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 .
This action is made final.
Claims 1-21 are pending. Claims 1, 18 and 20 are independent claims.
Response to Arguments
Applicant’s arguments, dated 2/23/2026, regarding the 35 U.S.C. 101 rejections of the previous office action, have been fully considered but are unpersuasive. However, due to the amendments, new grounds of rejection have been applied – see the updated 101 rejections below. Applicant argues that the claimed invention reflects a benefit of reducing training data size and therefore shorter training, but examiner argues that this alleged benefit is not reflected in the claims.
Applicant’s arguments, dated 2/23/2026, regarding the 35 U.S.C. 103 rejections of the previous office action, have been fully considered but are unpersuasive. However, due to the amendments, new grounds of rejection have been applied – see the updated 103 rejections below.
Applicant’s arguments, dated 2/23/2026, regarding the 35 U.S.C. 112 rejections of the previous office action, have been fully considered and are persuasive. The 112 rejections
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1, 2, 18, 19, 20 and 21 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.
Regarding claim 1, it recites: dividing the plurality of pieces of data for learning into one or more groups… dividing the plurality of pieces of data for input into one or more regions… dividing the plurality of pieces of data for learning belonging to one class into one or more groups. It is unclear what dividing data into one group or region means, because dividing data implies splitting it into at least two parts.
Claims 2, 18, 19, 20 and 21 also recite limitations that involve dividing data into one group/region.
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 1-21 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1:
Step 1: This part of the eligibility analysis evaluates whether the claim falls within any statutory category. See MPEP 2106.03. Claim 1 recites: A learning method… Claim 1 is directed to a method (i.e., a process) (Step 1: YES).
Step 2A prong 1: Does the claim recite a judicial exception? Claim 1 recites: dividing the plurality of pieces of data for learning into a plurality of groups to generate a plurality of input learning data groups (dividing data into groups is a mental process)… wherein the dividing of the plurality of pieces of data for learning includes dividing the plurality of pieces of data for learning into a plurality of regions to generate, as one of the plurality of input learning data groups, a collection of first type divided input data after the division belonging to a same region (dividing data into groups by region, i.e., sections of a graph, is a mental process), or dividing the plurality of pieces of data for learning belonging to one class into a plurality of groups to generate, as one of the plurality of input learning data groups, a collection of second type divided input data after the division (dividing input data by class is a mental process). These steps can be performed mentally or are mathematical calculations (Step 2A prong 1: YES).
Step 2A prong 2: Does the claim recite additional elements? Do those additional elements, considered individually and in combination, integrate the judicial exception into a practical application? Claim 1 recites: for M number of vector neural network type machine learning models including a plurality of vector neuron layers, the M number of vector neural network type machine learning models being used in determining a class of spectral data, representing spectral reflectance acquired from a target, to be determined, M being an integer of two or more, the learning method comprising: preparing a plurality of pieces of data for learning including the spectral data for input and a pre-label associated with the spectral data for input… and training the M number of vector neural network type machine learning models so that a correspondence between the spectral data for input and the pre-label associated with the spectral data for input is reproduced, by inputting the plurality of input learning data groups respectively into the M number of vector neural network type machine learning models… Describing M vector neural network type machine learning models being used for classification, and training them to reproduce a correspondence between data and their labels are mere instructions to implement an abstract idea on a generic computer, without any limits on how the neural networks operate – equivalent to adding the words “apply it” to the recited judicial exception. Preparing the pieces of data for learning and inputting the input learning data groups into the models is extra-solution activity of data gathering/inputting that does not add a meaningful limitation to the learning method. Specifying that the input data is spectral reflectance acquired from a target, otherwise referred to as spectral data, is an attempt to limit the field of use without significantly more (Step 2A prong 2: NO).
Step 2B: These elements are recited at such a high level of generality that they fail to integrate the abstract idea into a practical application, since they provide nothing more than mere instructions to implement an abstract idea on a generic computer (MPEP 2106.05(f)), only amount to data gathering or inputting without significantly more (MPEP 2106.05(g)), or limit the field of use without significantly more (MPEP 2106.05(h)). These limitations, taken either alone or in combination, fail to provide an inventive concept (Step 2B: NO). Thus, the claim is not patent eligible.
Regarding claim 2:
Step 1: This part of the eligibility analysis evaluates whether the claim falls within any statutory category. See MPEP 2106.03. Claim 2 recites: A determining method for determining… Claim 2 is directed to a process (Step 1: YES).
Step 2A prong 1: Does the claim recite a judicial exception? Claim 2 recites:
and the plurality of input learning data groups is obtained by dividing the plurality of pieces of data for learning (dividing training data into groups is a mental process)… the individual data is obtained using at least one of a similarity between a feature spectrum calculated from an output of the specific layer according to an input of the data to be determined for input into the M number of vector neural network type machine learning models and the known feature spectrum group (determining a similarity between two feature spectrums is a mental process or mathematical calculation)… executing class determination for the spectral data to be determined using M number of pieces of the individual data obtained respectively for the M number of vector neural network type machine learning models (performing classification based on spectral data is a mental process)… dividing the plurality of pieces of data for learning into a plurality of regions, and using a collection of first type divided input data after the division belonging to a same region as one of the plurality of input learning data groups, and executing division processing to divide the plurality of pieces of data for learning belonging to one class into a plurality of groups, and using a collection of second type divided input data after the division processing as one of the plurality of input learning data groups (dividing input data by region or dividing data by class, i.e., sorting, are mental processes). These steps can be performed mentally or are mathematical calculations (Step 2A prong 1: YES).
Step 2A prong 2: Does the claim recite additional elements? Do those additional elements, considered individually and in combination, integrate the judicial exception into a practical application? Claim 2 recites: a class of spectral data, representing spectral reflectance acquired from a target, to be determined using M number of vector neural network type machine learning models including a plurality of vector neuron layers, M being an integer of two or more, the determining method comprising: preparing the M number of vector neural network type machine learning models trained using a plurality of pieces of data for learning including the spectral data for input and a pre-label associated with the spectral data for input, wherein each one of the M number of vector neural network type machine learning models is trained using one corresponding group of a plurality of input learning data groups… preparing M number of known feature spectrum groups corresponding to the M number of vector neural network type machine learning models after training, wherein the M number of known feature spectrum groups include a known feature spectrum group obtained from an output of a specific layer from among the plurality of vector neuron layers by inputting the plurality of input learning data groups into the M number of vector neural network type machine learning models after the training; obtaining individual data used in class determination of the spectral data to be determined for each one of the M number of vector neural network type machine learning models by inputting data to be determined for input, generated from the spectral data to be determined, into each one of the M number of vector neural network type machine learning models after the training, wherein, for each one of the M number of vector neural network type machine learning models… and an activation value corresponding to a determination value for each class output from an output layer of the M number of vector neural network type machine learning models according to the input of the data to be determined for input; and… wherein the preparing of the M number of vector neural network type machine learning models includes one of… Specifying that the data being operated on is spectral reflectance data is an attempt to limit the field of use of the invention without significantly more. Describing M number of generic vector neural network type machine learning models being used for classification and training them with input data are mere instructions to implement an abstract idea on a generic computer, without any limits on how the neural networks operate – equivalent to adding the words “apply it” to the recited judicial exception. Preparing the pieces of data for learning, inputting the input learning data groups into the models, and inputting the input determination data into the models is extra-solution activity of data gathering/inputting that does not add a meaningful limitation to the learning method. Obtaining the known feature spectrum groups, individual data feature spectrums and activation values for input data from the model output is insignificant extra-solution activity of data outputting that does not add a meaningful limitation to the learning method, or recites only the idea of a solution or outcome i.e., the claim fails to recite details of how the solution to a problem is accomplished (explaining that data is input to a model and using the outputs for further processing) (Step 2A prong 2: NO).
Step 2B: These elements are recited at such a high level of generality that they fail to integrate the abstract idea into a practical application, since they limit the field of use without significantly more (MPEP 2106.05(h)), provide nothing more than mere instructions to implement an abstract idea on a generic computer (MPEP 2106.05(f)), or only amount to data gathering, inputting or outputting without significantly more (MPEP 2106.05(g)). These limitations, taken either alone or in combination, fail to provide an inventive concept (Step 2B: NO). Thus, the claim is not patent eligible.
Regarding claims 3-17, they recite limitations which further narrow the abstract idea by specifying more details of the mental and mathematical process that occurs (Claim 3, summing the activation values is a mathematical calculation and classifying the based off the summed activation values is a mental process; Claim 4, integrating similarities from multiple models is a mathematical calculation (i.e., a sum, mean, etc.), and classifying the data based off of a threshold is a mathematical calculation or mental process of comparison; Claim 5, classifying the data based on the class with the highest activation value is a mental process; Claim 6, classifying the data based on an activation value is a mental process, and setting the class as unknown when the similarity is less than a threshold is a mental process as well; Claim 7, calculating a weighted sum of similarities is a mathematical calculation, and setting the max or min similarity as the similarity is a mental process; Claim 8, clarifying that known feature spectrum groups have an associated class is a part of the insignificant extra-solution activity of data gathering, and calculating a similarity for each class known feature spectrum is a mathematical calculation; Claim 9, calculating a similarity for each class and calculating the representative similarity using statistical processing are mathematical calculations, while selecting the highest value of the representative similarities is a mental process; Claim 10, performing any of the statistical processing methods listed is a mathematical calculation or formula; Claim 11, classifying the data as an unknown class when similarity is below a threshold is a mental process; Claim 12, classifying the data based on how prevalent the class is within the model outputs is a mental process; Claim 13, generating a similarity between feature spectrums is a mathematical calculation, classifying data based on the highest similarity value is a mental process, and calculating a sum or product of the similarities is a mathematical calculation – the following classification is still mental; Claim 14, generating a similarity between feature spectrums and calculating a reference value for a model based on the output similarity and a weighting coefficient are mathematical calculations, and classifying data based on the reference values and class data is a mental process; Claim 15, calculating the reference value with multiplication is a mathematical formula, setting the class based on the highest sum or max/min values is a mathematical calculation combined with a mental process; Claim 16, setting the class of data as unknown when there is disagreement in model determined class as described is a mental process; Claim 17, clustering based on class is a mental process, and randomly sampling data belonging to one class via sampling with replacement is insignificant extra-solution activity – see MPEP 2106.05(g)(1)).
Regarding claims 18 and 20, they are apparatuses that implement a method similar to claim 1 and are rejected on the same grounds – see above.
Regarding claims 19 and 21, they are apparatuses that implement a method similar to claim 2 and are rejected on the same grounds – see above.
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, 18, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hinton et al. (US 20200285934 A1), herein Hinton in view of Lyman et al. (US 20200161005 A1), herein Lyman, Rao et al. (US 20190188584 A1), herein Rao, and Fan et al. (US 20200193597 A1), herein Fan.
Regarding claim 1, Hinton teaches: A learning method for M number of vector neural network type machine learning models including a plurality of vector neuron layers (¶42, A capsule layer refers to a neural network layer composed of capsules. Each capsule is configured to generate a capsule output… The capsule output includes two or more numbers and may be represented as a vector – Hinton teaches a single model, where M = 1), the M number of vector neural network type machine learning models being used in determining a class of
Hinton fails to teach: M being an integer of two or more... (examiner note: Hinton fails to teach multiple models) dividing the plurality of pieces of data for learning into a plurality of groups to generate a plurality of input learning data groups… wherein the dividing of the plurality of pieces of data for learning includes dividing the plurality of pieces of data for learning into a plurality of regions to generate, as one of the plurality of input learning data groups, a collection of first type divided input data after the division belonging to a same region…
However, in the same field of endeavor, Lyman teaches: M being an integer of two or more (¶338, In some embodiments, multiple models are trained by utilizing multiple sets of training data, for example, where each set of training data corresponds to a different modality and/or anatomical region – Lyman teaches multiple models, or M >= 2)... dividing the plurality of pieces of data for learning into a plurality of groups to generate a plurality of input learning data groups (¶517, image data of the medical scans is partitioned)… wherein the dividing of the plurality of pieces of data for learning includes dividing the plurality of pieces of data for learning into a plurality of regions to generate, as one of the plurality of input learning data groups, a collection of first type divided input data after the division belonging to a same region (¶517, image data of the medical scans is partitioned by a plurality of anatomical subregion types. The plurality of training sets includes a set of anatomical subregion type subsets that each include image data of the plurality of medical scans of a corresponding one of the plurality of anatomical subregion types)…
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 a multi-model approach with input data partitioning as disclosed by Lyman in the method disclosed by Hinton to increase model specialization and overall performance (¶382, allowing more precise inference data to be generated by utilizing particular models trained to process the type of scan presented and/or trained to process a particular type of detected abnormality in the type of scan presented).
Hinton in view of Lyman fails to teach: or dividing the plurality of pieces of data for learning belonging to one class into a plurality of groups to generate, as one of the plurality of input learning data groups, a collection of second type divided input data after the division.
However, in the same field of endeavor, Rao teaches: or dividing the plurality of pieces of data for learning belonging to one class into a plurality of groups to generate, as one of the plurality of input learning data groups, a collection of second type divided input data after the division (¶95, the method 100 (step 130) may use a stratified k-fold cross-validation approach based on the classification of process observations (normal, pre-failure, post-failure, etc.)).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to divide data from one class into at least one group as disclosed by Rao in the method disclosed by Hinton in view of Lyman to create representative data samples (¶95, Stratification involves rearranging the data as to ensure each fold is a good representative of the whole input sub-dataset).
Hinton in view of Lyman and Rao fails to teach: spectral data, representing spectral reflectance acquired from a target…
However, in the same field of endeavor, Fan teaches: spectral data, representing spectral reflectance acquired from a target (¶221, As described above, spectral images including reflectance data at an individual wavelength or a plurality of wavelengths can be analyzed using the machine learning techniques described herein… some of the methods disclosed herein predict wound healing parameters based at least in part on aggregate quantitative features… calculated based on a subset of pixels of a wound image that are determined to be the “wound pixels,” or the pixels that correspond to the wound tissue region rather than callus, normal skin, background, or other non-wound tissue regions)…
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 spectral reflectance data in classification as disclosed by Fan in the method disclosed by Hinton in view of Rao and Lyman to accurately and efficiently classify difficult visual data (¶222, However, such manual segmentation may be time consuming, inefficient, and potentially prone to human error. For example, the formulas used to compute area and volume lack the accuracy and precision required to measure the convex shape of wounds. In addition, identifying the true boundaries of the wound and classification of tissues within the wound, such as epithelial growth, requires a high level of competency).
Regarding claims 18 and 20, they are both apparatuses that recite similar limitations to claim 1 and are rejected on the same grounds – see above.
Claim(s) 2, 5, 6, 8, 11, 16, 17, 19 and 21 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hinton in view of Lyman, Shankar (US 20210256300 A1), Rao and Fan.
Regarding claim 2, Hinton teaches: A determining method for determining a class of (¶42, A capsule layer refers to a neural network layer composed of capsules. Each capsule is configured to generate a capsule output… The capsule output includes two or more numbers and may be represented as a vector – Hinton teaches a single model, or where M = 1) the determining method comprising: preparing the M number of vector neural network type machine learning models trained using a plurality of pieces of data for learning including the (¶64, if the training engine 126 trains the network 100 to perform image classification tasks, then the training data may include images and corresponding image class labels, where the image class label indicates the type of object depicted in the corresponding image), (¶63, the class capsule layer 116 includes a respective capsule for each class in the predetermined set, and the network output 118 is the output of each capsule of the class capsule layer… In some implementations, the network output 118 is the activation of each capsule of the class capsule layer 116), (¶63, the class capsule layer 116 includes a respective capsule for each class in the predetermined set, and the network output 118 is the output of each capsule of the class capsule layer… In some implementations, the network output 118 is the activation of each capsule of the class capsule layer 116); and executing class determination for the (¶24, In some implementations, the neural network is configured to receive a network input and to classify the network input as belonging to one or more of a predetermined set of classes),
Hinton fails to teach: M being an integer of two or more (examiner note: Hinton fails to teach multiple models)… wherein each one of the M number of vector neural network type machine learning models is trained using one corresponding group of a plurality of input learning data groups, and the plurality of input learning data groups is obtained by dividing the plurality of pieces of data for learning… wherein the preparing of the M number of vector neural network type machine learning models includes one of: dividing the plurality of pieces of data for learning into a plurality of regions, and using a collection of first type divided input data after the division belonging to a same region as one of the plurality of input learning data groups…
However, in the same field of endeavor, Lyman teaches: M being an integer of two or more (¶338, In some embodiments, multiple models are trained)… wherein each one of the M number of vector neural network type machine learning models is trained using one corresponding group of a plurality of input learning data groups (¶338, multiple models are trained by utilizing multiple sets of training data, for example, where each set of training data corresponds to a different modality and/or anatomical region – and – ¶382, Multiple models can be trained to process medical scans of different views of a patient, medical scans of different modalities, and/or medical scans of different anatomical regions – i.e., models each have their own learning data), and the plurality of input learning data groups is obtained by dividing the plurality of pieces of data for learning (¶517, image data of the medical scans is partitioned)… wherein the preparing of the M number of vector neural network type machine learning models includes one of: dividing the plurality of pieces of data for learning into a plurality of regions, and using a collection of first type divided input data after the division belonging to a same region as one of the plurality of input learning data groups (¶517, image data of the medical scans is partitioned by a plurality of anatomical subregion types. The plurality of training sets includes a set of anatomical subregion type subsets that each include image data of the plurality of medical scans of a corresponding one of the plurality of anatomical subregion types)…
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 a multi-model approach with input data partitioning as disclosed by Lyman in the method disclosed by Hinton to increase model specialization and overall performance (¶382, allowing more precise inference data to be generated by utilizing particular models trained to process the type of scan presented and/or trained to process a particular type of detected abnormality in the type of scan presented).
Hinton in view of Lyman fails to teach: preparing M number of known feature spectrum groups corresponding to the M number of vector neural network type machine learning models after training, wherein the M number of known feature spectrum groups include a known feature spectrum group obtained from an output of a specific layer from among the plurality of vector neuron layers by inputting the plurality of input learning data groups into the M number of vector neural network type machine learning models after the training… the individual data is obtained using at least one of a similarity between a feature spectrum calculated from an output of the specific layer according to an input of the data to be determined for input into the M number of vector neural network type machine learning models and the known feature spectrum group.
However, in the same field of endeavor, Shankar teaches: preparing M number of known feature spectrum groups corresponding to the M number of vector neural network type machine learning models after training, wherein the M number of known feature spectrum groups include a known feature spectrum group obtained from an output of a specific layer from among the plurality of vector neuron layers by inputting the plurality of input learning data groups into the M number of vector neural network type machine learning models after the training (¶24, The neural network associated with the robotic device compares the feature vector of the unknown fruit to feature vectors of the known fruits. Based on the comparison, the neural network may classify the unknown fruit)… the individual data is obtained using at least one of a similarity between a feature spectrum calculated from an output of the specific layer according to an input of the data to be determined for input into the M number of vector neural network type machine learning models and the known feature spectrum group (¶26, the neural network determines whether a difference and/or a similarity between the feature vector of the unknown fruit and the feature vector of a known fruit satisfies a threshold. The selected known fruit may be the known fruit that is most similar to the unknown fruit).
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 feature vector similarity measures to classify data as disclosed by Shankar in the method disclosed by Hinton in view of Lyman to identify previously unseen data (¶26, an unknown object that shares characteristics with an object of a training class can be identified based on the similarity to trained characteristics).
Hinton in view of Lyman and Shankar fails to teach: and executing division processing to divide the plurality of pieces of data for learning belonging to one class into a plurality of groups, and using a collection of second type divided input data after the division processing as one of the plurality of input learning data groups.
However, in the same field of endeavor, Rao teaches: and executing division processing to divide the plurality of pieces of data for learning belonging to one class into a plurality of groups, and using a collection of second type divided input data after the division processing as one of the plurality of input learning data groups (¶95, the method 100 (step 130) may use a stratified k-fold cross-validation approach based on the classification of process observations (normal, pre-failure, post-failure, etc.)).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to divide data from one class into at least one group as disclosed by Rao in the method disclosed by Hinton in view of Lyman and Shankar to create representative data samples (¶95, Stratification involves rearranging the data as to ensure each fold is a good representative of the whole input sub-dataset).
Hinton in view of Lyman, Shankar and Rao fails to teach: spectral data, representing spectral reflectance…
However, in the same field of endeavor, Fan teaches: spectral data, representing spectral reflectance (¶221, As described above, spectral images including reflectance data at an individual wavelength or a plurality of wavelengths can be analyzed using the machine learning techniques described herein… some of the methods disclosed herein predict wound healing parameters based at least in part on aggregate quantitative features…. calculated based on a subset of pixels of a wound image that are determined to be the “wound pixels,” or the pixels that correspond to the wound tissue region rather than callus, normal skin, background, or other non-wound tissue regions)…
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 spectral reflectance data in classification as disclosed by Fan in the method disclosed by Hinton in view of Lyman, Shankar and Rao to accurately and efficiently classify difficult visual data (¶222, However, such manual segmentation may be time consuming, inefficient, and potentially prone to human error. For example, the formulas used to compute area and volume lack the accuracy and precision required to measure the convex shape of wounds. In addition, identifying the true boundaries of the wound and classification of tissues within the wound, such as epithelial growth, requires a high level of competency). Fan’s teaching of spectral reflectance data being used in classification applies to the dependent claims as well.
Regarding claim 5, Hinton further teaches: The determining method according to claim 2, wherein the obtaining of the individual data includes generating a class corresponding to a highest activation value from among activation values corresponding to classes for each machine learning model as a pre-determined class as an element of the individual data (¶25, In some implementations, an activation of a respective capsule of the class capsule layer is greater than activations of other capsules of the class capsule layer; the network input is classified as belonging to a particular class corresponding to the respective capsule of the class capsule layer).
Regarding claim 6, Hinton further teaches: The determining method according to claim 2, wherein the obtaining of the individual data includes generating… a class corresponding to a highest activation value from among activation values corresponding to classes as a pre-determined class as an element of the individual data (¶25, In some implementations, an activation of a respective capsule of the class capsule layer is greater than activations of other capsules of the class capsule layer; the network input is classified as belonging to a particular class corresponding to the respective capsule of the class capsule layer)…
Hinton in view of Rao and Fan fails to teach: when the similarity is equal to or greater than a predetermined threshold… for each machine learning model, and generating, when the similarity is less than the predetermined threshold…
However, in the same field of endeavor, Shankar teaches: when the similarity is equal to or greater than a predetermined threshold (¶26, the neural network determines whether a difference and/or a similarity between the feature vector of the unknown fruit and the feature vector of a known fruit satisfies a threshold. The selected known fruit may be the known fruit that is most similar to the unknown fruit – i.e., classifying when the similarity threshold is exceeded)… for each machine learning model, and generating, when the similarity is less than the predetermined threshold… (¶28, In one aspect the unknown fruit may not be identifiable when the similarity between the feature vector of the sixth fruit and the one or more feature vectors of a specified known fruit (e.g., the lime) less than the similarity threshold – i.e., not classifying the item when the similarity is too low).
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 thresholds with feature vector similarity measures to classify data, or determine it as unclassifiable as disclosed by Shankar in the method disclosed by Hinton in view of Rao and Fan to account for situations where the unknown item is not similar to any of the known items (¶28, the unknown fruit may not be identifiable when the difference between the feature vector of the unknown fruit and the one or more feature vectors of the specified known fruit is greater than the difference threshold).
Hinton in view of Shankar, Rao and Fan fails to teach: the pre-determined class as an unknown class different from a class corresponding to the pre-label as the element of the individual data for each machine learning model.
However, in the same field of endeavor, Lyman teaches: the pre-determined class as an unknown class different from a class corresponding to the pre-label as the element of the individual data for each machine learning model (¶68, Some scans identified as normal scans can include identified abnormalities that are classified as benign, and include zero abnormalities classified as either unknown or malignant – Lyman describes the use of an “unknown” class for data that is separate from other classes. Classifying data as the most similar class as described by Shankar could be followed by reclassifying the data to the “unknown” class if similarity is low).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to utilize an unknown class as disclosed by Lyman in the method disclosed by Hinton in view of Shankar, Rao and Fan to describe unusual data that is unable to be classified (¶68, Some medical scans can include one or more abnormalities… can also include any other unknown, malignant or benign feature of a medical scan identified as not normal).
Regarding claim 8, Hinton in view of Lyman, Rao and Fan fails to teach: The determining method according to claim 2, wherein each known feature spectrum included in the known feature spectrum group is associated with class classification information indicating which class the known feature spectrum belongs to, and when the known feature spectrum associated with the class classification information is referred to as a by class known feature spectrum, the obtaining of the individual data includes calculating a by class similarity, which is a similarity between the by class known feature spectrum and the feature spectrum, for each class, and generating a class associated with the by class similarity with a highest value from among by class similarities calculated for classes as a pre-determined class as an element of the individual data.
However, in the same field of endeavor, Shankar teaches: wherein each known feature spectrum included in the known feature spectrum group is associated with class classification information indicating which class the known feature spectrum belongs to (¶49, the unknown fruit 410 is identified as a lime when the similarity between the feature vector of the unknown fruit 410 and the one or more feature vectors of the lime is greater than a similarity threshold – the known feature spectrums are associated with a specific class, in this case fruits), and when the known feature spectrum associated with the class classification information is referred to as a by class known feature spectrum, the obtaining of the individual data includes calculating a by class similarity, which is a similarity between the by class known feature spectrum and the feature spectrum, for each class (¶48, The neural network determines the differences and similarities between the feature vectors of the unknown fruit 410 and the known fruits), and generating a class associated with the by class similarity with a highest value from among by class similarities calculated for classes as a pre-determined class as an element of the individual data (¶18, When a classifier is confronted with an unknown object, the classifier may select a most similar known object).
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 similarity between labeled feature spectrums as disclosed by Shankar in the method disclosed by Hinton in view of Lyman, Rao and Fan to correctly classify the unknown data (¶29, a network that is capable of learning unknown objects based on elements of trained objects).
Regarding claim 11, Hinton in view of Rao and Fan fails to teach: The determining method according to claim 8, wherein the obtaining of the individual data further includes generating an unknown class different from a class corresponding to the pre-label instead of the class associated with the by class similarity as the pre-determined class as the element of the individual data….
However, in the same field of endeavor, Lyman teaches: wherein the obtaining of the individual data further includes generating an unknown class different from a class corresponding to the pre-label instead of the class associated with the by class similarity as the pre-determined class as the element of the individual data (¶68, Some scans identified as normal scans can include identified abnormalities that are classified as benign, and include zero abnormalities classified as either unknown or malignant – Lyman explicitly teaches the idea of a distinct “unknown” class).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to utilize an unknown class as disclosed by Lyman in the method disclosed by Hinton in view of Rao to describe unusual data that is unable to be classified (¶68, Some medical scans can include one or more abnormalities… can also include any other unknown, malignant or benign feature of a medical scan identified as not normal).
Hinton in view of Lyman, Rao and Fan fails to teach: when the highest value is less than a predetermined threshold.
However, in the same field of endeavor, Shankar teaches: when the highest value is less than a predetermined threshold (¶28, In one aspect the unknown fruit may not be identifiable when the similarity between the feature vector of the sixth fruit and the one or more feature vectors of a specified known fruit (e.g., the lime) less than the similarity threshold).
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 thresholds with feature vector similarity measures to classify data as disclosed by Shankar in the method disclosed by Hinton in view of Lyman, Rao and Fan to account for situations where the unknown item is not similar to any of the known items (¶28, the unknown fruit may not be identifiable when the difference between the feature vector of the unknown fruit and the one or more feature vectors of the specified known fruit is greater than the difference threshold).
Regarding claim 16, Hinton in view of Rao and Fan fails to teach: The determining method according to claim 5, wherein the executing of the class determination includes when one of a plurality of pre-determined classes corresponding to the M number of vector neural network type machine learning models indicates an unknown… regardless of classes indicated by other pre-determined classes.
However, in the same field of endeavor, Shankar teaches: wherein the executing of the class determination includes when one of a plurality of pre-determined classes corresponding to the M number of vector neural network type machine learning models indicates an unknown… regardless of classes indicated by other pre-determined classes (¶28, In one aspect the unknown fruit may not be identifiable when the similarity between the feature vector of the sixth fruit and the one or more feature vectors of a specified known fruit (e.g., the lime) less than the similarity threshold).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to indicate input data is unknown as disclosed by Shankar in the method disclosed by Hinton in view of Rao and Fan to account for situations where the data is not similar to any of the known data (¶28, the unknown fruit may not be identifiable when the difference between the feature vector of the unknown fruit and the one or more feature vectors of the specified known fruit is greater than the difference threshold).
Hinton in view of Shankar, Rao and Fan fails to teach: class different from a class corresponding to the pre-label, setting the unknown class as the class of the spectral data to be determined…
However, in the same field of endeavor, Lyman teaches: class different from a class corresponding to the pre-label, setting the unknown class as the class of the spectral data to be determined (¶68, Some scans identified as normal scans can include identified abnormalities that are classified as benign, and include zero abnormalities classified as either unknown or malignant – Lyman teaches a distinct “unknown” class assigned to data)…
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to utilize an unknown class as disclosed by Lyman in the method disclosed by Hinton in view of Shankar, Rao and Fan to draw attention to unusual data that is unable to be classified (¶68, Some medical scans can include one or more abnormalities… can also include any other unknown, malignant or benign feature of a medical scan identified as not normal).
Regarding claim 17, Hinton in view of Shankar and Rao fails to teach: The determining method according to claim 2, wherein the division processing is executed by performing clustering of the plurality of pieces of data for learning belonging to the one class, or randomly extracting the plurality of pieces of data for learning belonging to the one class via sampling with replacement.
However, in the same field of endeavor, Lyman teaches: wherein the division processing is executed by performing clustering of the plurality of pieces of data for learning belonging to the one class, or randomly extracting the plurality of pieces of data for learning belonging to the one class via sampling with replacement (¶137, The medical scan classifications selected to segregate the medical scans can be automatically determined by the medical scan image analysis system, for example, where an unsupervised clustering algorithm is applied to the original training set to determine appropriate medical scan classifications based on the output of the unsupervised clustering algorithm).
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 clustering of data as disclosed by Lyman in the method disclosed by Hinton in view of Shankar and Rao to provide data for multiple models (¶137, to segregate the medical scans into multiple training sets).
Regarding claims 19 and 21, they are both apparatuses that recite similar limitations to claim 1 and are rejected on the same grounds – see above.
Claim(s) 3-4, 9, 10 and 12 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hinton in view of Lyman, Shankar, Rao and Fan as applied to claim 2 above, and further in view of Velic et al. (US 20170304732 A1), herein Velic.
Regarding claim 3, Hinton in view of Lyman, Shankar and Rao fails to teach: The determining method according to claim 2, wherein each one of the M number of pieces of the individual data includes the activation value corresponding to each class and the executing of the class determination includes setting, as a determination class, a class with a highest activation value for determination calculated using a cumulative activation value obtained by adding together activation values of the M number of pieces of the individual data for each class.
However, in the same field of endeavor, Velic teaches: wherein each one of the M number of pieces of the individual data includes the activation value corresponding to each class and the executing of the class determination includes setting, as a determination class, a class with a highest activation value for determination calculated using a cumulative activation value obtained by adding together activation values of the M number of pieces of the individual data for each class (¶49, In model-based averaging, an ensemble of more than one classification model is used to classify the same image (or more images) and predictions are averaged (or otherwise combined) across all classes and thus a final decision is made based on a combination of the outputs of the different models).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine model outputs as disclosed by Velic in the method disclosed by Hinton in view of Lyman, Shankar, Rao and Fan to increase accuracy (¶49, Empirically, this method increases the accuracy of the whole system).
Regarding claim 4, Hinton in view of Rao and Fan fails to teach: The determining method according to claim 3, wherein each one of the M number of pieces of the individual data includes the similarity and the executing of the class determination further includes… and setting, when the similarity for determination is equal to or greater than a predetermined threshold, the class with the highest activation value for determination as the determination class, and setting, when the similarity for determination is less than the predetermined threshold, regardless of the activation value for determination, an unknown.
However, in the same field of endeavor, Shankar teaches: wherein each one of the M number of pieces of the individual data includes the similarity and the executing of the class determination further includes… and setting, when the similarity for determination is equal to or greater than a predetermined threshold, the class with the highest activation value for determination as the determination class (¶26, the neural network determines whether a difference and/or a similarity between the feature vector of the unknown fruit and the feature vector of a known fruit satisfies a threshold. The selected known fruit may be the known fruit that is most similar to the unknown fruit), and setting, when the similarity for determination is less than the predetermined threshold, regardless of the activation value for determination, an unknown (¶28, In one aspect the unknown fruit may not be identifiable when the similarity between the feature vector of the sixth fruit and the one or more feature vectors of a specified known fruit (e.g., the lime) less than the similarity threshold)…
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 thresholds with feature vector similarity measures to classify data as disclosed by Shankar in the method disclosed by Hinton in view of Rao and Fan to account for situations where the unknown item is not similar to any of the known items (¶28, the unknown fruit may not be identifiable when the difference between the feature vector of the unknown fruit and the one or more feature vectors of the specified known fruit is greater than the difference threshold).
Hinton in view of Rao, Shankar and Fan fails to teach: generating a similarity for determination by integrating the respective similarities of the M number of vector neural network type machine learning models...
However, in the same field of endeavor, Velic teaches: generating a similarity for determination by integrating the respective similarities of the M number of vector neural network type machine learning models (¶49, In model-based averaging, an ensemble of more than one classification model is used to classify the same image (or more images) and predictions are averaged (or otherwise combined) across all classes and thus a final decision is made based on a combination of the outputs of the different models).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine model outputs as disclosed by Velic in the method disclosed by Hinton in view of Shankar, Rao and Fan to increase accuracy (¶49, Empirically, this method increases the accuracy of the whole system).
Hinton in view of Rao, Shankar, Fan and Velic fails to teach: class different from a class corresponding to a pre-label, as the determination class…
However, in the same field of endeavor, Lyman teaches: class different from a class corresponding to a pre-label, as the determination class (¶68, Some scans identified as normal scans can include identified abnormalities that are classified as benign, and include zero abnormalities classified as either unknown or malignant)…
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to utilize an unknown class as disclosed by Lyman in the method disclosed by Hinton in view of Shankar, Rao, Fan and Velic to draw attention to unusual data that is unable to be classified (¶68, Some medical scans can include one or more abnormalities… can also include any other unknown, malignant or benign feature of a medical scan identified as not normal).
Regarding claim 9, Hinton in view of Lyman, Rao and Fan fails to teach: The determining method according to claim 8, wherein the calculating of the by class similarity includes calculating a similarity between each one of a plurality of by class known feature spectrums and the feature spectrum for each class… and the generation of the class includes generating the class associated with the representative similarity with a highest value from among representative similarities calculated for each class as the pre- determined class as the element of the individual data.
However, in the same field of endeavor, Shankar teaches: wherein the calculating of the by class similarity includes calculating a similarity between each one of a plurality of by class known feature spectrums and the feature spectrum for each class (¶29, Specifically, the network learns (e.g., identifies) unknown objects by determining similarities and differences between characteristics of trained objects and the unknown object)… and the generation of the class includes generating the class associated with the representative similarity with a highest value from among representative similarities calculated for each class as the pre- determined class as the element of the individual data (¶26, The selected known fruit may be the known fruit that is most similar to the unknown fruit).
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 similarity between labeled feature spectrums as disclosed by Shankar in the method disclosed by Hinton in view of Lyman, Rao and Fan to correctly classify the unknown data (¶29, a network that is capable of learning unknown objects based on elements of trained objects).
Hinton in view of Lyman, Shankar, Rao and Fan fails to teach: and calculating a representative similarity of a plurality of similarities for each class by executing statistical processing of the plurality of similarities calculated for each class.
However, in the same field of endeavor, Velic teaches: and calculating a representative similarity of a plurality of similarities for each class by executing statistical processing of the plurality of similarities calculated for each class (¶49, In model-based averaging, an ensemble of more than one classification model is used to classify the same image (or more images) and predictions are averaged (or otherwise combined) across all classes and thus a final decision is made based on a combination of the outputs of the different models).
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 statistical processing on model outputs as disclosed by Velic in the method disclosed by Hinton in view of Lyman, Shankar, Rao and Fan to increase accuracy (¶49, Empirically, this method increases the accuracy of the whole system).
Regarding claim 10, Hinton in view of Lyman, Shankar, Rao and Fan fails to teach: The determining method according to claim 9, wherein the statistical processing of the plurality of similarities includes calculating a maximum value, a median value, an average value, or a modal value of the plurality of similarities as the representative similarity.
However, in the same field of endeavor, Velic teaches: wherein the statistical processing of the plurality of similarities includes calculating a maximum value, a median value, an average value, or a modal value of the plurality of similarities as the representative similarity (¶49, In model-based averaging, an ensemble of more than one classification model is used to classify the same image (or more images) and predictions are averaged (or otherwise combined) across all classes and thus a final decision is made based on a combination of the outputs of the different models).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to perform statistical processing such as taking a mean as disclosed by Velic in the method disclosed by Hinton in view of Lyman, Shankar, Rao and Fan to increase accuracy (¶49, Empirically, this method increases the accuracy of the whole system).
Regarding claim 12, Hinton in view of Lyman, Shankar, Rao and Fan fails to teach: The determining method according to claim 5, wherein the executing of the class determination includes setting a most prevalent class from pre-determined classes included in the individual data of the M number of vector neural network type machine learning models as the class of the spectral data to be determined.
However, in the same field of endeavor, Velic teaches: wherein the executing of the class determination includes setting a most prevalent class from pre-determined classes included in the individual data of the M number of vector neural network type machine learning models as the class of the spectral data to be determined (¶49, In model-based averaging, an ensemble of more than one classification model is used to classify the same image (or more images) and predictions are averaged (or otherwise combined) across all classes and thus a final decision is made based on a combination of the outputs of the different models).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to select the most prevalent class from the data produced by the plurality of the machine learning models as disclosed by Velic in the method disclosed by Hinton in view of Lyman, Shankar, Rao and Fan to increase accuracy (¶49, Empirically, this method increases the accuracy of the whole system).
Claim(s) 7 and 13 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hinton in view of Lyman, Shankar, Rao, and Fan as applied to claim 2 above, and further in view of Lipsky et al. (US 20210104321 A1), herein Lipsky.
Regarding claim 7, Hinton in view of Lyman and Rao fails to teach: The determining method according to claim 2, wherein the obtaining of the individual data includes, when a plurality of specific layers is provided, one of… and setting a maximum value or a minimum value of similarities corresponding to the plurality of specific layers as the similarity used in class determination…
However, in the same field of endeavor, Shankar teaches: wherein the obtaining of the individual data includes, when a plurality of specific layers is provided, one of… and setting a maximum value or a minimum value of similarities corresponding to the plurality of specific layers as the similarity used in class determination (¶26, The selected known fruit may be the known fruit that is most similar to the unknown fruit – i.e., classification is based off of maximum similarity out of known fruit feature vectors)…
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 most similar item as the classification as disclosed by Shankar in the method disclosed by Hinton in view of Lyman, Rao and Fan to classify items that are similar to multiple known items (¶26, an unknown object that shares characteristics with an object of a training class can be identified based on the similarity to trained characteristics. The neural network determines the differences and similarities between the unknown object and the known objects).
Hinton in view of Lyman, Shankar, Rao and Fan fails to teach: for each one of the plurality of specific layers, calculating a multiplication value obtained by multiplying a weighting coefficient set for each one of the plurality of specific layers and the similarity corresponding to one of the plurality of specific layers and setting a sum of multiplication values calculated as the similarity used in class determination…
However, in the same field of endeavor, Lipsky teaches: for each one of the plurality of specific layers, calculating a multiplication value obtained by multiplying a weighting coefficient set for each one of the plurality of specific layers and the similarity corresponding to one of the plurality of specific layers and setting a sum of multiplication values calculated as the similarity used in class determination (¶499, a sum or a weighted sum of the plurality of classifications or outcome values of the plurality of classifiers may be calculated to produce a single classification or outcome value for the sample)…
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 a weighted sum of model outputs as disclosed by Lipsky in the method disclosed by Hinton in view of Lyman, Shankar, Rao and Fan to improve accuracy (¶499, a single classification or outcome value may be produced for the sample having greater confidence or statistical significance).
Regarding claim 13, Hinton in view of Lyman, Rao and Fan fails to teach: The determining method according to claim 5, wherein the obtaining of the individual data further includes generating the similarity between the feature spectrum calculated from the output of the specific layer and the known feature spectrum group as the element of the individual data, and the executing of the class determination includes one of: setting a class with a highest value for the similarity from among pre-determined classes included in the individual data of the M number of vector neural network type machine learning models as the class of the spectral data to be determined… setting the pre-determined class with a highest calculated value as the class of the spectral data to be determined.
However, in the same field of endeavor, Shankar teaches: wherein the obtaining of the individual data further includes generating the similarity between the feature spectrum calculated from the output of the specific layer and the known feature spectrum group as the element of the individual data, and the executing of the class determination (¶22, a neural network associated with the robotic device generates an output of the unknown fruit from the image. The output includes a feature vector of the unknown fruit. The feature vector may be a multiple dimensional vector)… includes one of: setting a class with a highest value for the similarity from among pre-determined classes included in the individual data of the M number of vector neural network type machine learning models as the class of the spectral data to be determined… setting the pre-determined class with a highest calculated value as the class of the spectral data to be determined (¶26, The selected known fruit may be the known fruit that is most similar to the unknown fruit).
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 most similar class as the classification output as disclosed by Shankar in the method disclosed by Hinton in view of Lyman, Rao and Fan to correctly classify unknown data (¶29, a network that is capable of learning unknown objects based on elements of trained objects).
Hinton in view of Lyman, Shankar, Rao and Fan fails to teach: and calculating a sum or product of similarities included in the individual data of the classes with a same pre-determined class of the pre- determined classes and…
However, in the same field of endeavor, Lipsky teaches: and calculating a sum or product of similarities included in the individual data of the classes with a same pre-determined class of the pre- determined classes and (¶499, a sum or a weighted sum of the plurality of classifications or outcome values of the plurality of classifiers may be calculated to produce a single classification or outcome value for the sample)…
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 a sum or weighted sum of model outputs as disclosed by Lipsky in the method disclosed by Hinton in view of Lyman, Shankar, Rao and Fan to improve accuracy (¶499, a single classification or outcome value may be produced for the sample having greater confidence or statistical significance).
Claim(s) 14 and 15 is/are rejected under 35 U.S.C. 103 as being unpatentable over Hinton in view of Lyman, Shankar, Rao, Fan and Lipsky as applied to claim 5 above, and further in view of Lee et al. (US 20210110259 A1), herein Lee.
Regarding claim 14, Hinton in view of Lyman and Rao fails to teach: The determining method according to claim 5, wherein the obtaining of the individual data further includes generating the similarity between a feature spectrum calculated from the output of the specific layer and the known feature spectrum group as the element of the individual data (¶29, Specifically, the network learns (e.g., identifies) unknown objects by determining similarities and differences between characteristics of trained objects and the unknown object) and the executing of the class determination includes calculating a reference value… using the similarity… and setting the class of the spectral data to be determined using… the reference value calculated (¶26, The selected known fruit may be the known fruit that is most similar to the unknown fruit).
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 similarity measures to classify data as disclosed by Shankar in the method disclosed by Hinton in view of Lyman, Rao and Fan to correctly classify unknown data (¶29, a network that is capable of learning unknown objects based on elements of trained objects).
Hinton in view of Lyman, Shankar, Rao and Fan fails to teach: for each one of the M number of vector neural network type machine learning models… and a weighting coefficient preset for each one of the M number of vector neural network type machine learning models… the pre-determined class and…
However, in the same field of endeavor, Lee teaches: a reference value for each one of the M number of vector neural network type machine learning models… and a weighting coefficient preset for each one of the M number of vector neural network type machine learning models… the pre-determined class and (¶70, The speech recognition model may generate a final recognition result by calculating a weighted sum of a probability of each of the candidate output tokens based on a preset ensemble weight – Lee uses multiple models with preset weights, the weights are multiplied with probabilities output by the models to recognize words)…
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 preset output weights as disclosed by Lee in the method disclosed by Hinton in view of Lyman, Shankar, Rao and Fan to balance model outputs and achieve higher accuracy (¶76, In such case, the acoustic model, which is the main model, may not have a greater weight than the language model, which is the auxiliary model, and thus the word “slots” which is selected by the main model may be determined to be an ensemble result even though the language model predicts the word “sloth” as having a greater probability than the word “slots.”).
Regarding claim 15, Hinton in view of Lyman, Shankar, Rao and Fan fails to teach: The determining method according to claim 14, wherein the reference value is calculated for each one of the M number of vector neural network type machine learning models by multiplying the similarity and the weighting coefficient and the setting of the pre-determined class includes setting the pre-determined class with a highest sum of reference values of the M number of vector neural network type machine learning models with a same pre-determined class as the class of the spectral data to be determined, or setting the pre-determined class of the machine learning model with a maximum or minimum value for the reference value as the class of the spectral data to be determined.
However, in the same field of endeavor, Lee teaches: wherein the reference value is calculated for each one of the M number of vector neural network type machine learning models by multiplying the similarity and the weighting coefficient and the setting of the pre-determined class includes setting the pre-determined class with a highest sum of reference values of the M number of vector neural network type machine learning models with a same pre-determined class as the class of the spectral data to be determined, (¶71, the candidate output token “eight” may have a final weight of 0.44… and the candidate output token “ate” may have a final weight of 0.3202... Thus, the speech recognition model may determine the candidate output token “eight” having the greater weight to be a final output token – the final weights are achieved by a weighted sum of the models output probabilities – see ¶71 for more), or setting the pre-determined class of the machine learning model with a maximum or minimum value for the reference value as the class of the spectral data to be determined.
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 preset output weights and a maximum value to classify data as disclosed by Lee in the method disclosed by Hinton in view of Lyman, Shankar, Rao and Fan to balance model outputs and achieve higher accuracy (¶76, In such case, the acoustic model, which is the main model, may not have a greater weight than the language model, which is the auxiliary model, and thus the word “slots” which is selected by the main model may be determined to be an ensemble result even though the language model predicts the word “sloth” as having a greater probability than the word “slots.”).
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 HARRISON CHAN YOUNG KIM whose telephone number is (571)272-0713. The examiner can normally be reached Monday - Friday 9:00 am - 5:00 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, Cesar Paula can be reached at (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.
/HARRISON C KIM/ Examiner, Art Unit 2145
/CESAR B PAULA/ Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights