DETAILED ACTION
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 § 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)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 12-13 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Myung Eun Lim et al. (hereinafter Lim) (US 20230297895 A1, 2023-09-21).
Regarding claim 12, Lim teaches;
providing an input to a plurality of prediction models; obtaining, for the input, a prediction from each of the plurality of prediction models;
([Abstract] collecting prediction values for input data of each of the prediction models)
determining a weight for each prediction from the plurality of prediction models to generate a plurality of weights;
([Abstract] calculating a model weight of each of the prediction models using a pre-trained ensemble model that uses the prediction value as an input)
generating a training dataset comprising the input labeled with the plurality of weights; training a weight model using the training data set to generate a plurality of output weights;
([0051] As illustrated in FIGS. 7A and 7B, the ensemble model is configured to receive prediction values and time series records of each prediction model and output model weights. In the situation where the model weights of the training data are pre-calculated based on the error between the prediction value of the prediction model and the actual observation value, the ensemble model may be trained through backpropagation optimization to output the pre-calculated weights of the training data.)
NOTE: Lim teaches that the training data for the ensemble / weight model can include the input (time-series records), that prediction values are collected for those inputs, and that the aforementioned weights are calculated based on the predictions and the actual observed value.
The ensemble model is then trained to output the weights, using the training data with the pre-calculated weights. Under the broadest reasonable interpretation, the pre-calculated weights function as target labels generated for the training dataset.
Thus, under the broadest reasonable interpretation, Lim teaches generating a training dataset comprising the input labeled with the plurality of weights, and training a weight model (the ensemble model) using the training data set to generate a plurality of output weights.
and determining a plurality of weighted predictions by weighting each of a plurality of initial predictions with a respective weight in the plurality of output weights.
([0042] Here, according to embodiments of the present disclosure, more accurate ensemble prediction may be performed by calculating a model weight using a machine learning ensemble model and applying an optimal model weight determined in a process of determining an optimal combination using the model weight.)
NOTE: The optimal model weights are determined from the aforementioned model weights output by the ensemble / weight model.
([0075] For example, as illustrated in FIG. 8, since the optimal model weights of prediction models A, B, C, and D are {0.0, 0.3, 0.7, 0.0}, when the prediction values of prediction models A, B, C, and D for the input data are {35, 41, 46, 17}, the ensemble prediction value of the input data may be calculated as 44.5 (=41×0.3+46×0.7))
NOTE: Lim teaches determining a plurality of weighted predictions (41×0.3+46×0.7) by weighting each of a plurality of initial predictions ({35, 41, 46, 17}) with a respective weight in the plurality of output weights ({0.0, 0.3, 0.7, 0.0}).
Regarding claim 13, Lim teaches;
The method of claim 12, wherein determining the weight for each prediction comprises determining the weight based upon a predetermined correct prediction for the input and a respective predictions from each of the plurality of prediction models.
([0050] Then, the ensemble model receives the prediction value and time series data of the prediction model to calculate a model weight of each prediction model… the model weight is a score indicating the accuracy, importance, or the like of the prediction model for each prediction value, and a higher model weight means that the prediction value of the prediction model is closer to the correct answer.)
NOTE: Lim indicates that the aforementioned weights for each prediction of the aforementioned plurality of prediction models are determined based upon how close the predictions for the input are to the correct answer.
Thus, Lim teaches wherein determining the weight for each prediction comprises determining the weight based upon a predetermined correct prediction / answer for the input and a respective prediction from each of the plurality of prediction models.
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.
Claim(s) 1-3, 5, 21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Michael I. Jordan et al. (hereinafter Jordan) (“Hierarchical mixtures of experts and the EM algorithm”, 1993) in view of Rafael M. O. Cruz et al. (hereinafter Cruz) (“META-DES.H: a dynamic ensemble selection technique using meta-learning and a dynamic weighting approach”, 2018).
Regarding claim 1, Jordan teaches;
a method comprising: providing input to a plurality of prediction models; obtaining an initial prediction from each of the plurality of prediction models;
([pg. 1340] Expert network (i, j) produces its output μ_ij as a generalized linear function of the input x: μ_ij = f(U_ij x) ... For regression problems, f(.) is the identity function ... For binary classification problems, f(.) is generally taken to be the logistic
function, ... Other models (e.g., multiway classification, counting, rate estimation and survival estimation) are handled readily by making other choices for f(.).)
NOTE: Jordan teaches providing input (x) to a plurality of prediction models (expert networks) obtaining an initial prediction from each of the plurality of prediction models (each expert network produces output, which can be regression output, classification results, or other kinds of predictions).
providing the input to a plurality of weight models;
([pg. 1340] The gating networks at the lower level are defined similarly, yielding outputs g_jli)
PNG
media_image1.png
232
725
media_image1.png
Greyscale
NOTE: Jordan teaches that the gating networks produce outputs g_j|i, which are used to weight the predictions of the expert networks, μ_ij. Thus, the gating networks can be considered weight models.
[pg. 1339]
PNG
media_image2.png
293
718
media_image2.png
Greyscale
NOTE: Jordan teaches providing the input x to a plurality of weight models (gating networks) as pictured.
obtaining from the plurality of weight models a weight for each initial prediction,
[pg. 1339]
PNG
media_image2.png
293
718
media_image2.png
Greyscale
NOTE: Jordan teaches obtaining from the plurality of weight models (gating networks) a weight for each initial prediction, because the gating networks generate the aforementioned gates g_j|i used to weight each of the initial predictions, μ_ij.
wherein the weight for each initial prediction is based upon the input
([pg. 1340] Note that both the g’s and the p’s depend on the input x, thus the total output is a nonlinear function of the input.)
NOTE: Jordan teaches that the weight for each initial prediction (g_j|i) are based upon the input x.
and determining a plurality of weighted predictions by weighting each initial prediction with a respective weight for the initial prediction.
[pg. 1340]
PNG
media_image1.png
232
725
media_image1.png
Greyscale
NOTE: Jordan weights each of the aforementioned initial predictions μ_ij with a respective weight for the initial prediction, g_j|i. Thus, Jordan teaches determining a plurality of weighted predictions by weighting each initial prediction with a respective weight for the initial prediction.
Jordan fails to teach but Cruz teaches;
wherein the weight for each initial prediction is based upon the input and behavior of each of the plurality of prediction models;
([Abstract] train a meta-classifier that is able to predict whether or not a base classifier from the pool is competent enough to classify an input instance… Dynamic weighting, where the meta-classifier estimates the competence of each classifier in the pool, and the outputs of all classifiers in the pool are weighted based on their level of competence)
NOTE: Cruz teaches a meta-classifier, which determines the competence of the classifiers based on an input instance. Cruz teaches assigning a weight to the output of each of the classifiers, based on the competence of the plurality of classifiers.
Thus, Cruz teaches the weight for each initial prediction / classifier output is based upon the input and behavior of each of the plurality of prediction models, because the weights are assigned to the classifier outputs based on the competence / behavior of the classifier, which is determined based on the input instances.
OBVIOUSNESS TO COMBINE CRUZ WITH JORDAN:
Jordan and Cruz are both analogous art to the present disclosure as they pertain to generating weighted predictions for a plurality of prediction models.
Jordan provides the base of weighting predictions of a plurality of prediction models using a plurality of weight models,
while Cruz provides a method for determining weights for predictions based on both the behavior of the plurality of prediction models and the input.
Cruz further states;
([pg. 2] In the dynamic weighting approach, the meta-classifier is used to estimate the weights of all base classifiers in the pool. Then, their decisions are aggregated using a weighted majority voting scheme [2]. Thus, classifiers that attain a higher level of competence, for the classification of the given query sample, have a greater impact on the final decision.)
NOTE: Cruz teaches that assigning weights based on classifier behavior / competence beneficially allows classifiers having a higher estimated competence for a given sample to have a greater impact on the final decision, thereby improving performance.
Therefore, 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 gating weights of Jordan to be additionally based on the competence / behavior measures taught by Cruz, so that the final weighted prediction gives greater influence to the expert / prediction models that are most competent for the current input, to improve prediction accuracy.
Regarding claim 2, Jordan teaches;
The method of claim 1, wherein each of the plurality of prediction models comprises a machine learning model.
[pg. 1343]
PNG
media_image3.png
14
681
media_image3.png
Greyscale
NOTE: Expert models have learnable parameters.
([pg. 1340] Expert network (i, j) produces its output μ_ij as a generalized linear function of the input x: μ_ij = f(U_ij x) ... For regression problems, f(.) is the identity function ... For binary classification problems, f(.) is generally taken to be the logistic
function, ... Other models (e.g., multiway classification, counting, rate estimation and survival estimation) are handled readily by making other choices for f(.).)
NOTE: Jordan teaches that each expert network can be a classifier, regression model, etc., with learnable parameters i.e., each expert network is a machine learning model.
Regarding claim 3, Jordan teaches;
The method of claim 1, wherein the plurality of weight models comprises a machine learning model.
[pg. 1343]
PNG
media_image3.png
14
681
media_image3.png
Greyscale
NOTE: The gating networks have learnable parameters.
[pg. 1340]
PNG
media_image4.png
125
724
media_image4.png
Greyscale
PNG
media_image5.png
109
542
media_image5.png
Greyscale
NOTE: Jordan teaches the plurality of gating networks being generalized linear models each having a set of learnable parameters. Thus, Jordan teaches that the plurality of weight models (gating networks) comprises a machine learning model.
Regarding claim 5, Jordan teaches;
The method of claim 1, further comprising providing an output that comprises all of the plurality of weighted predictions.
[pg. 1340]
PNG
media_image6.png
273
543
media_image6.png
Greyscale
NOTE: Jordan teaches providing a final output at the top level which is a summation of the weighted predictions of the of the experts. Thus, Jordan teaches providing an output (μ) that comprises all of the plurality of weighted predictions (μ is the combination of all the weighted predictions, g_j|i * μ_ij).
Regarding claim 21, Jordan teaches;
wherein each of the plurality of weight models is trained based on a training dataset.
[pg. 1342]
PNG
media_image7.png
380
426
media_image7.png
Greyscale
NOTE: Jordan teaches each of the plurality of the gating networks being trained by solving an IRLS problem using a set of observations, where the set of observations can be considered a training dataset. Thus, Jordan teaches that each of the plurality of weight models (lower-level gating networks) is trained based on a training dataset (the set of observations).
Claim(s) 6, 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jordan (“Hierarchical mixtures of experts and the EM algorithm”, 1993) in view of Cruz (“META-DES.H: a dynamic ensemble selection technique using meta-learning and a dynamic weighting approach”, 2018) as applied to claim 1 above, and further in view of Lim (US 20230297895 A1, 2023-09-21).
Regarding claim 6, Jordan and Cruz fail to teach but Lim teaches;
The method of claim 1, further comprising providing an output that comprises one of the plurality of weighted predictions.
([0004] In regression analysis ensemble prediction for predicting numerical values, as a method of determining ensemble prediction results by applying weights to a base model, a best selection method and a weighted sum method may be applied... The best selection method is a method of selecting a prediction result of a base model having a highest model weight as the ensemble prediction result.)
NOTE: Lim teaches a method of selecting a single prediction result as the prediction having the highest associated weight.
Thus, Lim teaches providing an output that comprises one of the plurality of weighted predictions.
OBVIOUSNESS TO COMBINE LIM WITH JORDAN AND CRUZ:
Lim is analogous art to the present disclosure as it pertains to an ensemble method using weighted predictions to generate a final prediction.
Jordan provides a base of generating a plurality of weighted predictions, Cruz provides a means of determining prediction weights based on model performance, while Lim provides a means for outputting a one prediction of a plurality of weighted predictions based on the associated weight.
Additionally, Lim states;
([0004] The best selection method may be advantageous in performance because it may exclude prediction results having large errors when the prediction accuracy of each model weight is high.)
NOTE: Lim teaches that selecting a single prediction having the highest weight as the output prediction improves the prediction performance by excluding less favorable predictions.
Therefore, 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 ensemble method of Jordan using prediction weights based both on input and model behavior (as taught by Cruz) to implement the process disclosed by Lim to select the prediction having the highest weight as output, to improve prediction performance by excluding less favorable predictions.
Regarding claim 22, Jordan and Cruz fail to teach but Lim teaches;
wherein the training dataset comprises one or more training inputs each labeled with a weight value.
([0051] As illustrated in FIGS. 7A and 7B, the ensemble model is configured to receive prediction values and time series records of each prediction model and output model weights. In the situation where the model weights of the training data are pre-calculated based on the error between the prediction value of the prediction model and the actual observation value, the ensemble model may be trained through backpropagation optimization to output the pre-calculated weights of the training data.)
NOTE: Lim teaches an ensemble method that outputs weights associated with model predictions, which can be considered a weight model.
The ensemble model is trained using a set of time series input data and prediction values to output weights associated with model predictions. In the training process, the weights are pre-calculated as the target outputs / labels of the training data for the ensemble model.
Thus, Lim teaches training a weight / ensemble model using a training dataset comprising one or more training inputs each labeled with a weight value.
OBVIOUSNESS:
Jordan teaches the base of training a plurality of weight models using an input dataset, Lim teaches training a weight model using a training dataset having weight values associated as target outputs / labels.
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the error-based weight targets of Lim to train the gating networks of Jordan, because doing so uses the known actual training outputs to teach the gates which expert predictions are accurate, thereby improving the gates ability to assign higher weights to important experts when generating weighted predictions, which would improve the accuracy of the predictions generated by the system.
Claim(s) 7-9 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jordan (“Hierarchical mixtures of experts and the EM algorithm”, 1993) in view of Cruz (“META-DES.H: a dynamic ensemble selection technique using meta-learning and a dynamic weighting approach”, 2018) as applied to claim 1 above, and further in view of Reiner Wilhelms-Tricarico et al. (hereinafter Wilhelms-Tricarico) (US 20130262096 A1, 2013-10-03).
Regarding claim 7, Jordan and Cruz fail to teach but Wilhelms-Tricarico teaches;
The method of claim 1, wherein the input comprises features extracted from text.
([0077] In a further example of applying hierarchical mixture of experts methods, the input data include both binary and continuous variables. The binary variables can be stored in a binary feature vector. The features can be obtained by specific rules that can be applied to the input text by computational processing in the front-end, for example, as described elsewhere herein.)
NOTE: Wilhelms-Tricarico teaches input for a hierarchical mixture of experts method comprising features extracted from text.
OBVIOUSNESS TO COMBINE WILHELMS-TRICARICO WITH JORDAN AND CRUZ: Wilhelms-Tricarico is analogous art to the present disclosure as it incorporates a hierarchical mixture of experts method, which utilizes a plurality of prediction models whose predictions are weighted by at least a weight model.
Jordan provides the base hierarchical mixture-of-experts model, while Wilhelms-Tricarico uses features of text data as input to a hierarchical mixture-of-experts model.
Wilhelms-Tricarico additionally states;
([0079] By analyzing both the texts with the front-end and the recorded signal by signal processing methods, the data then can be stored as a sequence of pairs (xi, yi) of data, which can be subsequently used in training the hierarchical mixture of experts model. In the hierarchical mixture of expert models, the basic building blocks can be experts which transform the input variable vectors x, a mixed binary and continuous feature vector, into a vector of acoustic features y… While each expert in the model provides one simple regression model of the data, specialization of the multiple experts according to a partitioning of the data domain can give the model useful advantages compared, for example, with a simple regression model.)
NOTE: This excerpt teaches that text-derived input creates a feature domain that a hierarchical mixture of experts can partition, letting different experts specialize on different text-feature patterns, to obtain more relevant expert outputs.
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to apply the Hierarchical mixture-of-experts model of Jordan to the text-derived feature vectors of Wilhelms-Tricarico to allow different experts to specialize in different regions of the text-feature input space, thereby improving the text-based prediction relative to a single model.
Regarding claim 8, Wilhelms-Tricarico teaches;
The method of claim 7, wherein the text is derived from human speech.
([0159] The invention includes computerized systems, or machines, for ... for recognizing utterances of human speech and rendering the utterances as text or other graphic characters, and computerized systems, or machines for other purposes relevant to the objects of the invention.)
NOTE: Wilhelms-Tricarico teaches text derived from human speech because they include a means for rendering utterances of human speech as text.
OBVIOUSNESS:
Using the same reasoning from claim 7, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine Wilhelms-Tricarico with Jordan and Cruz.
Regarding claim 9, Wilhelms-Tricarico teaches;
The method of claim 7, further comprising determining an overall prediction from a plurality of output predictions determined from different features extracted from the text.
([0077] In a further example of applying hierarchical mixture of experts methods, the input data include both binary and continuous variables. The binary variables can be stored in a binary feature vector. The features can be obtained by specific rules that can be applied to the input text)
NOTE: Wilhelms-Tricarico teaches input to the mixture of experts being features extracted from the input text. Additionally, ‘features’ is plural, indicating multiple different text derived features.
([0079] The predictions from each expert can be combined by means of a gating network. The gating network can compute from the input vector x a probability weighting for each expert's output, using a soft-max function. Using these weights, a weighted combination of the individual experts' outputs summing to the actual prediction of the group of experts can be computed.)
NOTE: Wilhelms-Tricarico teaches determining an overall prediction by summing the plurality of expert output predictions of the mixture of experts.
As previously taught, the mixture of experts method uses input data comprising different features extracted from text. From this, the plurality of expert output predictions are determined from different features extracted from the text.
Thus, Wilhelms-Tricarico teaches determining an overall prediction from a plurality of output predictions determined from different features extracted from the text.
OBVIOUSNESS:
Using the same reasoning from claim 7, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine Wilhelms-Tricarico with Jordan and Cruz.
Claim(s) 10-11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jordan (“Hierarchical mixtures of experts and the EM algorithm”, 1993) in view of Cruz (“META-DES.H: a dynamic ensemble selection technique using meta-learning and a dynamic weighting approach”, 2018) as applied to claim 1 above, and further in view of Christopher Lesner et al. (hereinafter Lesner) (US 20210406780 A1, 2021-12-30).
Regarding claim 10, Jordan and Cruz fail to teach but Lesner teaches;
The method of claim 1, wherein each initial prediction comprises a prediction class and a probability for the prediction class.
([0030] Examples of predictor MLMs (116) include any multiclass classification model that can output class probabilities)
NOTE: Teaches a predictor machine learning model (MLM) where a prediction comprises a prediction class and a probability for the prediction class, because the prediction of the predictor MLM includes at least a class with an associated probability.
OBVIOUSNESS TO COMBINE LESNER WITH JORDAN AND CRUZ:
Lesner is analogous art to the present disclosure as it pertains to an ensemble of machine learning models that generate classification outputs.
Jordan provides a base for weighting predictions of a plurality of prediction models (experts) using a plurality of weight models (gating networks).
Cruz provides a technique for incorporating the behavior/competence of a plurality of classifier models into the determination of the weights generated for model outputs.
Lesner provides a means for including a probability associated with classification predictions, which can provide more information regarding the behavior/competence of the models as opposed to a hard classification.
Additionally, Lesner states;
([0032] the input of a confidence MLM is the probabilities associated with the various classifications as determined by the predictor MLM, and the output of the confidence MLM is one or more probabilities that the prediction probabilities made by the corresponding predictor MLM were correct.)
NOTE: Lesner teaches that the output classifications and associated probabilities are useful in determining the competence / accuracy of the associated prediction model (predictor MLM).
Thus, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Jordans prediction models (experts) to use Lesner’s class-probability outputs as the predictions, to provide additional information regarding the competence / accuracy of the associated prediction model, thereby improving the accuracy of Cruz’s competence-based prediction weighting technique.
Regarding claim 11, Jordan and Cruz fail to teach but Lesner teaches;
The method of claim 10, wherein the probability is based upon the behavior of one of the plurality of prediction models and the input.
([0030] A predictor MLM is a MLM configured to take, as input, the data items in the training data set (102), or the data items in a subset of the training data set (102). The predictor MLM is configured to produce, as output, the predicted classifications of the data items. The output may take the form of all of the classifications with associated probabilities of classification.)
NOTE: Lesner teaches each of the predictor machine learning models (MLM) processing input to output the aforementioned classification and associated probability. Thus, Lesner teaches that the probability is based upon the behavior of one of the plurality of prediction models (predictor MLMs) and the input.
OBVIOUSNESS:
Using the same reasoning from claim 10, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Jordans prediction models (experts) to use Lesner’s class-probability outputs as the predictions, to provide additional information regarding the competence / accuracy of the associated prediction model, thereby improving the accuracy of Cruz’s competence-based prediction weighting technique.
Claim(s) 14-15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lim (US 20230297895 A1, 2023-09-21) as applied to claim 12 above, and further in view of Wilhelms-Tricarico (US 20130262096 A1, 2013-10-03).
Regarding claim 14, Lim fails to teach but Wilhelms-Tricarico teaches;
The method of claim 12, wherein the input comprises features extracted from text.
([0077] In a further example of applying hierarchical mixture of experts methods, the input data include both binary and continuous variables. The binary variables can be stored in a binary feature vector. The features can be obtained by specific rules that can be applied to the input text by computational processing in the front-end, for example, as described elsewhere herein.)
NOTE: Wilhelms-Tricarico teaches a mixture of experts framework comprising a plurality of prediction models (experts) where the input comprises feature extracted from text.
OBVIOUSNESS TO COMBINE WILHELMS-TRICARICO WITH LIM:
Lim and and Wilhelms-Tricarico are analogous art to the present disclosure as they pertain to ensemble methods including a plurality of prediction models whose predictions are weighted by a weight model.
Lim provides the base plurality of prediction models and training of the weight model, while Wilhelms-Tricarico uses features of text data as input to a plurality of prediction models.
Wilhelms-Tricarico additionally states;
([0079] By analyzing both the texts with the front-end and the recorded signal by signal processing methods, the data then can be stored as a sequence of pairs (xi, yi) of data, which can be subsequently used in training the hierarchical mixture of experts model. In the hierarchical mixture of expert models, the basic building blocks can be experts which transform the input variable vectors x, a mixed binary and continuous feature vector, into a vector of acoustic features y… While each expert in the model provides one simple regression model of the data, specialization of the multiple experts according to a partitioning of the data domain can give the model useful advantages compared, for example, with a simple regression model.)
NOTE: This excerpt teaches that text-derived input creates a feature domain that a hierarchical mixture of experts can partition, letting different experts specialize on different text-feature patterns, to obtain more relevant expert outputs.
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to apply the method of Lim using the text-derived feature vectors of Wilhelms-Tricarico as input to allow different experts to specialize in different regions of the text-feature input space, thereby improving the prediction performance relative to a single model.
Regarding claim 15, Lim fails to teach but Wilhelms-Tricarico teaches;
The method of claim 14, wherein the text comprises text derived from human speech.
([0159] The invention includes computerized systems, or machines, for ... for recognizing utterances of human speech and rendering the utterances as text or other graphic characters, and computerized systems, or machines for other purposes relevant to the objects of the invention.)
NOTE: Wilhelms-Tricarico teaches text comprising text derived from human speech.
OBVIOUSNESS:
Using the same reasoning from claim 14, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to apply the method of Lim using the text-derived feature vectors of Wilhelms-Tricarico as input to allow different experts to specialize in different regions of the text-feature input space, thereby improving the prediction performance relative to a single model.
Claim(s) 16-18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jordan (“Hierarchical mixtures of experts and the EM algorithm”, 1993) in view of Cruz (“META-DES.H: a dynamic ensemble selection technique using meta-learning and a dynamic weighting approach”, 2018) further in view of Lim (US 20230297895 A1, 2023-09-21).
Regarding claim 16,
Claim 16 is a non-transitory computer readable medium claim that is substantially similar to method claim 1, with one added limitation, which is taught by Lim;
One or more tangible, non-transitory computer readable storage media encoded with instructions that, when executed by one or more processors, cause the one or more processors to:
([0109] The scope of the present disclosure includes software or machine-executable instructions ... a non-transitory computer-readable medium in which such software, instructions, etc., are stored and executable on a device or computer… [0108] various embodiments of the present disclosure may be implemented by... processors)
OBVIOUSNESS TO COMBINE LIM WITH JORDAN AND CRUZ:
Lim, Jordan, and Cruz are all analogous art to the present disclosure as they pertain to ensemble machine learning methods.
Jordan provides the base of weighting a plurality of predictions using a plurality of prediction models, Cruz provides a technique for incorporating model behavior into the determination of prediction weights, and Lim provides hardware capable of implementing an ensemble machine learning architecture.
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the hardware taught by Lim to implement the ensemble techniques disclosed by Jordan and Cruz to provide a physical medium allow the techniques to actually be performed.
The remaining limitations are substantially similar to the limitations of method claim 1, and are taught using the same reasoning.
Regarding claim 17-18,
Claims 17-18 are non-transitory computer readable medium claims that are substantially similar to method claims 2-3, respectively, and are taught using the same reasoning.
Claim(s) 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Jordan (“Hierarchical mixtures of experts and the EM algorithm”, 1993) in view of Cruz (“META-DES.H: a dynamic ensemble selection technique using meta-learning and a dynamic weighting approach”, 2018) further in view of Lim (US 20230297895 A1, 2023-09-21) as applied to claim 16 above, and further in view of Lesner (US 20210406780 A1, 2021-12-30).
Regarding claim 20,
Claim 20 is a non-transitory computer readable medium claim that is substantially similar to method claim 10, and is taught using the same reasoning.
Response to Arguments
Applicant’s arguments, filed 04/10/2026, with respect to the 35 USC 101 rejections have been fully considered and are persuasive. The 101 rejections of claims 1-20 have been withdrawn.
Applicant’s arguments, filed 04/10/2026, with respect to the 35 USC 102 and 103 rejections regarding claim 1-20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
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 Matthew Alan Cady whose telephone number is (571) 272-7229. The examiner can normally be reached Monday - Friday, 7:30 am - 5:00 pm ET.
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, Cesar Paula can be reached on (571)272-4128. 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.
/MATTHEW ALAN CADY/ Examiner, Art Unit 2145
/CESAR B PAULA/ Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights