Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This office action is in response to submission of application on 1/25/2024.
Claims 1-15 are presented for examination.
Specification
The title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed.
The following title is suggested: Probabilistic Data Reduction Filter with Dynamic Feedback Learning.
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 1-15 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Independent Claims
Step 2A Prong One: Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes, independent claim 1 recites an abstract idea in the form of mental processes. A mental process is a process that “can be performed in the human mind, or by a human using a pen and paper” (MPEP§ 2106.04(a)(2)(III), paragraph 1). Examples of mental processes include “observations, evaluations, judgments, and opinions” (MPEP § 2106.04(a)(2)(III), paragraph 2).
The following limitations of claim 1 are mental processes:
determining, […], selection estimation values for the data records; [This is a mental process that can be performed by observations, evaluations, judgments, and opinions. Although this limitation recites “data records,” it does not define the structure of such “data records.”]
determining, […], pass-through probabilities for the data records based on the selection estimation values; [This is a mental process that can be performed by observations, evaluations, judgments, and opinions. Although this limitation recites “data records,” it does not define the structure of such “data records.”]
generating a subset of data records by discarding at least a portion of the dataset based on the pass-through probabilities; [This is a mental process that can be performed by observations, evaluations, judgments, and opinions. No specific methodology for generating and discarding data is recited in the claim; therefore, it broadly encompasses processing that can be performed as a mental process.]
Therefore, the independent claims recite a judicial exception.
Step 2A Prong Two: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No. The judicial exception recited in the above discussed claims is not integrated into a
practical application.
“by an estimation module” [These elements constitute no more than mere instructions to apply the judicial exception using generic computer functions (MPEP§2106.04(d)(I)), namely the generic computer function of machine learning. These additional elements merely invoke the use of generic machine learning as a tool to apply an abstract idea, namely an estimation module, which is a generic machine learning component.]
determining, by a pass-through function, pass-through probabilities for the data records based on the selection estimation values; [determining pass-through probabilities are mere instructions to apply the abstract idea. Mere recitation that a judicial exception is to be performed using generic class of computer algorithms in their ordinary capacity, cannot meaningfully integrate the judicial exception into a practical application. See MPEP 2106.05(f). Additionally, this is a description of how the abstract idea is performed, using a pass-through function. As such, this merely describes a technological environment. See MPEP 2106.05(h).]
processing, by a selection module, the subset of data records, wherein the selection module selects one or more data records of the subset of data records [This element recites selecting using a selection module. Only the idea of these selections is recited, without details as to how the make the selections. Therefore, the claim does not include additional limitations that integrate the abstract idea into a practical application. See MPEP 2106.05(f)(1).]
assigning weights and labels to the data records of the subset of data records, wherein the weights reflect a data distribution of the data records with respect to the subset of data records, and wherein the labels represent a selection of the data records by the selection module; [Assigning weights and labels are mere instructions to apply the abstract idea. Mere recitation that a judicial exception is to be performed using generic class of computer algorithms in their ordinary capacity, cannot meaningfully integrate the judicial exception into a practical application. See MPEP 2106.05(f)]
updating the estimation module and the pass-through function based on the subset of data records including the weights and labels [Updating the estimation module and pass-through function based on data are mere instructions to apply the abstract idea. Mere recitation that a judicial exception is to be performed using generic class of computer algorithms in their ordinary capacity, cannot meaningfully integrate the judicial exception into a practical application. See MPEP 2106.05(f)]
Therefore, under MPEP 2106.04(d), the additional elements of the claims do not integrate
the judicial exception into a practical application.
Step 2B: Does the claim recite additional elements that amount to significantly more than the judicial exception?
No. The claims do not include additional elements that are sufficient for the claims to
amount to significantly more than the judicial exception.
Additional elements that are mere instructions to apply an exception or merely generally
linking or generally linking the use of a judicial exception to a particular technological
environment or field of use do not constitute significantly more than a judicial exception under
MPEP§2106.05(I)(A). Since the additional elements in the independent claims are all are mere
instructions to apply an exception or are merely generally linking or generally linking the use of
a judicial exception to a particular technological environment or field of use, they do not
constitute significantly more than a judicial exception.
Therefore, the additional elements identified in the Step 2A Prong Two analysis do not
constitute significantly more than a judicial exception.
Dependent Claims
The remaining dependent claims being rejected do not recite additional elements, whether considered individually or in combination, that are sufficient to integrate the judicial exception into a practical application or amount to significantly more than the judicial exception.
Claims 2
the weights are determined on the pass-through probabilities [These further limitations merely further define the mental process recited in the parent claim and are therefore considered to be part of the mental process of the parent claim. This claim does not recite any additional non-abstract elements for purposes of Step 2A Prong Two and Step 2B analysis.”]
Claims 3
the pass-through function is a trainable function, [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the environment of a trainable function].
wherein updating the pass-through function comprises retraining the pass-through function based on the subset of data records including the weights and labels [Updating the pass-through function based on data are mere instructions to apply the abstract idea. Mere recitation that a judicial exception is to be performed using generic class of computer algorithms in their ordinary capacity, cannot meaningfully integrate the judicial exception into a practical application. See MPEP 2106.05(f)]
Claims 4
the estimation module is a machine learning routine, [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the environment of a trainable function].
wherein updating the estimation module comprises retraining the estimation module based on the subset of data records including the weights and labels [Updating the estimation module based on data are mere instructions to apply the abstract idea. Mere recitation that a judicial exception is to be performed using generic class of computer algorithms in their ordinary capacity, cannot meaningfully integrate the judicial exception into a practical application. See MPEP 2106.05(f)]
Claims 5
the estimation module and the pass-through function are periodically updated [This further limitation is also a mental process. This claim does not recite any additional non-abstract elements for purposes of Step 2A Prong Two and Step 2B analysis.”]
Claims 6
the data records are preprocessed before determining the selection estimation values, wherein preprocessing comprises at least one of merging a data record with additional information from a database, reducing the amount of data in a data record, and adding further information determined based on the data record to the data record [Preprocessing by merging, reducing, and adding data are mere instructions to apply the abstract idea. Mere recitation that a judicial exception is to be performed using generic class of computer algorithms in their ordinary capacity, cannot meaningfully integrate the judicial exception into a practical application. See MPEP 2106.05(f)]
Claims 7
the data records comprise fields and values [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the data records comprise fields and values].
Claims 8
the estimation module is a trained predictive model [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the estimation module is a trained predictive model].
Claims 9
the estimation module is a gradient boosted decision tree or a logistic regression model [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the estimation module is a boosted decision tree or logistic regression].
Claims 10
the pass-through function is a monotonically increasing function [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the pass-through function is monotonically increasing function].
Claims 11
the pass-through probabilities are non-zero probabilities above a threshold [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the pass-through probabilities are non-zero probabilities above a threshold].
Claims 12
the pass-through function comprises two or more trainable parameters. [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the pass-through function comprises two or more trainable parameters].
Claims 13
two trainable parameters of the two or more trainable parameters are the slope of the transition from low to high pass-through probabilities and the location of the transition [This additional element does no more than generally link the use of a judicial exception to a particular technological environment or field of use (MPEP § 2106.05(h)). This element merely indicates a field of use or technological environment in which a judicial exception is applied, namely the trainable parameters are the slope and location].
Claims 14
A system of filtering a dataset configured to execute the method according to claim 1 [A system to execute the method are components recited at a high level are construed as generic computer components and algorithms used to implement the abstract idea. See MPEP 2106.05(f)(2). As such, the limitations do not integrate the abstract idea into a practical application. Nor to do they amount to significantly more
Claims 15
A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of claim 1 [A computer executing a computer program comprising instruction to carry out the method are components recited at a high level are construed as generic computer components and algorithms used to implement the abstract idea. See MPEP 2106.05(f)(2). As such, the limitations do not integrate the abstract idea into a practical application. Nor to do they amount to significantly more.]
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1-12 and 14-15 are rejected under 35 U.S.C. 102(a)(l) as being anticipated by Importance Weighted Active Learning to Beygelzimer et al. (hereinafter Beygelzimer).
Per claim 1, Beygelzimer discloses A computerized method of filtering a dataset for processing, wherein the dataset comprises a plurality of data records, the method comprising [Beygelzimer, pg. 1 “We present a practical and statistically consistent scheme for actively learning binary classifiers under general loss functions.”; pg. 3 “We consider active learning in the streaming setting where at each step t, a learner observes an unlabeled point. xt
∈
X and has to decide whether to ask for the label yt
∈
Y”. (note: the algorithm operates on a stream (dataset) of unlabeled points xt drawn from a distribution D (data records), and performs a filtering method. The computerized filtering method of a dataset (stream of unlabeled data points) for the purpose of processing (classification).)]:
determining, by an estimation module, selection estimation values for the data records [Beygelzimer, pg. 4 “Upon seeing xt, the learner calls a subroutine rejection-threshold (instantiated in later sections), which looks at xt and past history to return the probability pt of requesting yt”; pg. 5 “The subroutine maintains an effective hypothesis class Ht, which is initially all of H and then gradually shrinks…”. (note: the subroutine rejection-threshold is the estimation module, and the probability pt it returns is the selection estimation value for each data record xt. The subroutine uses the existing hypothesis set Ht for this computation.)];
determining, by a pass-through function, pass-through probabilities for the data records based on the selection estimation values [Beygelzimer, pg. 2 “Our strategy, roughly, is to make it proportional to the spread of values h(xt), as h ranges over the remaining candidate hypotheses”. (note: this identifies the selection estimation value as a distinct quantity computed before pt is determined. The spread of values h(xt) is the range of predictions made by the current hypothesis set Ht on the data record xt. This spread is a property of the data record xt computed by the estimation module (the maintained hypothesis set Ht). It is not itself a probability. It is a real-valued score that characterizes how much disagreement exists about the record. This is the selection estimation value. It is the output of the estimation module, and it is distinct from and prior to the pass-through probability.); pg. 6 “More precisely,
p
t
=
m
a
x
f
,
g
∈
H
t
max
l
f
x
t
,
y
-
l
g
x
t
,
y
.
y
”. (note: this formula is the pass-through function. It takes the spread of loss values across hypotheses in Ht for the data record xt (selection estimation value) and maps it to a probability pt. This formula is the pass-through function and the selection estimation value is the spread it operates on. These are two distinct stages. First, the estimation module maintains Ht and the spread of h(xt) is the estimation value. Second, the pass-through function computes pt from that spread via the max formula.); pg. 4 “Upon seeing xt, the learner calls a subroutine rejection-threshold (instantiated in later sections), which looks at xt and past history to return the probability pt of requesting yt”. (note: the rejection-threshold subroutine is the pass-through function. It is a distinct functional step that takes as input xt and past history (which encodes the estimation module’s maintained state Ht and therefore the selection estimation value) and returns the pass-through probability pt. The subroutine’s input includes the estimation module’s accumulated information and its output is the probability pt.)];
generating a subset of data records by discarding at least a portion of the dataset based on the pass-through probabilities [Beygelzimer, pg. 4 “Flip a coin Qt ∈ {0, 1} with E[Qt] = pt. If Qt = 1, request yt and set St = St−1 ∪ {(xt, yt, 1/pt)}, else St = St−1.”. (note: data records for which Qt = 0 are discarded, only those with Qt = 1 enter the subset St. This is generating a subset by discarding data records based on pass-through probabilities (pt))];
processing, by a selection module, the subset of data records, wherein the selection module selects one or more data records of the subset of data records [Beygelzimer, pg. 4 “Let ht = arg minh∈H ∑(x,y,c)∈ St c · l(h(x), y)”. (note: the algorithm processes the subset St (the labeled weighted records) and selects the best hypothesis); pg. 4 “Qt = 1, request yt”. (note: the querying step is the selection module selecting which records to process and label)];
assigning weights and labels to the data records of the subset of data records, wherein the weights reflect a data distribution of the data records with respect to the subset of data records, and wherein the labels represent a selection of the data records by the selection module [Beygelzimer, pg. 2 “The points that end up getting labeled are then weighted according to the reciprocals of these probabilities (that is, 1/pt), in order to remove sampling bias.”. (note: this shows assigning weights (1/pt) to the data records of the subset. The weights 1/pt correct for the fact that the subset was drawn by selection probabilities pt and non-uniformly. Therefore, each record’s weight reflects its relative underrepresentation in the subset compared to the full dataset. This is what is claimed, the weights correct the distributional mismatch between the full dataset and the selected subset.); pg. 4 “If Qt = 1, request yt and set St = St−1 ∪ {(xt, yt, 1/pt)}, else St = St−1.”. (note: the binary variable Qt is the selection decision, Qt = 1 means that data record xt is selected (selection module) and then the label yt is requested and assigned to that record. The selected record is then added to the subset St with both its label yt and its weight 1/pt. The selection module is what flips Qt and when Qt = 1, selects the data record xt and assigns it its label yt. The label yt is only assigned and present in St when the selection module chose that record (Qt = 1)); pg. 4 “The algorithm maintains a set of labeled examples seen so far, each with an importance weight: if yt ends up being queried, its weight is set to 1/pt”. (note: the data records of the subset are the labeled examples, they have been selected (queried) and have received a label yt. Each record has been assigned both a label (yt, which is assigned upon selection) and a weight (1/pt, reflecting the distribution).’ If yt ends up being queried’ shows that label assignment is connected to the selection event (Qt = 1), which shows that the label represents the selection by the selection module. The weight 1/pt is the distributional relationship between the selected subset and the full dataset.)];
updating the estimation module and the pass-through function based on the subset of data records including the weights and labels [Beygelzimer, pg. 4 “Let ht = arg minh∈H ∑(x,y,c)∈ St c · l(h(x), y)”. (note: the hypothesis ht is updated at each step based on the weighted labeled subset St. The rejection-threshold subroutine that generates pt at the next step uses this updated hypothesis ht.); pg. 5 “The subroutine maintains an effective hypothesis class Ht, which is initially all of H and then gradually shrinks by setting Ht+1 to the subset of Ht”. (note: the hypothesis set (estimation module) and the rejection threshold (pass-through function) are both updated based on the weighted labeled subset)].
Per claim 2, Beygelzimer discloses claim 1, further disclosing the weights are determined on the pass-through probabilities [Beygelzimer, pg. 4 “if yt ends up being queried, its weight is set to 1/pt”. (note: the weight (1/pt) is directly determined by the pass-through probability pt)].
Per claim 3, Beygelzimer discloses claim 1, further disclosing the pass-through function is a trainable function, wherein updating the pass-through function comprises retraining the pass-through function based on the subset of data records including the weights and labels [Beygelzimer, pg. 2 “…this scheme queries its label with a carefully chosen probability pt, taking into account the identity of the point and the history of labels seen so far.”; pg. 4“Set pt = rejection-threshold (xt, {xi, yi, pi, Qi : 1 ≤ i < t})”; pg. 4 “Let ht = arg minh∈H ∑(x,y,c)∈ St c · l(h(x), y)” (note: the rejection-threshold subroutine is a function that is retrained based on the growing labeled set.); pg. 5 “The subroutine maintains an effective hypothesis class Ht, which is initially all of H and then gradually shrinks by setting Ht+1 to the subset of Ht”. (note: the pass-through function is updated at each step based on weighted labeled examples)].
Per claim 4, Beygelzimer discloses claim 1, further disclosing the estimation module is a machine learning routine, wherein updating the estimation module comprises retraining the estimation module based on the subset of data records including the weights and labels [Beygelzimer, pg. 3 “The learner works with a hypothesis space H = {h : X → Z}, where Z is a prediction space. The algorithm is evaluated with respect to a given loss function l : Z × Y → [0,∞).”. (note: this shows a machine learning routine consisting of a model selector operating over a hypothesis space H using a loss function to make predictions); pg. 4 “3. Flip a coin Qt ∈ {0, 1} with E[Qt] = pt. If Qt = 1, request yt and set St = St−1 ∪ {(xt, yt, 1/pt)}, else St = St−1.”, “4. Let ht = arg minh∈H ∑(x,y,c)∈ St c · l(h(x), y)” . (note: step 3 defines the generation of a data subset St consisting of the data record xt , the label yt , and the inverse-probability importance weight (1/pt). Step 4 performs the machine learning retraining loop by optimizing the objective function across the subset St containing the weights and labels)].
Per claim 5, Beygelzimer discloses claim 1, further disclosing the estimation module and the pass-through function are periodically updated [Beygelzimer, pg. 4 “For t from 1, 2, . . . until the data stream runs out:
1. Receive xt .
2. Set pt = rejection-threshold (xt, {xi, yi, pi, Qi : 1 ≤ i < t}).
3. Flip a coin Qt ∈ {0, 1} with E[Qt] = pt. If Qt = 1, request yt and set St = St−1 ∪ {(xt, yt, 1/pt)}, else St = St−1
4. Let ht = arg minh∈H ∑(x,y,c)∈ St c · l(h(x), y)”.
(note: the estimations module (ht) and pass-through function (rejection-threshold producing pt) are updated periodically (at every step t))].
Per claim 6, Beygelzimer discloses claim 1, further disclosing the data records are preprocessed before determining the selection estimation values, wherein preprocessing comprises at least one of merging a data record with additional information from a database, reducing the amount of data in a data record, and adding further information determined based on the data record to the data record [Beygelzimer, pg. 3 “We consider active learning in the streaming setting where at each step t, a learner observes an unlabeled point. xt
∈
X…”; pg. 14 “We used PCA to reduce the dimension from 784 to 25”. (note: the input space X includes pre-processed feature vectors. PCA is used for dimensionality reduction as preprocessing. This is reducing the amount of data in a data record.)].
Per claim 7, Beygelzimer discloses claim 1, further disclosing the data records comprise fields and values [Beygelzimer, pg. 3 “Let X be the input space and Y the output space…unlabeled point xt ∈ X”. (note: each data record (unlabeled point xt ∈ X) is a vector of features (fields and values). This is the standard representation of structured data records in machine learning.].
Per claim 8, Beygelzimer discloses claim 1, further disclosing wherein the estimation module is a trained predictive model [Beygelzimer, pg. 14 “We implemented IWAL with loss-weighting for linear separators under logistic loss.”. (note: the logistic loss linear separator is a trained predictive model); pg. 4 “Let ht = arg minh∈H ∑(x,y,c)∈ St c · l(h(x), y)”.”. (note: ht is a trained predictive model/estimation module. )].
Per claim 9, Beygelzimer discloses claim 1, further disclosing the estimation module is a gradient boosted decision tree or a logistic regression model [Beygelzimer, pg. 14 “We implemented IWAL with loss-weighting for linear separators under logistic loss.”. (note: a linear separator under logistic loss is a logistic regression model)].
Per claim 10, Beygelzimer discloses claim 1, further disclosing the pass-through function is a monotonically increasing function [Beygelzimer, pg. 2 “…this scheme queries its label with a carefully chosen probability pt, taking into account the identity of the point and the history of labels seen so far.”; pg. 4 “…as long as pt is bounded away from 0…”. (note: the rejection-threshold (pass-through function) maps increasing uncertainty scores to increasing probabilities pt, which is a monotonically increasing function); pg. 2 “Our strategy, roughly, is to make it proportional to the spread of values h(xt), as h ranges over the remaining candidate hypotheses”. (note: higher spread (higher estimation value) leads to higher pt, which is monotonically increasing)].
Per claim 11, Beygelzimer discloses claim 1, further disclosing the pass-through probabilities are non-zero probabilities above a threshold [Beygelzimer, pg. 4 “We prove that IWAL algorithms are consistent, as long as pt is bounded away from 0”; pg. 5 “…if there is a constant pmin > 0 such that pt ≥ pmin for all 1 ≤ t ≤ T” (note: the pass-through probabilities are required to be non-zero and above a minimum threshold pmin)].
Per claim 12, Beygelzimer discloses claim 1, further disclosing the pass-through function comprises two or more trainable parameters [Beygelzimer, pg. 5 “Algorithm 2 gives a particular instantiation of the rejection threshold subroutine in IWAL. The subroutine maintains an effective hypothesis class Ht…”. (note: the rejection-threshold function is parameterized by multiple parameters of the hypothesis class Ht (which is a function class, having at least two trainable parameters); pg. 14 “We implemented IWAL with loss-weighting for linear separators under logistic loss… the algorithm involves two convex optimizations as subroutines”. (note: this shows that the pass-through subroutine involves optimizing at least two trainable parameters)].
Per claim 14, Beygelzimer discloses claim 1, further disclosing A system of filtering a dataset configured to execute the method according to claim 1 [Beygelzime, pg. 3, “We conduct practical experiments with two IWAL algorithms”; pg. 14 “We implemented IWAL with loss-weighting for linear separators under logistic loss”. (note: this is a computer implemented system executing the method)].
Per claim 15, Beygelzimer discloses claim 1, further disclosing A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of claim 1 [Beygelzime, pg. 3, “We conduct practical experiments with two IWAL algorithms”; pg. 14 “We implemented IWAL with loss-weighting for linear separators under logistic loss”. (note: Beygelzime describes a computer program carrying out the IWAL algorithm)].
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.
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.
Claim 13 is rejected under 35 U.S.C. 103 as being unpatentable over Beygelzimer in view of Probabilistic Outputs for Support Vector Machines and Comparisons to Regularized Likelihood Methods to Platt et al. (hereinafter Platt).
Per claim 13, Beygelzimer discloses claim 12.
Beygelzimer does not fully disclose, but with Platt does teach two trainable parameters of the two or more trainable parameters [Platt, pg. 3
“
P
y
=
1
f
=
1
1
+
e
x
p
(
A
f
+
B
)
”. (note: this sigmoid function takes a real valued score, the output of a classifer (selection estimation value) and maps it to a probability between 0 and 1. As f increases from negative to positive values, P transitions from near 0 (low pass-through probability) to near 1 (high pass-through probability); pg. 3 “This sigmoid model… has two parameters trained discriminatively….”. (note: this shows the model has two trainable parameters); pg. 4 “The parameters A and B of (9) are fit using maximum likelihood estimation from a training set (fi, yi).”. (note: this shows both A (slope) and B (location) are trainable parameters. They are determined by optimizing on a training set.)] are the slope of the transition from low to high pass-through probabilities [Platt, pg. 3 “The posterior probability rule P(y = 1|f) is thus a sigmoid, whose slope is determined by the tied variance.”. (note: this shows that a parameter of a sigmoid controls the slope. This identifies the slope of P(y = 1|f) as the property controlled by a sigmoid parameter.); pg. 4 “As long as A < 0, the monotonicity of (9) is assured”. (note: parameter A dictates the monotonic behavior of the sigmoid transition, it is A that determines the direction and steepness of the transition from low to high probability. This shows A is the slope-controlling parameter.)] and the location of the transition [Platt, pg. 3 “Hastie and Tibshirani [7] then adjust the bias of the sigmoid so that the point P(y = l|f) = 0.5 occurs at f = 0.”. (note: adjust the bias of the sigmoid describes shifting the location of the transition point. In Platt’s two parameter model, the location is controlled by parameter B, the inflection point occurs at f = -B/A. Changing B shifts the transition left or right along the f axis.)].
Beygelzime and Platt are analogous art because they are from the same field of endeavor of machine learning probability estimation from classifier scores.
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to implement the rejection-threshold function using the two-parameter sigmoid formula as taught by Platt.
The suggestion/motivation for using a trainable sigmoid with two parameters is the sigmoid shape naturally expresses the transition from low to high probabilities as a function of a classifier score and is the effective approach for any system doing so. This demonstrates a simple substitution of one known element for another to obtain predictable results having reasonable-expectation-of-success (MPEP 2143.01(b)), namely substituting Platt’s two-parameter trainable sigmoid for Beygelzimer’s score-to-probability rejection-threshold pass-through function [Platt, pg. 1 “Instead, we train an SVM, then train the parameters of an additional sigmoid function to map the SVM outputs into probabilities.”].
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Sayed M Shah whose telephone number is (571)272-9406. The examiner can normally be reached Monday-Friday 6:00 am - 2:00 pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Miranda Huang can be reached at (571) 270-7092. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/SAYED MUNEER SHAH/Examiner, Art Unit 2124
/Kevin W Figueroa/Primary Examiner, Art Unit 2124
Prosecution Insights