DETAILED ACTION
Examiner Remarks
In regards to Applicant’s Arguments and Amendments in the Remarks submitted on 10/21/2025, Examiner has withdrawn the claim objections and rejections issued in the Non-Final Office Action of 07/29/2025.
Response to Arguments
101
Applicant argues that the 101 rejection for claims 1-181 have been overcome since the amended claims are directed to an improvement in dynamic debiasing of artificial intelligence models. See pg. 10 of Applicant’s Remarks submitted on 10/21/2025.
Respectfully, Examiner disagrees. As MPEP §2106.05(a) states:
After the examiner has consulted the specification and determined that the disclosed invention improves technology, the claim must be evaluated to ensure the claim itself reflects the disclosed improvement in technology...[t]hat is, the claim must include the components or steps of the invention that provide the improvement described in the specification. However, the claim itself does not need to explicitly recite the improvement described in the specification (e.g., "thereby increasing the bandwidth of the channel")...[a]n important consideration in determining whether a claim improves technology is the extent to which the claim covers a particular solution to a problem or a particular way to achieve a desired outcome, as opposed to merely claiming the idea of a solution or outcome. McRO, 837 F.3d at 1314-15, 120 USPQ2d at 1102-03; DDR Holdings, 773 F.3d at 1259, 113 USPQ2d at 1107(Emphasis added).
In this case Examiner has identified paras. [0008-0009] and [0011-0012] of Applicant’s Specification as disclosing the invention’s improvement in technology with respect to dynamic de-biasing of AI models to be fair when the data distribution used by the model has evolved dynamically over time due to updates in the data distribution. While Applicant’s Specification provides the necessary details for this improvement, Applicant’s claims do not reflect this improvement, since the independent claims do not reflect the dynamic nature of dealing with data distributions that change over time. Accordingly, the 101 rejection for claims 1-20 have not been withdrawn.
103
Applicant argues that the prior art of Chaloulos does not teach the amended limitations of receiving a dataset having a plurality of entries with one or more attributes and data corresponding to the one or more attributes according to a trained machine learning model, the data including a set of reward probabilities and a bias tolerance threshold, the trained machine learning model being deemed fair at an initial state and determining bias has developed in the trained machine learning model at a later state, the bias being toward the first attribute based on a change to the received reward probability of the first attribute being outside the bias tolerance threshold. See pgs. 10-12 of Applicant’s Remarks submitted on 10/21/2025.
Respectfully, Examiner disagrees. Chaloulos in view of Wang teaches both amended claim limitations see the Current Office Action for the detailed teachings. Furthermore, with respect to the newly added dependent claims of 21-22 Chaloulos in view of Wang also teaches those limitations and Examiner directs Applicant to see the Current Office Action for the detailed teaching.
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 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 17-18 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.
Claims 17 and 18 recite the limitation "The computer readable medium" of claim 15 and claim 17 however claim 15 does not state computer readable medium but rather states one or more computer readable storage medium. There is insufficient antecedent basis for this limitation in the claim.
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 1-18 and 21-22 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claim 1 partly recites the following limitations:
...the data including a set of reward probabilities and a bias tolerance threshold...selecting a first entry having a maximum reward based on a received reward probability of a first attribute;...updating the reward probability of the first attribute based on the predicted decision; and determining bias has developed… at a later state, the bias being toward the first attribute based on a a change to the received reward probability of the first attribute being outside the bias tolerance threshold.
These limitations as drafted are a process under Step 1 that under its broadest
reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A,
Prong Two because the claim recites the following additional elements:
receiving a dataset having a plurality of entries with one or more attributes and data corresponding to the one or more attributes according to a trained machine learning model;
predicting, using an artificial intelligence agent, a decision of the trained machine learning model based on the first entry;
in the trained machine learning model
The additional claim elements of receiving a dataset having a plurality of entries with one or more attributes and data corresponding to the one or more attributes according to a trained machine learning model amount to mere insignificant extra-solution activity in which the limitations amount to general data gathering, manipulation and/or outputting of data (i.e., receiving and/or providing data).
The additional claim elements of predicting, using an artificial intelligence agent, a decision of the trained machine learning model based on the first entry; in the trained machine learning model amounts to the general linking of the judicial exception to the particular technological environment of machine learning since it merely confines the use of the abstract idea to the particular technological environment of machine learning.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under Step 2B because as discussed above predicting, using an artificial intelligence agent, a decision of the trained machine learning model based on the first entry; in the trained machine learning model amounts to the general linking of the judicial exception to the particular technological environment of machine learning and thus does not recite significantly more than the judicial exception. And the additional elements of receiving a dataset having a plurality of entries with one or more attributes and data corresponding to the one or more attributes according to a trained machine learning model are well-understood, routine, conventional activity that court decisions, such as Symantec and buySAFE cited in MPEP 2106.05(d)(II) have indicated that the mere receiving and/or sending of data over a network using a generic computer are well- understood, routine, and conventional functions when claimed in a merely generic manner (as it is here). Even when considered in combination, these additional elements represent mere instructions to apply an exception with well understood, routine, and conventional activity, which does not provide significantly more to the abstract idea.
Accordingly, claim 1 is not patent eligible.
Claim 2 partly recites the following limitations:
selecting a sample associated with the first entry; and wherein: updating the reward probability of the first attribute is further based on observed rewards associated with the selected sample; and the change to the reward probability of the first attribute is, based on comparing the updated reward probability to the received reward probability of the first attribute in the set of reward probabilities
These limitations as drafted are a process under Step 1 that under its broadest
reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A, Prong
Two because there are no additional elements recited in the claim beyond the judicial exception. And the claims do not include additional elements that are sufficient to amount to significantly
more than the judicial exception under Step 2B because there are no additional elements recited in the claim beyond the judicial exception.
Accordingly, claim 2 is not patent eligible.
Claim 3 partly recites the following limitations:
wherein the plurality of entries in the dataset correspond to one or more borrowers and the set of reward probabilities correspond to credit scores associated with each of the one or more borrowers.
These limitations, as drafted, are a process under Step 1 that under its broadest reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A, Prong
Two because there are no additional elements recited in the claim beyond the judicial exception. And the claims do not include additional elements that are sufficient to amount to significantly
more than the judicial exception under Step 2B because there are no additional elements recited in the claim beyond the judicial exception.
Accordingly, claim 3 is not patent eligible.
Claim 4 partly recites the following limitations:
determining a distribution of credit scores associated with one or more groups of the borrowers having a common attribute; updating the distribution of credit scores based on a prediction … and a repayment probability associated with each group of the borrowers; detecting whether the updated distribution of credit scores crosses a pre-defined distribution tolerance threshold; and updating the credit score of the borrowers based on the repayment probability.
These limitations as drafted are a process under Step 1 that under its broadest
reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A,
Prong Two because the claim recites the following additional elements:
of the trained machine learning
The additional claim elements of of the trained machine learning to the general linking of the judicial exception to the particular technological environment of machine learning since it merely confines the use of the abstract idea to the particular technological environment of machine learning.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under Step 2B because as discussed above of the trained machine learning amounts to the general linking of the judicial exception to the particular technological environment of machine learning and thus does not recite significantly more than the judicial exception.
Accordingly, claim 4 is not patent eligible.
Claim 5 partly recites the following limitations:
determining, for each iteration of selecting and predicting, whether the bias has developed in the…model.
These limitations as drafted are a process under Step 1 that under its broadest
reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A,
Prong Two because the claim recites the following additional elements:
trained machine learning
The additional claim elements of trained machine learning amounts to the general linking of the judicial exception to the particular technological environment of machine learning since it merely confines the use of the abstract idea to the particular technological environment of machine learning.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under Step 2B because as discussed above trained machine learning amounts to the general linking of the judicial exception to the particular technological environment of machine learning and thus does not recite significantly more than the judicial exception.
Accordingly, claim 5 is not patent eligible.
Claim 6 partly recites the following limitations:
notifying a user of the bias within the…model.
These limitations as drafted are a process under Step 1 that under its broadest
reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A,
Prong Two because the claim recites the following additional elements:
trained machine learning
The additional claim elements of trained machine learning amounts to the general linking of the judicial exception to the particular technological environment of machine learning since it merely confines the use of the abstract idea to the particular technological environment of machine learning.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under Step 2B because as discussed above trained machine learning amounts to the general linking of the judicial exception to the particular technological environment of machine learning and thus does not recite significantly more than the judicial exception.
Accordingly, claim 6 is not patent eligible.
Claim 7 partly recites the following limitations:
remediating the bias within the…model.
These limitations as drafted are a process under Step 1 that under its broadest
reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A,
Prong Two because the claim recites the following additional elements:
trained machine learning
The additional claim elements of trained machine learning amounts to the general linking of the judicial exception to the particular technological environment of machine learning since it merely confines the use of the abstract idea to the particular technological environment of machine learning.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under Step 2B because as discussed above trained machine learning amounts to the general linking of the judicial exception to the particular technological environment of machine learning and thus does not recite significantly more than the judicial exception.
Accordingly, claim 7 is not patent eligible.
Because claim 8 is directed to a machine the claimed invention is directed to statutory subject matter under Step 1 and under Step 2A Prong Two and Step 2B the additional claim elements of a computer readable storage medium configured to store computer program code; and one or more computer processors configured to access said computer program code and operate as instructed by said computer program code, are not sufficient to amount to significantly more than the judicial exception since these additional claim elements are recited at a high level of generality (i.e. using a generic processor and generic memory) and for all other claim elements of claim 8 they are rejected using the PEG analysis of claim 1 since they are analogous claims.
Because dependent claims 9-14 are directed to a machine the claimed invention is directed to statutory subject matter under Step 1 and for all other claim limitations they are rejected on the same basis as dependent claims 2-7 since they are analogous claims.
Because claim 15 is directed to a manufacture the claimed invention is directed to statutory subject matter under Step 1 and under Step 2A Prong Two and Step 2B the additional claim elements of one or more computer readable storage medium having stored thereon a computer program for online fairness monitoring, the computer program, when executed by a processor are not sufficient to amount to significantly more than the judicial exception since these additional claim elements are recited at a high level of generality (i.e. using a generic processor and generic memory) and for all other claim elements of claim 15 they are rejected using the PEG analysis of claim 1 since they are analogous claims.
Because dependent claims 16-18 are directed to a manufacture the claimed invention is directed to statutory subject matter under Step 1 and for all other claim limitations they are rejected on the same basis as dependent claims 2-4 since they are analogous claims.
Claim 21 partly recites the following limitations:
updating a distribution of reward probabilities based on the received set of reward
probabilities; and determining the bias has developed...is further based on a change in the distribution of reward probabilities being above a distribution tolerance threshold.
These limitations as drafted are a process under Step 1 that under its broadest
reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A,
Prong Two because the claim recites the following additional elements:
in the training machine learning model
The additional claim elements of in the training machine learning model
amounts to the general linking of the judicial exception to the particular technological environment of machine learning since it merely confines the use of the abstract idea to the particular technological environment of machine learning.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under Step 2B because as discussed above in the training machine learning model amounts to the general linking of the judicial exception to the particular technological environment of machine learning and thus does not recite significantly more than the judicial exception.
Accordingly, claim 21 is not patent eligible.
Claim 22 partly recites the following limitations:
establishing an initial reference distribution of reward probabilities...which is initially determined to be fair; and determining the bias is further based on comparing the updated distribution of reward probabilities with the initial reference distribution of reward probabilities to establish the change in the distribution of reward probabilities.
These limitations as drafted are a process under Step 1 that under its broadest
reasonable interpretation can be performed in the human mind through the use of observations, evaluations, judgements and opinion and falls under the mental process grouping. Thus, the claim recites a mental process under Step 2A, Prong One.
This judicial exception is not integrated into a practical application under Step 2A,
Prong Two because the claim recites the following additional elements:
for the machine learning model
The additional claim elements of for the machine learning model amounts to the general linking of the judicial exception to the particular technological environment of machine learning since it merely confines the use of the abstract idea to the particular technological environment of machine learning.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception under Step 2B because as discussed above for the machine learning model amounts to the general linking of the judicial exception to the particular technological environment of machine learning and thus does not recite significantly more than the judicial exception.
Accordingly, claim 22 is not patent eligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
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.
Claims 1-2, 5, 7-9, 12, 14-16 and 21-22 are rejected under 35 U.S.C. 103 as being unpatentable over Chaloulos et al., US 2020/0320428 Al (“Chaloulos’) in view of Wang, Lequn, et al. "Fairness of exposure in stochastic bandits." International Conference on Machine Learning. PMLR, 2021(“Wang”).
Regarding claim 1, Chaloulos teaches a method of online fairness monitoring in a machine learning model, executable by a processor(Chaloulos, paras. 0076, see also fig. 8, “As shown in the figure, computer system/server 800 is shown in the form of a general-purpose computing device. The components of computer system/server 800 may include, but are not limited to, one or more processors or processing units 802….”), comprising:
receiving a dataset having a plurality of entries with one or more attributes and data corresponding to the one or more attributes according to a trained machine learning model, the data including [a set of reward probabilities] and a bias tolerance threshold, the trained machine learning model being deemed fair at an initial state(Chaloulos, paras. 0057-0061, see also fig. 5, “The evaluation portion may be symbolized by the blocks 504, 506, 508, 510, 512. After the supervised machine-learning model has been trained[and data corresponding to the one or more attributes according to a trained machine learning model,], it is evaluated with respect to both, prediction performance and prediction fairness. This is achieved by letting the trained supervised machine-learning model perform a set of predictions on a validation dataset that has not been used in the training process of the SML model…[t]his may be done by splitting a test dataset of pre-selected critical/protected features, 506, and by evaluating a fairness measurement on each of the subsets, 508[receiving a dataset having a plurality of entries with one or more attributes and data corresponding to the one or more attributes according to a trained machine learning model].” & Chaloulos, paras. 0062-0069, see also fig. 6, “The performance metric of the model is evaluated, 604, by computing the F-score on the test dataset. To illustrate the example, an F-score of 0.8 is assumed for this iteration[the trained machine learning model being deemed fair at an initial state]...[t]he bias metric is in this example defined as (-1) if fairness metric>0.1%, 0 if fairness metric<0.1%[the data including and a bias tolerance threshold].” );2
selecting a first entry [having a maximum reward] based on a received reward probability of a first attribute(Chaloulos, paras. 0062-0069, see also fig. 6, “Next, the bias on the protected metric (e.g., gender) is measured. In this example, this is done by randomly
sampling a set of 100 women and a matching set of 100 men that match the selected women as closely as possible on all attributes except the gender, 606[selecting a first entry; of a first attribute]… for the later used reinforcement learning algorithm a reward function is computed, 612, which is a combination of performance and bias[based on a received reward probability].”);3
predicting, using an artificial intelligence agent, a decision of the trained machine learning model based on the first entry(Chaloulos, paras. 0070-0071, “[T]he RL engine 616 therefore finds an optimal combination of parameters and/or hyper-parameters that decreases the bias under the specified threshold while maintaining a high performance which results in a fair credit scoring across gender[predicting, using an artificial intelligence agent, a decision of the trained machine learning model based on the first entry].”);
updating the reward probability of the first attribute based on the predicted decision(Chaloulos, paras. 0070-0071, “[T]he RL engine 616 therefore finds an optimal combination of parameters and/or hyper-parameters that decreases the bias under
the specified threshold while maintaining a high performance which results in a fair credit scoring across gender. In a next step, the RL engine uses the computed reward to efficiently optimize the parameters and/or hyperparameters of the SML model[updating the reward probability of the first attribute based on the predicted decision].”);
and determining bias has developed in the trained machine learning model at a later state, the bias being toward the first attribute based on a change to the received reward probability of the first attribute being outside the bias tolerance threshold(Chaloulos, paras. 0069-0073, see also figs. 6 and 7, “In the case of the example, therefore a reward of R=0.8-
1=-0.2 (since 10%>0.1%) is fed to the RL network[the bias being toward the first attribute based on a change to the received reward probability of the first attribute being outside the bias tolerance threshold]. A termination engine 614 uses the computed reward to decide if any iteration is to be initiated[and determining bias has developed in the trained machine learning model at a later state].”).4
While Chaloulos does teach selecting a first entry based on a received reward probability of a first attribute, Chaloulos does not teach: a set of reward probabilities; having a maximum reward.
However, Wang teaches:
a set of reward probabilities(Wang, pg., 5, “[E]ach arm a at round t comes
with a context vector
x
t
,
a
∈
R
d
. A stochastic linear bandits instance
v
=
(
P
x
:
x
∈
R
d
)
is a collection of reward distributions for each context vector[a set of reward probabilities].”);
having a maximum reward(Wang, pg., 4, see also algorithm 1, “At each round t, the algorithm constructs a confidence region
C
R
t
which contains the true
parameter μ* with high probability. Then the algorithm optimistically selects a parameter
μ
t
∈
R
K
within the confidence region
C
R
t
that maximizes the estimated expected reward subject to the constraint that we construct a fair policy as if the selected parameter is the true parameter[having a maximum reward]”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Chaloulos with the teachings of Wang the motivation to do so would be to modify the learning problem associated with standard linear contextual bandit reinforcement learning to include fairness towards during the learning process(Wang, pgs. 1-2, “To overcome these problems of the conventional bandit objective,
we propose a new formulation of the bandit problem that implements the principle of Merit-based Fairness of Exposure… [i]t incorporates the additional fairness requirement that each item/arm receives a share of exposure that is proportional to its merit.”).
Regarding claim 2, Chaloulos in view of Wang teaches the method of claim 1, wherein determining the bias exists toward the first attribute comprises:
selecting a sample associated with the first entry(Chaloulos, paras. 0062-0069, see also fig. 6, “Next, the bias on the protected metric (e.g., gender) is measured. In this example, this is done by randomly sampling a set of 100 women and a matching set of 100 men that match the selected women as closely as possible on all attributes except the gender, 606[selecting a sample associated with the first entry]….”);
and wherein: updating the reward probability(Wang, pgs. 4-5, see also line 7 of algorithm 2, “Finally, the algorithm observes the feedback and updates the posterior distribution of the true parameter.”) of the first attribute is further based on observed rewards associated with the selected sample(Chaloulos, paras. 0069-0073, see also figs. 6 and 7, “Chaloulos, paras. 0062-0069, see also fig. 6, “To illustrate the example, a fairness metric value of 10% (average relative difference) is assumed… a reward function is computed, 612, which is a combination of performance and bias…[t]he bias metric is in this example defined as ( -1) if fairness metric>0.1%, 0 if fairness metric<0.l %. In the case of the example, therefore a reward of R=0.8-1=-0.2 (since 10%>0.1%) is fed to the RL network. A termination engine 614 uses the computed reward to decide if any iteration is to be initiated. It terminates the process if…the reward is not improving anymore”);
and the change to the reward probability of the first attribute is, based on comparing the updated reward probability to the received reward probability of the first attribute in the set of reward probabilities (Chaloulos, paras. 0062-0069, see also fig. 6, “To illustrate the example, an F-score of 0.8 is assumed for this iteration. Next, the bias on the protected metric (e.g., gender) is measured… a fairness metric value of 10% (average relative difference) is assumed… a reward function is computed, 612, which is a combination of performance and bias…[t]he bias metric is in this example defined as ( -1) if fairness metric>0.1%, 0 if fairness metric<0.l %. In the case of the example, therefore a reward of R=0.8-1=-0.2 (since 10%>0.1%) is fed to the RL network. A termination engine 614 uses the computed reward to decide if any iteration is to be initiated. It terminates the process if…the reward is not improving anymore (no convergence). It may also be specified that the fairness metric must be lower than 0.1% to terminate[and the change to the reward probability of the first attribute is, based on comparing the updated reward probability to the received reward probability of the first attribute in the set of reward probabilities].”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Wang with the above teachings of Chaloulos for the same rationale stated at Claim 1.
Regarding claim 5, Chaloulos in view of Wang teaches the method of claim 1, further comprising: determining, for each iteration of selecting and predicting, whether the bias has developed in the trained machine learning model (Chaloulos, paras. 0057-0061, see also fig. 5, “To this end, the RL engine navigates the space of possible machine-learning models…and model parameters as well as hyper-parameters by calculating a reward which is based both, on performance and fairness of the predictions by the supervised machine-learning model. The reinforcement learning engine 516 will iteratively optimize this reward function value by exploring several different configurations of the supervised machine-learning model, and the termination engine 514 determines that a suitable predictor has been found based on the convergence of the reward function[determining, for each iteration of selecting and predicting, whether the bias has developed in the trained machine learning model].”).
Regarding claim 7, Chaloulos in view of Wang teaches the method of claim 1, further comprising: remediating the bias within the trained machine learning model(Chaloulos, paras. 0057-0061, see also fig. 5, “The reinforcement learning engine 516 will iteratively optimize this reward function value by exploring several different configurations of the supervised machine-learning model, and the termination engine 514 determines that a suitable predictor has beenfound based on the convergence of the reward function. The termination engine 514 may also have other explicit criteria that must be fulfilled before termination occurs, e.g., constraints on the achieved fairness. If the reward function outcomes have converged and the supervised machine learning model meets the pre-specified performance and fairness requirements, the optimized supervised machine learning model (SML) is presented as a result, 518[remediating the bias within the trained machine learning model].”).
Regarding claim 8, Chaloulos teaches a computer readable storage medium configured to store computer program code; and one or more computer processors configured to access said computer program code and operate as instructed by said computer program code(Chaloulos, paras. 0074-0080, “As shown in the figure, computer system/server 800 is shown in the form of a general-purpose computing device. The components of computer system/server 800 may include, but are not limited to, one or more processors or
processing units 802, a system memory 804…[t]he system memory 804 may include computer
system readable media[a computer readable storage medium configured to store computer program code; and one or more computer processors configured to access said computer program code and operate as instructed by said computer program code]”) and for all other claim limitations they are rejected on the same basis as independent claim 1 since they are analogous claims.5
Referring to dependent claims 9, 12, and 14, they are rejected on the same basis as
dependent claims 2, 5, and 7 since they are analogous claims.
Regarding claim 15, Chaloulos teaches a computer program product comprising one or more computer readable storage medium having stored thereon a computer program for online fairness monitoring, the computer program, when executed by a processor(Chaloulos, paras. 0072-0080, “[A] repetition unit 708 is adapted for triggering the selector unit and the controller iteratively for an improvement of a fairness value of the supervised machine-learning model[a computer program for online fairness monitoring]…[a]s shown in the figure, computer system/server 800 is shown in the form of a general-purpose computing device. The components of computer system/server 800 may include, but are not limited to, one or more processors or processing units 802, a system memory 804…[t]he system memory 804 may include computer system readable media[one or more computer readable storage medium having stored thereon a computer program for online fairness monitoring, the computer program, when executed by a processor]”) and for all other claim limitations they are rejected on the same basis as independent claim 1 since they are analogous claims.
Referring to dependent claim 16 it is rejected on the same basis as
dependent claims 2 since they are analogous claims.
Regarding claim 21, Chaloulos in view of Wang teaches the method of claim 1, further comprising: updating a distribution of reward probabilities based on the received set of reward probabilities(Wang, pg., 3, “[R]eward distributions
v
=
(
P
a
:
a
∈
[
K
]
), where
P
a
is the reward distribution of arm a with mean
μ
a
*
=
E
r
~
P
a
[
r
]
... the goal of learning is to maximize
the cumulative expected reward
∑
t
=
1
T
E
a
t
~
π
t
μ
a
*
[updating a distribution of reward probabilities based on the received set of reward probabilities]”); and
determining the bias has developed in the training machine learning model is further based on a change in the distribution of reward probabilities being above a distribution tolerance threshold(Chaloulos, paras. 0068-0069, “In the case of the example, therefore a reward of R=0.8-1=-0.2 (since 10%>0.1%) is fed to the RL network[determining the bias has developed in the training machine learning model]. A termination engine 614 uses the computed reward to decide if any iteration is to be initiated. It terminates the process...if the reward is not improving anymore (no convergence)[ is further based on a change in the distribution of reward probabilities being above a distribution tolerance threshold].”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Wang with the above teachings of Chaloulos for the same rationale stated at Claim 1.
Regarding claim 21, Chaloulos in view of Wang teaches the method of claim 21, further comprising: establishing an initial reference distribution of reward probabilities for the machine learning model, which is initially determined to be fair(Wang, pg., 3, “The merit function
f
is an input to the bandit algorithm, and it provides a design choice that permits tailoring the fairness criterion to different applications. The following theorem shows that there is a unique policy that satisfies the above fairness constraints. For any mean reward
parameter
μ
*
and any choice of merit function
f
⋅
>
0
there exist a unique policy... that fulfills the merit-based fairness....[ establishing an initial reference distribution of reward probabilities for the machine learning model, which is initially determined to be fair]”);
and determining the bias is further based on comparing the updated distribution of reward probabilities with the initial reference distribution of reward probabilities to establish the change in the distribution of reward probabilities(Wang, pg., 3, “We thus define the reward regret
R
R
T
at round T as the gap[and determining the bias is further based on comparing; to establish the change in the distribution of reward probabilities] between the expected reward of the deployed policy[the updated distribution of reward probabilities] and the expected reward of the optimal fair policy
π
*
[with the initial reference distribution of reward probabilities]”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Wang with the above teachings of Chaloulos for the same rationale stated at Claim 1.
Claims 3-4, 10-11, and 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Chaloulos et al., US 2020/0320428 Al (“Chaloulos’) in view of Wang, Lequn, et al. "Fairness of exposure in stochastic bandits." International Conference on Machine Learning. PMLR, 2021(“Wang”) and in view of D'Amour, Alexander, et al. "Fairness is not static: deeper understanding of long term fairness via simulation studies." Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency. 2020(“D'Amour”)
Regarding claim 3, Chaloulos in view of Wang teaches the method of claim 1, but do not teach: wherein the entries in the dataset correspond to one or more borrowers and the reward probability corresponds to a credit score associated with each of the one or more borrowers.
However, D'Amour teaches:
wherein the plurality of entries in the dataset correspond to one or more borrowers and the set of reward probabilities correspond to credit scores associated with the one or more borrowers(D'Amour, pgs. 3-5, “In particular, we consider the lending scenario… where an agent representing a bank makes decisions about whether to approve or reject applications for loans from a stream of individuals[wherein the plurality of entries in the dataset correspond to one or more borrowers]… [i]n this environment, each loan applicant has an observable group membership variable A and a discrete credit score C
∈
1
,
…
,
C
m
a
x
… [i]f the applicant defaults, the agent’s profit decreases by
r
-
and the applicant’s C value is decreased by
c
-
. If the applicant pays back, the agent’s profit is increased by
r
+
and the applicant’s C value is increased by
c
+
. In this simulation, probability of repaying is a deterministic function of credit score π(C); when an applicant’s score increases or decreases, so, too, does their probability of repaying[and the set of reward probabilities correspond to credit scores associated with the one or more borrowers].”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Chaloulos in view of Wang with the teachings of D'Amour the motivation to do so would be to simulate the long-term behavior characteristics of deployed machine learning models with respect to fairness in decision making (D'Amour, pg. 1, “As machine learning becomes increasingly incorporated within high impact decision ecosystems, there is a growing need to understand the long-term behaviors of deployed ML-based decision systems and their potential consequences. Most approaches to understanding or improving the fairness of these systems have focused on static settings without considering long-term dynamics… long term dynamics are hard to assess, particularly because they do not align with the traditional supervised ML research framework that uses fixed data sets. To address this structural difficulty in the field, we advocate for the use of simulation as a key tool in studying the fairness of algorithms.”).
Regarding claim 4, Chaloulos in view of Wang and in view of D'Amour teaches the method of claim 3, wherein determining the bias exists toward the first attribute comprises: determining a distribution of credit scores associated with one or more groups of the borrowers having a common attribute(Chaloulos, paras. 0062-0069, see also fig. 6, “An SML model for predicting a credit score based on a set of attributes (protected ones and other attributes) is set up… [n]ext, the bias on the protected metric (e.g., gender) is measured[of credit scores associated with one or more groups of the borrowers having a common attribute]…[a] fair evaluation (608) would require each pair to have a similar credit score, any differences are summarized, 610 into the fairness metric. In this example, the fairness metric is calculated as the average relative difference between all pairs. In any other method of measuring bias is acceptable as well. To illustrate the example, a fairness metric value of 10% (average relative difference) is assumed[determining a distribution].”);
updating the distribution of credit scores based on a prediction of the trained machine learning model(Chaloulos, paras. 0062-0070, see also fig. 6, “By going through multiple iterations, the RL engine 616 therefore finds an optimal combination of parameters and/or hyper-parameters that decreases the bias under the specified threshold while maintaining a high performance which results in a fair credit scoring across gender[updating the distribution of credit scores based on a prediction of the trained machine learning model].”)
and a repayment probability associated with each group of the borrowers(D'Amour, pgs. 3-5, see also fig. 4, “In this simulation, probability of repaying is a deterministic function of credit score π(C); when an applicant’s score increases or decreases, so, too, does their probability of repaying[and a repayment probability associated with each group of the borrowers].”);
detecting whether the updated distribution of credit scores crosses a pre-defined distribution tolerance threshold(Chaloulos, paras. 0062-0069, see also fig. 6, “A termination engine 614 uses the computed reward to decide if any iteration is to be initiated…[i]t may also be specified that the fairness metric must be lower than 0.1 % to terminate[detecting whether the updated distribution of credit scores crosses a pre-defined distribution tolerance threshold]”);
and updating the credit score of the borrowers based on the repayment probability(D'Amour, pgs. 3-5, see also fig. 4, “In this environment, each loan applicant has an observable group membership variable A and a discrete credit score C
∈
1
,
…
,
C
m
a
x
… [i]f the applicant pays back, the agent’s profit is increased by
r
+
and the applicant’s C value is increased by
c
+
. In this simulation, probability of repaying is a deterministic function of credit score π(C)[ and updating the credit score of the borrowers based on the repayment probability]….”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Chaloulos in view of Wang with the above teachings of D'Amour for the same rationale stated at Claim 3.
Referring to dependent claims 10-11 they are rejected on the same basis as
dependent claims 3-4 since they are analogous claims.
Referring to dependent claims 17-18 they are rejected on the same basis as
dependent claims 3-4 since they are analogous claims.
Claims 6 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Chaloulos et al., US 2020/0320428 Al (“Chaloulos’) in view of Wang, Lequn, et al. "Fairness of exposure in stochastic bandits." International Conference on Machine Learning. PMLR, 2021(“Wang”) and in view of Ramchandani et al. US 2023/0060452 Al (“Ramchandani”)
Regarding claim 6, Chaloulos in view of Wang teaches the method of claim 1, but does not teach further comprising notifying a user of bias within the machine learning model.
However, Ramchandani teaches:
further comprising notifying a user of the bias within the trained machine learning model(Ramchandani, para. 0061, see also fig. 4, “The model analysis system 402
executes the model 410 for each subgroup of a plurality of subgroups. In some examples, the model analysis system 402 may include tools (e.g., accessible via one or more graphical user interfaces (GUis)) that are provided to users (e.g., such as users of the issuer system 108, merchants, and/or the like) to screen for certain parameters (e.g., sensitive attributes) and detecting a level of bias by a subgroup[notifying a user of the bias within the trained machine learning model].”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Chaloulos in view of Wang with the teachings of Ramchandani the motivation to do so would be to remove discrimination and unethical practices in models used for classification and propensity(Ramchandani, paras. 0003-0004, “Models, such as classification and propensity models, are often used for targeted offers and personalization of products/services. Such models may have unintended consequences
such as discrimination in who receives offers, overwhelming customers with offers, sending unethical offers to vulnerable demographic groups (e.g., alcohol promotion in teenage groups), and/or the like.”).
Referring to dependent claim 13 it is rejected on the same basis as
dependent claim 6 since they are analogous claims.
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 ADAM C STANDKE whose telephone number is (571)270-1806. The examiner can normally be reached Gen. M-F 9-9PM EST.
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, Michael J Huntley can be reached at (303) 297-4307. 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.
/Adam C Standke/
Primary Examiner
Art Unit 2129
1 Examiner Remarks: The Non-Final Office Action of 07/29/2025 rejected claim 1-20 under 101.
2 Examiner Remarks: the claim limitations that are not in bold and contained within square parenthesis i.e., [ ] are claim limitations not taught by Chaloulos
3 Examiner Remarks: the claim limitations that are not in bold and contained within square parenthesis i.e., [ ] are claim limitations not taught by Chaloulos
4 Examiner Remarks: According to the broadest reasonable interpretation (BRI), the use of alternative language amounts to the claim requiring one or more elements but not all.
5 Examiner Remarks: 112(f) was not applied since the first step of the analysis as determined by Examiner was not satisfied since “code” connotes structure. See Dyfan, LLC v. Target Corp., 28 F.4th 1360, 1368-69 (Fed., Cir. 2022); Zeroclick, LLC v. Apple Inc., 891 F.3d 1003, 1008-1009 (Fed. Cir. 2018).
Prosecution Insights