
obtaining the inference using the tree based inference model; and (This step for obtaining an inference using an inference model is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process. The tree-based inference model is shown in FIG. 2B, a person can mentally go through each step of the tree until they reach the terminal node to obtain the inference.)
Step 2A prong 2: This judicial exception is not integrated into a practical application. The claim further recites:
training an instance of a tree based inference model using training data, (This additional element recites a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the abstract idea. Examiner’s note – high level recitation of training a model with data)
…for providing computer implemented services using inference models, and …providing computer implemented services using the inference. (This additional element recites a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the abstract idea. The additional elements amount to nothing more than an instruction to apply the abstract idea using a generic computer. The claim doesn’t specify how the computer is improved.)
Step 2B:
training an instance of a tree based inference model using training data, (This additional element recites a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the abstract idea. Examiner’s note – high level recitation of training a model with data)
…for providing computer implemented services using inference models, and …providing computer implemented services using the inference. (This additional element recites a mere instruction to apply an exception with a recitation of the words "apply it" (or an equivalent) as identified in MPEP 2106.05(f), and does not provide integration into a practical application or significantly more than the abstract idea. The additional elements amount to nothing more than an instruction to apply the abstract idea using a generic computer. The claim doesn’t specify how the computer is improved.)
Regarding claim 2:
Step 1: A method, as above.
Step 2A prong 1: See the rejection of Claim 1 above, which claim 2 depends on.
Step 2A Prong 2:
wherein obtaining the tree based inference model comprises: reading the tree based inference model from storage. (Adding insignificant extra-solution activity to the judicial exception — see MPEP 2106.05(g) This step of reading the inference model from storage amounts to mere data gathering which is known to be insignificant extra solution activity.)
Step 2B:
wherein obtaining the inference model comprises: reading the inference model from storage. ((MPEP 2106.05(d)(II) indicates that merely “Storing and retrieving information in memory” is a well-understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed limitation is well-understood, routine, conventional activity is supported under Berkheimer.)
For the reasons above, claim 2 is rejected as being directed to an abstract idea without significantly more.
Regarding claim 4:
Step 1: A method, as above.
Step 2A prong 1: See the rejection of Claim 1 above, which claim 4 depends on. The claim further recites:
wherein each of the records comprises:
at least one feature value;
at least one label value associated with the at least one feature value;
and at least one bias label value associated with the at least one feature value. (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can gather records like rows on a worksheet. For each of those records they can identify a feature value (example, column 1: “Income”). Identify a label value (example, column 2: “Loan Approved?”). Identify a bias feature value. (example, column 3: “Gender”).
Step 2A prong 2: The claim does not recite additional elements therefore the judicial exception is not integrated into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception. Therefore, the claim is not patent eligible.
Regarding claim 5:
Step 1: A method, as above
Step 2A prong 1: See the rejection of Claim 4 above, which claim 5 depends on. The claim further recites:
wherein training the instance of the tree based inference model comprises: (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment}. Training is understood to be a mental process of learning or formulating a decision-making strategy based on observation. A person mentally or using their judgement can review a collection of past files (training data) to ‘train’ themselves to create a logical framework (the tree model as depicted in Fig. 2) for how to handle similar files in the future.)
obtaining, based on the training data and the splitting rule, a root node and a question; (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person if given training data on a worksheet can follow a splitting rule to obtain a question and assign that question to a root node.)
obtaining two answer to the question that partitions the records into two groups; (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can mentally or using their judgement to come up with two answers to the question to separate into two groups.)
obtaining a second node and a third node based on the two groups; and (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can obtain a second and a third node that’s based on the two groups.)
and establishing a first edge between the root node and the second node based on a first of the two answers; and (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can establish by drawing a first edge between the root node and the second node which is based on the first of the two answers.)
establishing a second edge between the root node and the third node based on a second of the two answers. (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can establish by drawing a second edge between the root node and the third node which is based on the second of the two answers.)
Step 2A prong 2: The claim does not recite additional elements therefore the judicial exception is not integrated into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception. Therefore, the claim is not patent eligible.
Regarding claim 8:
Claim 8 is a non-transitory computer-readable storage medium claim that recites identical imitations to method claim 1. Therefore, claim 8 is rejected using the same rationale as claim 1. Regarding claim 9:
Claim 9 is a non-transitory computer-readable storage medium claim that recites identical imitations to method claim 2. Therefore, claim 9 is rejected using the same rationale as claim 2.Regarding claim 11:
Claim 11 is a non-transitory computer-readable storage medium claim that recites identical imitations to method claim 4. Therefore, claim 8 is rejected using the same rationale as claim 4.Regarding claim 12:
Claim 12 is a non-transitory computer-readable storage medium claim that recites identical imitations to method claim 5. Therefore, claim 8 is rejected using the same rationale as claim 5.
Regarding claim 15:
Claim 15 is a system (machine) claim that recites identical imitations to method claim 1. Therefore, claim 15 is rejected using the same rationale as claim 1.
Regarding claim 16:
Claim 16 is a system (machine) claim that recites identical imitations to method claim 2. Therefore, claim 16 is rejected using the same rationale as claim 2.
Regarding claim 18:
Claim 18 is a system (machine) claim that recites identical imitations to method claim 4. Therefore, claim 18 is rejected using the same rationale as claim 4.
Regarding claim 19:
Claim 19 is a system (machine) claim that recites identical imitations to method claim 5. Therefore, claim 19 is rejected using the same rationale as claim 5.
Regarding claim 21:
Step 1: A method, as per claim 1.
Step 2A prong 1: See the rejection of Claim 1 above, which claim 21 depends on. The claim further recites:
wherein predictive power of the labels and predictive power of the bias labels are compared to a threshold. (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can perform a comparison of something against a threshold.
Step 2A prong 2: The claim does not recite additional elements therefore the judicial exception is not integrated into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception. Therefore, the claim is not patent eligible.
Regarding claim 22:
Step 1: A method, as above.
Step 2A prong 1: See the rejection of Claim 21 above, which claim 22 depends on. The claim further recites:
wherein satisfying the threshold indicates a decision tree of the tree based inference model is complete. (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can perform a comparison of something against a threshold and once that threshold is satisfied, they know the model is complete.
Step 2A prong 2: The claim does not recite additional elements therefore the judicial exception is not integrated into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception. Therefore, the claim is not patent eligible.
Regarding claim 23:
Step 1: A method, as above.
Step 2A prong 1: See the rejection of Claim 1 above, which claim 23 depends on. The claim further recites:
wherein each group provided by the tree based inference model includes an average of the labels of the corresponding group. (This step is practically implementable in the human mind with the aid of a pen and paper and is understood to be a recitation of a mental process {i.e., judgment). A person can average the labels of corresponding groups and provide them to the tree based model.
Step 2A prong 2: The claim does not recite additional elements therefore the judicial exception is not integrated into a practical application.
Step 2B: The claim does not recite additional elements that amount to significantly more than the judicial exception. Therefore, the claim is not patent eligible.
Regarding claim 24:
Claim 24 is a non-transitory machine-readable medium claim that recites identical imitations to method claim 21. Therefore, claim 24 is rejected using the same rationale as claim 21.
Regarding claim 25:
Claim 25 is a non-transitory machine-readable medium claim that recites identical imitations to method claim 22. Therefore, claim 25 is rejected using the same rationale as claim 22.
Regarding claim 26:
Claim 26 is a non-transitory machine-readable medium claim that recites identical imitations to method claim 23. Therefore, claim 26 is rejected using the same rationale as claim 23.
Regarding claim 27:
Claim 27 is a system (machine) claim that recites identical imitations to method claim 21. Therefore, claim 27 is rejected using the same rationale as claim 21.
Regarding claim 28:
Claim 28 is a system (machine) claim that recites identical imitations to method claim 23. Therefore, claim 28 is rejected using the same rationale as claim 23.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 2, 4, 5, 8, 9, 11, 12, 15, 16, 18, 19, 21-28 are rejected under 35 U.S.C. 103 as being unpatentable over non-patent literature Kamiran et al. ("Discrimination Aware Decision Tree Learning", hereinafter "Kamiran") in view of patent publication Gauci et al. (US-20170068408-A1), hereinafter "Gauci" and non-patent literature Ignatov et al. (“Decision Stream: Cultivating Deep Decision Trees”, hereinafter “Ignatov”)
Claim 1
“A method (…), the method comprising:” (section 1 Introduction, “The solution is to develop new techniques which we call discrimination aware – we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making…” (Examiners Note (EN): This denotes a method (“techniques”) for learning an inference model (“classification model”))
training an instance of a tree-based inference model using training data,
(Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making” EN: this denotes the model is trained on historical data and this trained model is used for future decision making. Therefore, the training happens “prior to” the future decision. As for “identifying the occurrence”, that is taught in claim 1 above. Additionally, the paper is titled “Discrimination Aware Decision Tree Learning” learning is synonymous to training in the BRI.)
wherein the tree based inference model is based on:
a splitting rule that partition the training data for predictive ability that incentivizes labels of the training data and that disincentivizes bias labels of the training data; and (…)
(Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”, Section 5.1: “Suppose that a certain split divides the data D into D1, . . . , Dk”, Section 4.2,“Let a labeled dataset D and a sensitive attribute B be given. The discrimination aware classification problem is to learn a classifier such that (a) The accuracy of C is high, and (b) the discrimination of C w. r. t. B is low.”, Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making”)
wherein the splitting rule partitions records of the training data into two groups using a function that rewards predictability of the labels by the instance of the tree-based inference model and discourages predictability of the bias labels by the instance of the tree-based inference model, (Algorithm 1, Step 5 and step 7-8. And Section 5.1 page 15, The IGC-IGS function. EN: this denotes the function IGC-IGS that is inserted into Step 5 of the Algorithm 1 to determine which attribute is selected. Based on that selection, Step 8 of Algorithm 1 partitions the data into groups. Additionally, in section 5.1, it denotes calculating IGC (information Gain w.r.t. Class). The information gain is the standard metric for measuring the predictability of labels (accuracy). Higher IGC represented as a higher positive term in the formulas, thus the function “rewards” predictive power for the labels. Also in section 5.1, IGS measures how well a split predicts a bias feature. By subtracting IGS from the total score (IGC – IGS), the function penalizes or discourages splits that result in high predictability of the bias feature. Therefore, the (IGC – IGS) formula denotes a function that rewards label accuracy (+IGC) and discourages bias predictability (-IGS).)
wherein the function assigns a numerical value based on division of the records among the two groups,
PNG
media_image2.png
571
943
media_image2.png
Greyscale
(Section 5.1 EN: This denotes a formula (function) Gain = IGC – IGS. The resulting “Gain” is a score which is a number maximized by the algorithm. The above screenshot is a section from 5.1 which also denotes the division of data (records). )
and the records are partitioned through optimization of the function, (Algorithm 1: Decision Tree Induction in Section 4, Line 5: “Select test att from att list and test s.t. gain(test att, test) is maximized” and Section 5.1, particularly, “Based on these two measures IGC and IGS, we introduce three alternative criteria for determining the best split:” EN: This denotes maximizing the gain function (IGC - IGS) which falls under mathematical optimization.)
“obtaining the tree based inference model from among the inference models,” (Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”)
obtaining the inference using the tree-based inference model; and (Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making” Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”)
Kamiran does not distinctly disclose:
“providing computer implemented services using inference models”,
“identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services; based on the occurrence: (…) providing computer implemented services using the inference.”
However, Gauci teaches:
“for providing computer implemented services using inference models”,
(Gauci, Figure 1 block 106 which states “perform action associated with application”. EN: This denotes opening an application for the user from the outcome of the prediction model. The opening of an application is the computer implemented services and the prediction model is the inference.) Additionally, to make the record clear, a prediction model in the broadest reasonable interpretation includes inference models. See paragraph 27 of Gauci, which states “Examples of prediction models include neural networks, decision trees, multi-label logistic regression, and combinations thereof.”)
“identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services;” (Figure 1 and paragraph 20-22, EN: this denotes detecting the triggering event which is the occurrence of a condition. This detection indicates that the system must now use a “prediction model” which is the inference model used to suggest the correct app to the user (the opening of the suggested app is the computer implemented service).
based on the occurrence: (…) (Gauci, Paragraph 97, “At block 506, the device selects a prediction model… The selected prediction model may depend on the triggering event.” )
providing computer implemented services using the inference. (Gauci, Figure 1 block 106 which states “perform action associated with application”. EN: This denotes opening an application for the user from the outcome of the prediction model. The opening of an application is the computer implemented services and the prediction model is the inference.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran with the event-triggered system that selects prediction models to provide computer-implemented services of Gauci. Combining the teachings of Kamiran and Gauci allows the theoretical fairness benefits of Kamiran to be realized in a scalable, and automated manner, effectively upgrading a static research model into a functioning, real-time computer service.
Kamiran in view of Gauci does not explicitly teach:subsuming nodes to limit a total number of nodes of the tree-based inference model,
However, Ignatov teaches:
subsuming nodes to limit a total number of nodes of the tree-based inference model,
(Page 1 Abstract, “we propose merging nodes from different branches based on their similarity that is estimated with two sample test statistics, which leads to generation of a deep directed acyclic graph of decision rules” Page 3, “In this section, we describe the proposed Decision Stream algorithm. The main concept of this method consists in merging similar nodes after each splitting iteration […] fuses statistically similar nodes (Fig. 2(b-c)) using an input parameter—significance threshold Plim” – EN: this denotes merging, analogous to subsuming leaf nodes after each splitting iteration based on statistical similarity, which reduces the total number of nodes in the model structure as shown in Figure 5 of Ignatov.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran and the event-triggered system that selects prediction models to provide computer-implemented services of Gauci with the node merging technique of Ignatov. The motivation for doing so would be to reduce model complexity and prevent by addressing the issue of data exhaustion in decision tree leaf nodes, thereby improving prediction accuracy. Ignatov Abstract, “Tree node splitting based on relevant feature selection is a key step of decision tree learning, at the same time being their major shortcoming: the recursive nodes partitioning leads to geometric reduction of data quantity in the leaf nodes, which causes an excessive model complexity and data overfitting.” Ignatov further teaches that merging nodes directly addresses this problem, as stated in section 5 (conclusion), “a Decision Stream, which avoids the problems of data exhaustion and formation of unrepresentative data samples in decision tree nodes by merging the leaves from the same and/or different levels of the predictive model structure.”
Claim 2
Gauci further teaches:
“The method of claim 1, wherein obtaining the tree based inference model comprises: reading the tree based inference model from storage” (Paragraph 73, “The prediction engine 302 may be program code stored on a memory device. In embodiments, the prediction engine 302 includes one or more prediction models.” AND paragraph 126, “One or more processors 818 run various software components stored in medium 802… [including] prediction module 830…” Para 27, “Various types of prediction models can be used. Examples of prediction models include neural networks, decision trees, multi-label logistic regression, and combinations thereof.”)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method of Kamiran with the storage and retrieval of decision tree prediction models of Gauci to be able to store and retrieve the models, thereby enabling the device to immediately provide the suggested service upon detection of a trigger, which improves the speed and streamlines the operations. Gauci paragraph 22, “A user interface can be provided to a user in an opportune manner or in an opportune screen, which can save time and streamline device operation.”
Claim 4
Kamiran further teaches:
The method of claim 1, wherein each of the records comprises: at least one feature value; (Section 4.1,
PNG
media_image3.png
289
382
media_image3.png
Greyscale
(table 1 – example of training data with feature values)
at least one label value associated with the at least one feature value; (Section 4.1 “a labeled dataset is a finite set of tuples over the schema (A1, . . . , An, Class). dom(Class) = {+, −}.” EN: this denotes the outcome or class the model tries to predict. This denotes the training data having a class label.)
and at least one bias label value associated with the at least one feature value. (Section 1, “a labeled dataset D is given, and one Boolean discriminatory attribute B (e.g., gender) is specified.” This denotes the training data having a sensitive (bias) attribute. In Kamiran, B’s values in each record serve as the target labels for the IGS computation (Section 5.1), and therefore ach record’s value for B constitutes a “bias label value” under BRI, it is the value used as a label for measuring and penalizing discrimination.)
Claim 5
Kamiran further teaches:The method of claim 4, wherein training the instance of the tree based inference model comprises: obtaining, based on the training data and the splitting rule, a root node and a question; (Algorithm 1: Decision Tree Induction in Section 4, “1: Create a node N” and “5: Select test att from att list and test s.t. gain(test att, test) is maximized” (EN: This denotes Algorithm 1 which creates a node (root node) in line 1 and selects a test attribute (question) based on maximizing a gain calculation (splitting criterion) in line 5.) obtaining two answer to the question that partitions the records into two groups;
PNG
media_image4.png
457
696
media_image4.png
Greyscale
(Figure 3 - Decision tree in Section 5.2 and Algorithm 1: Decision Tree Induction in Section 4, line 7-9 (EN: this denotes creation of branches and the partitioning of the data into subgroups based on the answers to the test question.) obtaining a second node and a third node based on the two groups;
PNG
media_image5.png
457
696
media_image5.png
Greyscale
(Figure 3 - Decision tree in Section 5.2 and Algorithm 1: Decision Tree Induction in Section 4, line 13. (EN: this denotes for each new group of data, the algorithm recursively runs again, creating new nodes, (second, third, etc. that’s attached to the root node.) and establishing a first edge between the root node and the second node based on a first of the two answers; and establishing a second edge between the root node and the third node based on a second of the two answers.
PNG
media_image6.png
457
696
media_image6.png
Greyscale
(Figure 3 - Decision tree in Section 5.2 and Algorithm 1: Decision Tree Induction in Section 4, line 8. (EN: this denotes the edges are the branches that connect the nodes together, representing the answers to the question.)
Claim 8
Kamiran teaches:
(…) training an instance of a tree-based inference model using training data,
(Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making” EN: this denotes the model is trained on historical data and this trained model is used for future decision making. Therefore, the training happens “prior to” the future decision. As for “identifying the occurrence”, that is taught in claim 1 above. Additionally, the paper is titled “Discrimination Aware Decision Tree Learning” learning is synonymous to training in the BRI.)
wherein the tree based inference model is based on:
a splitting rule that partition the training data for predictive ability that incentivizes labels of the training data and that disincentivizes bias labels of the training data; and
(Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”, Section 5.1: “Suppose that a certain split divides the data D into D1, . . . , Dk”, Section 4.2,“Let a labeled dataset D and a sensitive attribute B be given. The discrimination aware classification problem is to learn a classifier such that (a) The accuracy of C is high, and (b) the discrimination of C w. r. t. B is low.”, Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making”)
wherein the splitting rule partitions records of the training data into two groups using a function that rewards predictability of the labels by the instance of the tree-based inference model and discourages predictability of the bias labels by the instance of the tree-based inference model, (Algorithm 1, Step 5 and step 7-8. And Section 5.1 page 15, The IGC-IGS function. EN: this denotes the function IGC-IGS that is inserted into Step 5 of the Algorithm 1 to determine which attribute is selected. Based on that selection, Step 8 of Algorithm 1 partitions the data into groups. Additionally, in section 5.1, it denotes calculating IGC (information Gain w.r.t. Class). The information gain is the standard metric for measuring the predictability of labels (accuracy). Higher IGC represented as a higher positive term in the formulas, thus the function “rewards” predictive power for the labels. Also in section 5.1, IGS measures how well a split predicts a bias feature. By subtracting IGS from the total score (IGC – IGS), the function penalizes or discourages splits that result in high predictability of the bias feature. Therefore, the (IGC – IGS) formula denotes a function that rewards label accuracy (+IGC) and discourages bias predictability (-IGS).)
wherein the function assigns a numerical value based on division of the records among the two groups,
PNG
media_image2.png
571
943
media_image2.png
Greyscale
(Section 5.1 EN: This denotes a formula (function) Gain = IGC – IGS. The resulting “Gain” is a score which is a number maximized by the algorithm. The above screenshot is a section from 5.1 which also denotes the division of data (records). ) and the records are partitioned through optimization of the function, (Algorithm 1: Decision Tree Induction in Section 4, Line 5: “Select test att from att list and test s.t. gain(test att, test) is maximized” and Section 5.1, particularly, “Based on these two measures IGC and IGS, we introduce three alternative criteria for determining the best split:” EN: This denotes maximizing the gain function (IGC - IGS) which falls under mathematical optimization.)
“obtaining the tree based inference model from among the inference models,” (Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”)
obtaining the inference using the tree-based inference model; and (Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making” Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”))
Kamiran does not distinctly disclose:
“A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations for providing computer implemented services using inference models, the operations comprising:” “identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services; based on the occurrence:”
“providing computer implemented services using the inference.”
However, Gauci teaches:
A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations for providing computer implemented services using inference models, the operations comprising: (Paragraph 126, “One or more processors 818 run various software components stored in medium 802 to perform various functions for device 800. In some embodiments, the software components include… a prediction module 830…”)
“identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services;” (Figure 1 and paragraph 20-22, EN: this denotes detecting the triggering event which is the occurrence of a condition. This detection indicates that the system must now use a “prediction model” which is the inference model used to suggest the correct app to the user (the opening of the suggested app is the computer implemented service).
based on the occurrence: (…) (Gauci, Paragraph 97, “At block 506, the device selects a prediction model… The selected prediction model may depend on the triggering event.” )
providing computer implemented services using the inference. (Gauci, Figure 1 block 106 which states “perform action associated with application”. EN: This denotes opening an application for the user from the outcome of the prediction model. The opening of an application is the computer implemented services and the prediction model is the inference.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran with the event-triggered system that selects prediction models to provide computer-implemented services that’s part of an over encompassing machine-readable medium of Gauci. Combining the teachings of Kamiran and Gauci allows the theoretical fairness benefits of Kamiran to be realized in a scalable, and automated manner, that utilizes machine, and thus effectively upgrading a static research model into a functioning, real-time computer service.
Kamiran in view of Gauci does not explicitly teach:subsuming nodes to limit a total number of nodes of the tree-based inference model,
However, Ignatov teaches:
subsuming nodes to limit a total number of nodes of the tree-based inference model,
(Page 1 Abstract, “we propose merging nodes from different branches based on their similarity that is estimated with two sample test statistics, which leads to generation of a deep directed acyclic graph of decision rules” Page 3, “In this section, we describe the proposed Decision Stream algorithm. The main concept of this method consists in merging similar nodes after each splitting iteration […] fuses statistically similar nodes (Fig. 2(b-c)) using an input parameter—significance threshold Plim” – EN: this denotes merging, analogous to subsuming leaf nodes after each splitting iteration based on statistical similarity, which reduces the total number of nodes in the model structure as shown in Figure 5.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran with the event-triggered system that selects prediction models to provide computer-implemented services that’s part of an over encompassing machine-readable medium of Gauci with the node merging technique of Ignatov. The motivation for doing so would be to reduce model complexity and prevent by addressing the issue of data exhaustion in decision tree leaf nodes, thereby improving prediction accuracy. Ignatov Abstract, “Tree node splitting based on relevant feature selection is a key step of decision tree learning, at the same time being their major shortcoming: the recursive nodes partitioning leads to geometric reduction of data quantity in the leaf nodes, which causes an excessive model complexity and data overfitting.” Ignatov further teaches that merging nodes directly addresses this problem, as stated in section 5 (conclusion), “a Decision Stream, which avoids the problems of data exhaustion and formation of unrepresentative data samples in decision tree nodes by merging the leaves from the same and/or different levels of the predictive model structure.”
Claim 9 is a non-transitory machine-readable medium claim that recite substantially the same limitation as claim 2. Therefore, claim 9 is rejected under the same rationale as claim 2.
Claim 11 is a non-transitory machine-readable medium claim that recite substantially the same limitation as claim 4. Therefore, claim 11 is rejected under the same rationale as claim 4.
Claim 12 is a non-transitory machine-readable medium claim that recite substantially the same limitation as claim 5. Therefore, claim 12 is rejected under the same rationale as claim 5.
Claim 15
(…) training an instance of a tree-based inference model using training data,
(Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making” EN: this denotes the model is trained on historical data and this trained model is used for future decision making. Therefore, the training happens “prior to” the future decision. As for “identifying the occurrence”, that is taught in claim 1 above. Additionally, the paper is titled “Discrimination Aware Decision Tree Learning” learning is synonymous to training in the BRI.)
wherein the tree based inference model is based on:
a splitting rule that partition the training data for predictive ability that incentivizes labels of the training data and that disincentivizes bias labels of the training data; and
(Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”, Section 5.1: “Suppose that a certain split divides the data D into D1, . . . , Dk”, Section 4.2,“Let a labeled dataset D and a sensitive attribute B be given. The discrimination aware classification problem is to learn a classifier such that (a) The accuracy of C is high, and (b) the discrimination of C w. r. t. B is low.”, Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making”)
wherein the splitting rule partitions records of the training data into two groups using a function that rewards predictability of the labels by the instance of the tree-based inference model and discourages predictability of the bias labels by the instance of the tree-based inference model, (Algorithm 1, Step 5 and step 7-8. And Section 5.1 page 15, The IGC-IGS function. EN: this denotes the function IGC-IGS that is inserted into Step 5 of the Algorithm 1 to determine which attribute is selected. Based on that selection, Step 8 of Algorithm 1 partitions the data into groups. Additionally, in section 5.1, it denotes calculating IGC (information Gain w.r.t. Class). The information gain is the standard metric for measuring the predictability of labels (accuracy). Higher IGC represented as a higher positive term in the formulas, thus the function “rewards” predictive power for the labels. Also in section 5.1, IGS measures how well a split predicts a bias feature. By subtracting IGS from the total score (IGC – IGS), the function penalizes or discourages splits that result in high predictability of the bias feature. Therefore, the (IGC – IGS) formula denotes a function that rewards label accuracy (+IGC) and discourages bias predictability (-IGS).)
wherein the function assigns a numerical value based on division of the records among the two groups,
PNG
media_image2.png
571
943
media_image2.png
Greyscale
(Section 5.1 EN: This denotes a formula (function) Gain = IGC – IGS. The resulting “Gain” is a score which is a number maximized by the algorithm. The above screenshot is a section from 5.1 which also denotes the division of data (records).)
and the records are partitioned through optimization of the function, (Algorithm 1: Decision Tree Induction in Section 4, Line 5: “Select test att from att list and test s.t. gain(test att, test) is maximized” and Section 5.1, particularly, “Based on these two measures IGC and IGS, we introduce three alternative criteria for determining the best split:” EN: This denotes maximizing the gain function (IGC - IGS) which falls under mathematical optimization.)
“obtaining the tree based inference model from among the inference models,” (Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”)
obtaining the inference using the tree-based inference model; and (Section 1 introduction “we want to learn a classification model from the potentially biased historical data such that it generates accurate predictions for future decision making” Section 5, “The first solution is based on the adaptation of splitting criterion for tree construction to build a discrimination-aware decision tree”))
Kamiran does not distinctly disclose:
“A data processing system, comprising: a processor; and a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations for providing computer implemented services using inference models, the operations comprising: identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services; based on the occurrence:” and “providing computer implemented services using the inference.”
However, Gauci teaches:
A data processing system, comprising: a processor; and a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations for providing computer implemented services using inference models, (Figure 8, with a non-transitory memory (802) storing a prediction module (830) that is executed by a processor (818) to perform the service using prediction models. The prediction models being a part of the prediction module, the prediction module is linked to the prediction engine (figure 3) and the prediction engine is has prediction models (paragraph 73).) the operations comprising: identifying an occurrence of a condition that indicates an inference is necessary to provide the computer implemented services;(Figure 1 and paragraph 20-22, EN: this denotes detecting the triggering event which is the occurrence of a condition. This detection indicates that the system must now use a “prediction model” which is the inference to suggest the correct app to the user (the opening of the suggested app is the computer implemented service). based on the occurrence (Gauci, Paragraph 97, “At block 506, the device selects a prediction model… The selected prediction model may depend on the triggering event.” )
Gauci further teaches: providing computer implemented services using the inference. (Gauci, Figure 1 block 106 which states “perform action associated with application”. EN: This denotes opening an application for the user from the outcome of the prediction model. The opening of an application is the computer implemented services and the prediction model which includes a prediction/inference.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran with the event-triggered system that selects prediction models to provide computer-implemented services that’s part of an over encompassing machine-readable medium of Gauci. Combining the teachings of Kamiran and Gauci allows the theoretical fairness benefits of Kamiran to be realized in a scalable, and automated manner, that utilizes machine, and thus effectively upgrading a static research model into a functioning, real-time computer service.
Kamiran in view of Gauci does not explicitly teach:subsuming nodes to limit a total number of nodes of the tree-based inference model,
However, Ignatov teaches:
subsuming nodes to limit a total number of nodes of the tree-based inference model,
(Page 1 Abstract, “we propose merging nodes from different branches based on their similarity that is estimated with two sample test statistics, which leads to generation of a deep directed acyclic graph of decision rules” Page 3, “In this section, we describe the proposed Decision Stream algorithm. The main concept of this method consists in merging similar nodes after each splitting iteration […] fuses statistically similar nodes (Fig. 2(b-c)) using an input parameter—significance threshold Plim” – EN: this denotes merging, analogous to subsuming leaf nodes after each splitting iteration based on statistical similarity, which reduces the total number of nodes in the model structure as shown in Figure 5.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran with the event-triggered system that selects prediction models to provide computer-implemented services that’s part of an over encompassing machine-readable medium of Gauci with the node merging technique of Ignatov. The motivation for doing so would be to reduce model complexity and prevent by addressing the issue of data exhaustion in decision tree leaf nodes, thereby improving prediction accuracy. Ignatov Abstract, “Tree node splitting based on relevant feature selection is a key step of decision tree learning, at the same time being their major shortcoming: the recursive nodes partitioning leads to geometric reduction of data quantity in the leaf nodes, which causes an excessive model complexity and data overfitting.” Ignatov further teaches that merging nodes directly addresses this problem, as stated in section 5 (conclusion), “a Decision Stream, which avoids the problems of data exhaustion and formation of unrepresentative data samples in decision tree nodes by merging the leaves from the same and/or different levels of the predictive model structure.”
Claim 16 is a data processing system claim that recite substantially the same limitation as claim 2. Therefore, claim 9 is rejected under the same rationale as claim 2.
Claim 18 is a data processing system claim that recite substantially the same limitation as claim 4. Therefore, claim 18 is rejected under the same rationale as claim 4.
Claim 19 is a data processing system claim that recite substantially the same limitation as claim 5. Therefore, claim 19 is rejected under the same rationale as claim 5.
Claim 21
Kamiran teaches:
The method of claim 1, wherein predictive power of the labels and predictive power of the bias labels are (…) (Section 5.1,
PNG
media_image7.png
554
597
media_image7.png
Greyscale
EN: this denotes IGC which is the information gain with respect to the class label, which is the metric for measuring the predictive power of the labels. IGS is the information gain with respect to the sensitive attribute B, which measures the predictive power of the bias labels.)
Kamiran does not explicitly teach:
“compared to a threshold.”
However, Ignatov teaches:
“compared to a threshold.” (Page 5,
PNG
media_image8.png
275
773
media_image8.png
Greyscale
Page 5, “The splitting and merging operations are performed according to significance threshold Plim. We take as the null hypothesis that labels of two nodes are from the same distribution and have the same mean value. The null hypothesis is rejected at the significance level Plim” – EN: this denotes comparing computed statistical measures to a threshold value (P_lim) to make training decisions.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran and the event-triggered system that selects prediction models to provide computer-implemented services of Gauci with the threshold comparison of Ignatov. The motivation for doing so would be to prevent unnecessary splits that do not improve the model and avoiding overfitting. See Ignatov, Section 3.2, page 4, “A leaf is marked as terminal if it cannot be split into statistically different child nodes.” Page 5, “Then the similarity function Sst is calculated for each split, and the one with the lowest significance of similarity is selected. If this significance is smaller than the input threshold Plim, the selected best split is returned, otherwise — splitting is rejected and the node becomes terminal. Though this method is rather computationally expensive, it provides the best split quality and is reasonable for compact datasets.”
Claim 22
Ignatov teaches:
The method of claim 21, wherein satisfying the threshold indicates a decision tree of the tree based inference model is complete. (Page 4, “A leaf is marked as terminal if it cannot be split into statistically different child nodes. The pair of splitting and merging steps is iteratively performed till the stopping criterion is met. If all leaves are terminal or the prediction accuracy is not improved, the DS training is finished and Algorithm 2 returns the reference to the root node of the generated DS.” (Page 4, Algorithm 2, Page 5 Algorithm 3,
PNG
media_image9.png
450
775
media_image9.png
Greyscale
PNG
media_image10.png
264
756
media_image10.png
Greyscale
EN: this denotes where satisfying the (P_lim) threshold causes a node’s split to be rejected (Algorithm 3 returns Ø in line 7), which causes that node to be marked terminal (Algorithm 2 lines 10-11), and when all the nodes reach this terminal state through satisfying this threshold, the decision tree is complete.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran and the event-triggered system that selects prediction models to provide computer-implemented services of Gauci with the threshold comparison for tree completion of Ignatov. The motivation for doing so would be to provide an automatic stopping criteria that avoids underfitting and overfitting of the tree. See Ignatov Page 5, “If this significance is smaller than the input threshold Plim, the selected best split is returned, otherwise — splitting is rejected and the node becomes terminal. Though this method is rather computationally expensive, it provides the best split quality and is reasonable for compact datasets”
Claim 23
Ignatov teaches:
The method of claim 1, wherein each group provided by the tree-based inference model includes an average of the labels of the corresponding group. (Page 4, “Function Sst returns the significance level p representing the probability that the mean values of labels associated with these two nodes are identical.” Page 5, “We take as the null hypothesis that labels of two nodes are from the same distribution and have the same mean value.” – EN: this denotes that each node is associated with a mean value of its labels (average of the levels of the samples in that node), because the entire merging and splitting framework operates by comparing whether the mean values of labels between nodes are identical, which requires each group/node to include a computed average of the labels of its corresponding records.)
Before the effective filing date of the invention, it would have been obvious to one of ordinary skill in the art to combine the discrimination-aware decision tree learning method that modifies splitting criteria to reduce bias of Kamiran and the event-triggered system that selects prediction models to provide computer-implemented services of Gauci with the per-node mean label computation of Ignatov. The motivation for doing so would be to enable statistically grounded evaluation of each group’s label distribution, which is necessary for both making accurate predictions and for determining which nodes are sufficiently similar to be merged. See Ignatov, Page 5, “We take as the null hypothesis that labels of two nodes are from the same distribution and have the same mean value.”
Claim 24-26 are non-transitory machine-readable medium claims that recite substantially the same limitations as claims 21-23. Therefore, claims 24-26 are rejected under the same rationale as claim 21-23.
Claim 27, 28 are data processing system claims that recite substantially the same limitations as claims 21, 23. Therefore, claims 27, 28 are rejected under the same rationale as claim 21, 23.
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 NAYMUR RAHMAN ALI whose telephone number is (571)272-0007. The examiner can normally be reached Mon-Fri. 9:30-6:30 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, Alexey Shmatov can be reached at (571)270-3428. 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.
/NAYMUR RAHMAN ALI/Examiner, Art Unit 2123
/ALEXEY SHMATOV/Supervisory Patent Examiner, Art Unit 2123
Prosecution Insights