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 .
Response to Amendment
The following is in response to the amendment filed on October 6, 2025.
Claim Objections
Claim 13 is objected to because of the following informalities: Claim 13 depends on a cancelled claim. Appropriate correction is required.
Claim 21 is objected to under 37 CFR 1.75 as being a substantial duplicate of claim 13 (assuming the dependency addressed above is corrected). When two claims in an application are duplicates or else are so close in content that they both cover the same thing, despite a slight difference in wording, it is proper after allowing one claim to object to the other as being a substantial duplicate of the allowed claim. See MPEP § 608.01(m).
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, 3-11, 13-22 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 2A Prong 1:
Genetically filtering a labeled dataset to obtain a subset of the labeled dataset, wherein the label dataset and the subset each comprise data tagged with one or more classes; (Mental process, a person could filter a labeled dataset, genetical filtering just recites filtering based on certain criteria)
Based on the classification vector, selecting one or more media objects from a plurality of media objects for delivery to the user. (Mental process, a person could select media objects from a plurality of media objects)
Step 2A Prong 2:
Obtaining a multi-layer neural network comprising a neural network model pre-trained with an unlabeled training dataset; (Obtaining a model adds insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g). This element represents generic computer functions)
Fine-tuning the neural network model by training the neural network model with the subset of the labeled dataset; (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f) – Examiner’s note: high level recitation of training a machine learning model by modifying weights)
Receiving the input through a communication interface; (Receiving data adds insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g). This element represents generic computer functions)
Providing the input to the multi-layer neural network to obtain a classification vector, the classification vector having one or more entries, wherein each of the one or more entries is associated with a class of the feature; (Providing data to the multi-layer neural network adds insignificant extra-solution activity to the judicial exception, as discussed in MPEP 2106.05(g). This element represents generic computer functions)
Step 2B:
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception.
Obtaining a multi-layer neural network comprising a neural network model pre-trained with an unlabeled training dataset; (Obtaining step is well understood routine and conventional, See MPEP 2106.05(d))
Fine-tuning the neural network model by training the neural network model with the subset of the labeled dataset; (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea – see MPEP 2106.05(f) – Examiner’s note: high level recitation of training a machine learning model by modifying weights)
Receiving the input through a communication interface; (Receiving step is well understood routine and conventional, See MPEP 2106.05(d))
Providing the input to the multi-layer neural network to obtain a classification vector, the classification vector having one or more entries, wherein each of the one or more entries is associated with a class of the feature; (Providing step is well understood routine and conventional, See MPEP 2106.05(d))
The additional elements as disclosed above alone or in combination do not integrate the judicial exception into practical application as they are mere insignificant extra solution activity in combination of generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above.
The claim is ineligible.
Regarding Claim 3:
Claim 3 recites
The method of claim 2, wherein the genetic training comprises:
initializing a genetic filtering data vector, the genetic filtering data vector comprising data selected from the labeled dataset;
obtaining an average validation accuracy measurement of the genetic filtering data vector by propagating the genetic filtering data vector through the pre-trained neural network model; and
generating one or more new genetic filtering data vectors based on the average validation accuracy measurement.
This provides a further description of the data within the abstract idea, as discussed with regards to claim 2.
The limitations of dependent claim 3, as drafted, is a process that, under broadest reasonable interpretation, recited Mathematical concepts which is also an abstract idea. Initializing a genetic training data vector, obtaining the average validation accuracy measurement of the genetic data vector and generating one or more new genetic vectors are mathematical calculations. See MPEP 2106.04(2)(I)(C).
Claim 3 does not include any additional elements (i.e. elements other than the abstract idea) that integrate the abstract idea into a practical application or amount to significantly more than the abstract idea, and is therefore also ineligible under 35 U.S.C 101. See MPEP 2106.05.
Claim 3 is not patent eligible.
Regarding Claim 4:
Claim 4 recites
The method of claim 1, wherein the labeled training dataset is smaller than the unlabeled training dataset.
This provides a further description of the data within the abstract idea, as discussed with regards to claim 1.
The limitation of dependent claim 4, as drafted, is a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind. A user could easily, with the use of pen and paper, determine that the labeled training dataset is smaller than the unlabeled training dataset. This is considered as Mental process under Abstract ideas as they can be performed in the human mind, including concepts, observation, evaluation, judgement and opinion.
Claim 4 does not include any additional elements (i.e. elements other than the abstract idea) that integrate the abstract idea into a practical application or amount to significantly more than the abstract idea, and is therefore also ineligible under 35 U.S.C 101. See MPEP 2106.05.
Claim 4 is not patent eligible.
Regarding Claim 5:
Claim 5 recites
The method of claim 4 comprising pre-training the neural network model, wherein the pre-training comprises bidirectional training by applying a missing words mask to the unlabeled dataset.
This provides a further description of the data within the abstract idea, as discussed with regards to claim 4.
The limitation of dependent claim 5, as drafted, is a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind. A user could easily, with the use of pen and paper, apply a missing words mask to the unlabeled dataset. This is considered as Mental process under Abstract ideas as they can be performed in the human mind, including concepts, observation, evaluation, judgement and opinion.
Claim 5 does not include any additional elements (i.e. elements other than the abstract idea) that integrate the abstract idea into a practical application or amount to significantly more than the abstract idea, and is therefore also ineligible under 35 U.S.C 101. See MPEP 2106.05.
Claim 5 is not patent eligible.
Regarding Claim 6:
Claim 6 recites
The method of claim 5, wherein the pre-training comprises training through sentence prediction.
This provides a further description of the data within the abstract idea, as discussed with regards to claim 5.
The limitation of dependent claim 6, as drafted, is a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind. A user could easily, with the use of pen and paper, perform sentence prediction. This is considered as Mental process under Abstract ideas as they can be performed in the human mind, including concepts, observation, evaluation, judgement and opinion.
Claim 6 does not include any additional elements (i.e. elements other than the abstract idea) that integrate the abstract idea into a practical application or amount to significantly more than the abstract idea, and is therefore also ineligible under 35 U.S.C 101. See MPEP 2106.05.
Claim 6 is not patent eligible.
Regarding Claim 7:
Claim 7 recites
The method of claim 4, wherein the fine-tuning comprises training through back- propagation.
This provides a further description of the data within the abstract idea, as discussed with regards to claim 4.
The limitation of dependent claim 7, as drafted, is a process that, under broadest reasonable interpretation, recited Mathematical concepts which is also an abstract idea. Back-propagation is a mathematical calculation. See MPEP 2106.04(2)(I)(C).
Claim 7 does not include any additional elements (i.e. elements other than the abstract idea) that integrate the abstract idea into a practical application or amount to significantly more than the abstract idea, and is therefore also ineligible under 35 U.S.C 101. See MPEP 2106.05.
Claim 7 is not patent eligible.
Regarding Claim 8:
Claim 8 recites
The method of claim 1, wherein the one or more media objects comprise video segments that are selected by a multi-class media object selector and combined into a dynamic video response for delivery to the user.
This provides a further description of the data within the abstract idea, as discussed with regards to claim 1.
Claim 8 does not include any additional elements (i.e. elements other than the abstract idea) that integrate the abstract idea into a practical application or amount to significantly more than the abstract idea, and is therefore also ineligible under 35 U.S.C 101. See MPEP 2106.05.
Claim 8 is not patent eligible.
Regarding Claim 9:
Claim 9 recites
The method of claim 1, wherein the input is a text string, and wherein the feature extracted from the input is an emotion associated with the text string.
This provides a further description of the data within the abstract idea, as discussed with regards to claim 1.
The limitation of dependent claim 9, as drafted, is a process that, under broadest reasonable interpretation, covers performance of the limitation in the mind. A user could easily, with the use of pen and paper, extract an emotion feature from the input text string. This is considered as Mental process under Abstract ideas as they can be performed in the human mind, including concepts, observation, evaluation, judgement and opinion.
Claim 9 does not include any additional elements (i.e. elements other than the abstract idea) that integrate the abstract idea into a practical application or amount to significantly more than the abstract idea, and is therefore also ineligible under 35 U.S.C 101. See MPEP 2106.05.
Claim 9 is not patent eligible.
Regarding Claim 10:
Claim 10 recites
A non-transitory computer-readable medium comprising instructions executable by a processor to perform the method of claim 1.
This provides a further description of the data within the abstract idea, as discussed with regards to claim 1.
Additional element “A non-transitory computer-readable medium” is generically recited, thus amounts to mere instructions to apply the judicial exception on a generic computer as discussed in MPEP 2106.05(f).
Claim 10 does not include any additional elements (i.e. elements other than the abstract idea) that integrate the abstract idea into a practical application or amount to significantly more than the abstract idea, and is therefore also ineligible under 35 U.S.C 101. See MPEP 2106.05.
Claim 10 is not patent eligible.
Claims 11 and 13-22 are rejected according to claims 1, 3-10.
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.
Claims 1, 4, 9, 10, 11, 14, 19 and 20 are rejected under 35 U.S.C 103 as being unpatentable over Liu et al. United States Patent Application Publication US 2021/0326751 A1 in view of Turkelson et al. United States Patent Application Publication US 2020/0193206 A1 further in view of Lanzi “Fast Feature Selection with Genetic Algorithms: A Filter Approach”.
Regarding Claim 1, Liu discloses
A computer-implemented method for providing media to a user based on a feature extracted from an input of the user, the method comprising: (Para [0015] Line 3 Neural network is disclosed (i.e. computers). Para [0106] Line 1-6 FIG. 8 shows an example review page 800 for laptop 802, where review page is one example of a user interface. The user has selected to filter negative reviews. This implies an input from the user. Para [0103] Line 1-9 At Block 710 input data is processed to obtain a result which is equivalent to extracting feature from the input. Examples are given. Para [0016] Line 1-4 Input layer, an output layer and one or more intermediate layers of the neural network are equivalent to multi-layer neural network).
obtaining a multi-layer neural network comprising a neural network model pre-trained with an unlabeled training dataset and fine-tuning the neural network model by training the neural network model with the subset of the labeled dataset; (Para [0016] Line 1-4 Input layer, an output layer and one or more intermediate layers of the neural network are equivalent to multi-layer neural network. Para [0019] Line 1-14 Pretraining refers to model training with unlabeled training data and tuning refers to using supervised learning from labeled training data. Para [0021] Line 1-6 The disclosed implementation offers a mechanism for pretraining of one or more mapping layers of the model and then the pretrained layers can be tuned with a specific layer to perform a specific task using supervised learning which is equivalent to obtaining a multi-layer neural network by pre-training a neural network model with an unlabeled training dataset and fine-tuning the neural network model with a labeled dataset. Para [0029] Line 1-13 The tuning example 204 can include labeled examples where the labels identify the correct classification for example sentences is equivalent to labeled dataset which is tagged with one or more classes.)
receiving the input through a communication interface; (Para [0090] Line 1-7 The example system 600 shows client devices 610, 640 and server devices 620, 630 connected by one or more servers 650. Client devices can be mobile devices like smartphones or tablets as well as stationary devices like desktops, server devices, etc. Server also can be implemented using various types of computing devices. Basically, the entire setup can be equivalent to receiving input through communication interface.)
providing the input to the multi-layer neural network to obtain a classification vector, the classification vector having one or more entries, wherein each of the one or more entries is associated with a class of the feature; (Para [0024] Line 7-15 The lexicon encoder 104(1) and the transformer encoder 104(2) are defined to operate to produce representations (e.g. vectors). Para [0040] Line 3-13 First Input data 402 is input to the lexicon encoder 104(1) to obtain first embeddings 106 which is provided to transformer encoder 104(2) to produce second embeddings 108 which is provided to target task layer 202. The runtime prediction output can provide a distribution of probabilities for a set of defined classifications. Basically, the output 108 is a classification vector where one or more entries are associated with a class of the feature. Para [0042] Line 2-11 The Lexicon and transformer encoder maps the input layer into embedding vectors which is equivalent to providing the input to the multi-layer neural network to obtain a classification vector.)
Liu does not disclose:
Genetically filtering a labeled dataset to obtain a subset of the labeled dataset, wherein the labeled dataset and the subset each comprise data tagged with one or more classes;
based on the classification vector, selecting one or more media objects from a plurality of media objects for delivery to the user.
Turkelson disclose:
based on the classification vector, selecting one or more media objects from a plurality of media objects for delivery to the user. (Para [0027] Line 3-6 The context classification model outputs a classification vector. Para [0061] Line 12-19 The classification vector serves as an input to the object recognition model which is equivalent to the step of basing on the classification vector. The output of the object recognition model may be one or more object identifiers indicating objects within a given image which is equivalent to selecting one or more objects from a plurality of media objects. Para [0052] Line 8-14 Computer system 102 from system 100 of Fig. 1 maybe a mobile computing device which is same or similar to mobile computing device 104. 102 may also refer to a server-side system that receives data from one or more devices and outputs data to the devices which is equivalent to delivery to user.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the training of machine learning models taught by Liu to include Turkelson’s classification vector method to select one or more media objects for delivery to user. The motivation for doing so would have been to provide enhanced performance as the classification vector’s representation can be used to optimize the model’s ability to made accurate predictions. (Turkelson, Para [0061]).
The combination of Liu and Turkelson does not appear to explicitly disclose:
Genetically filtering a labeled dataset to obtain a subset of the labeled dataset, wherein the labeled dataset and the subset each comprise data tagged with one or more classes;
Lanzi teaches:
Genetically filtering a labeled dataset to obtain a subset of the labeled dataset, wherein the labeled dataset and the subset each comprise data tagged with one or more classes; (Introduction Section, discloses a dataset that contains a plurality of attributes/features (labeled dataset) and using a genetic algorithm to filter said dataset. Experimental Results Section (Page 3) discloses generating a subset that also has labeled data)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the training of machine learning models taught by Liu and Turkelson to include Lanzi’s filtering using a genetic algorithm method. The motivation for doing so would have been to provide faster results without loss of predictive accuracy. (Lanzi, Abstract)
Regarding Claim 4, Liu in view of Turkelson and Lanzi discloses the method of Claim 1. Liu additionally discloses wherein the labeled training dataset is smaller than the unlabeled training dataset. (Para [0036] Line 1-9 Pretraining examples 308 are selected from the unlabeled training data. Masking is applied to the pretraining examples to obtain masked pretraining examples which is equivalent to the labeled training dataset is smaller than the unlabeled training dataset.)
Regarding Claim 9, Liu in view of Turkelson and Lanzi discloses the method of Claim 1. Liu additionally discloses wherein the input is a text string, and wherein the feature extracted from the input is an emotion associated with the text string. (Para [103] Line 3-9 Different examples of input data are query, document, sentence which is equivalent to the input being a text string. When the input data is a sentence, then the output can be characterized as a sentiment of the sentence which is equivalent to extracting emotion from the text string.)
Regarding Claim 10, Liu in view of Turkelson and Lanzi discloses the method of Claim 1. Liu additionally discloses A non-transitory computer-readable medium comprising instructions executable by a processor to perform (Para [0005] Line 1-4 A computer-readable storage medium storing instructions is included which when executed by one or more processing devices performs acts.)
Regarding Claim 11, Liu discloses
A system for providing media to a user based on a feature extracted from an input of the user, the system comprising: (Para [0015] Line 3 Neural network is disclosed (i.e. computers). Para [0106] Line 1-6 FIG. 8 shows an example review page 800 for laptop 802, where review page is one example of a user interface. The user has selected to filter negative reviews. This implies an input from the user. Para [0103] Line 1-9 At Block 710 input data is processed to obtain a result which is equivalent to extracting feature from the input. Examples are given. Para [0016] Line 1-4 Input layer, an output layer and one or more intermediate layers of the neural network are equivalent to multi-layer neural network).
a communication interface for receiving the input of the user; (Para [0090] Line 1-7 The example system 600 shows client devices 610, 640 and server devices 620, 630 connected by one or more servers 650. Client devices can be mobile devices like smartphones or tablets as well as stationary devices like desktops, server devices, etc. Server also can be implemented using various types of computing devices. Basically, the entire setup can be equivalent to receiving input through communication interface.)
one or more memory storage for storing a neural network model, a plurality of media objects and training data, the training data comprising an unlabeled training dataset and a labeled training dataset, the labeled dataset including data tagged with one or more classes; and (Para [0112] Line 1-6 Memory Storage devices are mentioned. Para [0016] Line 1-4 Input layer, an output layer and one or more intermediate layers of the neural network are equivalent to multi-layer neural network. Para [0019] Line 1-14 Pretraining refers to model training with unlabeled training data and tuning refers to using supervised learning from labeled training data. Para [0021] Line 1-6 The disclosed implementation offers a mechanism for pretraining of one or more mapping layers of the model and then the pretrained layers can be tuned with a specific layer to perform a specific task using supervised learning which is equivalent to obtaining a multi-layer neural network by pre-training a neural network model with an unlabeled training dataset and fine-tuning the neural network model with a labeled dataset. Para [0029] Line 1-13 The tuning example 204 can include labeled examples where the labels identify the correct classification for example sentences is equivalent to labeled dataset which is tagged with one or more classes.)
a processor configured to: train the neural network model using the training data to obtain a multi-layer neural network, the neural network model trained in a pre- training step with the unlabeled training dataset and fine-tuned by further training with a subset of the labeled training dataset; (Para [0111] Line 5-8 Processors are mentioned. Para [0016] Line 1-4 Input layer, an output layer and one or more intermediate layers of the neural network are equivalent to multi-layer neural network. Para [0019] Line 1-14 Pretraining refers to model training with unlabeled training data and tuning refers to using supervised learning from labeled training data. Para [0021] Line 1-6 The disclosed implementation offers a mechanism for pretraining of one or more mapping layers of the model and then the pretrained layers can be tuned with a specific layer to perform a specific task using supervised learning which is equivalent to obtaining a multi-layer neural network by pre-training a neural network model with an unlabeled training dataset and fine-tuning the neural network model with a labeled dataset. Para [0029] Line 1-13 The tuning example 204 can include labeled examples where the labels identify the correct classification for example sentences is equivalent to labeled dataset which is tagged with one or more classes.)
provide the input to the multi-layer neural network to obtain a classification vector, the classification vector having one or more entries, wherein each of the one or more entries is associated with a class of the feature; (Para [0024] Line 7-15 The lexicon encoder 104(1) and the transformer encoder 104(2) are defined to operate to produce representations (e.g. vectors). Para [0040] Line 3-13 First Input data 402 is input to the lexicon encoder 104(1) to obtain first embeddings 106 which is provided to transformer encoder 104(2) to produce second embeddings 108 which is provided to target task layer 202. The runtime prediction output can provide a distribution of probabilities for a set of defined classifications. Basically, the output 108 is a classification vector where one or more entries are associated with a class of the feature. Para [0042] Line 2-11 The Lexicon and transformer encoder maps the input layer into embedding vectors which is equivalent to providing the input to the multi-layer neural network to obtain a classification vector.)
Liu does not disclose:
The subset of the labeled dataset obtained from the labeled training dataset by generic filtering.
based on the classification vector, selecting one or more media objects from a plurality of media objects for delivery to the user.
Turkelson disclose:
based on the classification vector, selecting one or more media objects from a plurality of media objects for delivery to the user. (Para [0027] Line 3-6 The context classification model outputs a classification vector. Para [0061] Line 12-19 The classification vector serves as an input to the object recognition model which is equivalent to the step of basing on the classification vector. The output of the object recognition model may be one or more object identifiers indicating objects within a given image which is equivalent to selecting one or more objects from a plurality of media objects. Para [0052] Line 8-14 Computer system 102 from system 100 of Fig. 1 maybe a mobile computing device which is same or similar to mobile computing device 104. 102 may also refer to a server-side system that receives data from one or more devices and outputs data to the devices which is equivalent to delivery to user.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the training of machine learning models taught by Liu to include Turkelson’s classification vector method to select one or more media objects for delivery to user. The motivation for doing so would have been to provide enhanced performance as the classification vector’s representation can be used to optimize the model’s ability to made accurate predictions. (Turkelson, Para [0061]).
The combination of Liu and Turkelson does not appear to explicitly disclose:
The subset of the labeled dataset obtained from the labeled training dataset by generic filtering.
Lanzi teaches:
The subset of the labeled dataset obtained from the labeled training dataset by generic filtering; (Introduction Section, discloses a dataset that contains a plurality of attributes/features (labeled dataset) and using a genetic algorithm to filter said dataset. Experimental Results Section (Page 3) discloses generating a subset that also has labeled data)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the training of machine learning models taught by Liu and Turkelson to include Lanzi’s filtering using a genetic algorithm method. The motivation for doing so would have been to provide faster results without loss of predictive accuracy. (Lanzi, Abstract)
Regarding Claim 14, Liu in view of Turkelson discloses the system of Claim 10. Liu additionally discloses wherein the labeled training dataset is smaller than the unlabeled training dataset. (Para [0036] Line 1-9 Pretraining examples 308 are selected from the unlabeled training data. Masking is applied to the pretraining examples to obtain masked pretraining examples which is equivalent to the labeled training dataset is smaller than the unlabeled training dataset.)
Regarding Claim 19, Liu in view of Turkelson discloses the system of Claim 10. Liu additionally discloses wherein the input is a text string, and wherein the feature extracted from the input is an emotion associated with the text string. (Para [103] Line 3-9 Different examples of input data are query, document, sentence which is equivalent to the input being a text string. When the input data is a sentence, then the output can be characterized as a sentiment of the sentence which is equivalent to extracting emotion from the text string.)
Regarding Claim 20, Liu discloses
A computer-implemented method for communicating with a user in response to a detected emotional state of the user, the method comprising: (Para [0015] Line 3 Neural network is disclosed (i.e. computers). Para [103] Line 3-9 Different examples of input data are query, document, sentence. When the input data is a sentence, then the output can be characterized as a sentiment of the sentence which is equivalent to extracting emotion from the text string. Para [0108] Line 1-9 FIG. 10 shows an example scenario 1000, where user 1002 interacts with the client device 640. This example shows that the user is communicating in response to state of the user.).
obtaining an input text string from input provided by the user; (Para [103] Line 3-9 Different examples of input data are query, document, sentence which is equivalent to the input being a text string. Para [0108] Line 1-2 FIG. 10 shows an example scenario 1000, where user 1002 interacts with the client device 640. This example shows that the input is provided by the user.)
providing the input text string to a multi-layer neural network to obtain a classification vector representing the detected emotional state of the user, (Para [0024] Line 7-15 The lexicon encoder 104(1) and the transformer encoder 104(2) are defined to operate to produce representations (e.g. vectors). Para [0040] Line 3-13 First Input data 402 is input to the lexicon encoder 104(1) to obtain first embeddings 106 which is provided to transformer encoder 104(2) to produce second embeddings 108 which is provided to target task layer 202. The runtime prediction output can provide a distribution of probabilities for a set of defined classifications. Basically, the output 108 is a classification vector where one or more entries are associated with a class of the feature. Para [0042] Line 2-11 The Lexicon and transformer encoder maps the input layer into embedding vectors which is equivalent to providing the input to the multi-layer neural network to obtain a classification vector. Para [103] Line 3-9 Different examples of input data are query, document, sentence. When the input data is a sentence, then the output can be characterized as a sentiment of the sentence which is equivalent to extracting emotion from the text string)
the multi- layer neural network obtained by training a neural network model with a first dataset and fine-tuning the neural network model with a second dataset, the second dataset comprising data tagged with one or more classes of emotion; (Para [0016] Line 1-4 Input layer, an output layer and one or more intermediate layers of the neural network are equivalent to multi-layer neural network. Para [0019] Line 1-14 Pretraining refers to model training with unlabeled training data and tuning refers to using supervised learning from labeled training data. Para [0021] Line 1-6 The disclosed implementation offers a mechanism for pretraining of one or more mapping layers of the model and then the pretrained layers can be tuned with a specific layer to perform a specific task using supervised learning which is equivalent to obtaining a multi-layer neural network by training a neural network model with a first dataset and fine-tuning the neural network model with a second dataset. Para [0029] Line 1-13 The tuning example 204 can include labeled examples where the labels identify the correct classification for example sentences is equivalent to second dataset which is tagged with one or more classes. Para [103] Line 3-9 Different examples of input data are query, document, sentence. When the input data is a sentence, then the output can be characterized as a sentiment of the sentence which is equivalent to extracting emotion from the text string.)
receiving the input through a communication interface; (Para [0090] Line 1-7 The example system 600 shows client devices 610, 640 and server devices 620, 630 connected by one or more servers 650. Client devices can be mobile devices like smartphones or tablets as well as stationary devices like desktops, server devices, etc. Server also can be implemented using various types of computing devices. Basically, the entire setup can be equivalent to receiving input through communication interface.)
Liu does not disclose:
Wherein the second dataset comprises a subset of a labelled dataset, the subset obtained from the labelled dataset by generic filtering.
based on the classification vector, selecting one or more media objects from a library of media objects and communicating the selected one or more media objects to the user.
Turkelson disclose:
based on the classification vector, selecting one or more media objects from a library of media objects and communicating the selected one or more media objects to the user. (Para [0027] Line 3-6 The context classification model outputs a classification vector. Para [0061] Line 12-19 The classification vector serves as an input to the object recognition model which is equivalent to the step of basing on the classification vector. The output of the object recognition model may be one or more object identifiers indicating objects within a given image which is equivalent to selecting one or more objects from a plurality of media objects. Para [0052] Line 8-14 Computer system 102 from system 100 of Fig. 1 maybe a mobile computing device which is same or similar to mobile computing device 104. 102 may also refer to a server-side system that receives data from one or more devices and outputs data to the devices which is equivalent to delivery to user.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the training of machine learning models taught by Liu to include Turkelson’s classification vector method to select one or more media objects for delivery to user. The motivation for doing so would have been to provide enhanced performance as the classification vector’s representation can be used to optimize the model’s ability to made accurate predictions. (Turkelson, Para [0061]).
The combination of Liu and Turkelson does not appear to explicitly disclose:
Wherein the second dataset comprises a subset of a labelled dataset, the subset obtained from the labelled dataset by generic filtering.
Lanzi teaches:
Wherein the second dataset comprises a subset of a labelled dataset, the subset obtained from the labelled dataset by generic filtering. (Introduction Section, discloses a dataset that contains a plurality of attributes/features (labeled dataset) and using a genetic algorithm to filter said dataset. Experimental Results Section (Page 3) discloses generating a subset that also has labeled data)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the training of machine learning models taught by Liu and Turkelson to include Lanzi’s filtering using a genetic algorithm method. The motivation for doing so would have been to provide faster results without loss of predictive accuracy. (Lanzi, Abstract)
Claims 2 and 12 are rejected under 35 U.S.C 103 as being unpatentable over Liu et al. United States Patent Application Publication US 2021/0326751 A1 in view of Turkelson et al. United States Patent Application Publication US 2020/0193206 A1 in view of Lanzi “Fast Feature Selection with Genetic Algorithms: A Filter Approach” in view of Pasupalak et al. United Sates Patent Application Publication US 2015/0066479 A1.
Regarding Claim 2,
The combination of Liu in view of Turkelson and Lanzi does not disclose:
wherein the neural network model is genetically trained with the labeled dataset to obtain a subset of the labeled dataset, and wherein the subset of the labeled dataset is used for fine-tuning the neural network model.
Pasupalak discloses wherein the neural network model is genetically trained with the labeled dataset to obtain a subset of the labeled dataset, and wherein the subset of the labeled dataset is used for fine-tuning the neural network model. (Para [0100] Line 3-7 Module 610 is a 2-layer neural network. Fig. 6 displays the neural network. Para [0117] Line 1-5 Labeled data 708 is passed along to genetic algorithm 704. The output of the genetic algorithm is used as training data to optimize the operation of conditional random fields 714 and 716. Para [0129] Line 1-5 Genetic algorithm 704 working on the labeled corpus 708 to get a general feature set 704. (The labels are mislabeled in Para [0129], so have following the figure nomenclature according to the Fig. 7). Basically, the above process is equivalent to genetically training the neural network with the labeled dataset to get a subset of the labeled dataset. Figure 20 also shows the training process with genetic algorithm.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to modify the training of machine learning models taught by Liu to include genetic training based on the teachings of Pasupalak. The motivation for doing so would have been to determine areas to perform additional training to increase the accuracy of the trained neural network (Pasupalak, Para [0119]).
Regarding Claim 12,
The combination of Liu in view of Turkelson and Lanzi does not disclose:
wherein the processor is configured to genetically train the neural network model with the labeled dataset to obtain a subset of the labeled dataset, and wherein the subset of the labeled dataset is used in the fine-tuning of the neural network model.
Pasupalak discloses wherein the processor is configured to genetically train the neural network model with the labeled dataset to obtain a subset of the labeled dataset, and wherein the subset of the labeled dataset is used in the fine-tuning of the neural network model. (Para [0100] Line 3-7 Module 610 is a 2-layer neural network. Fig. 6 displays the neural network. Para [0117] Line 1-5 Labeled data 708 is passed along to genetic algorithm 704. The output of the genetic algorithm is used as training data to optimize the operation of conditional random fields 714 and 716. Para [0129] Line 1-5 Genetic algorithm 704 working on the labeled corpus 708 to get a general feature set 704. (The labels are mislabeled in Para [0129], so have following the figure nomenclature according to the Fig. 7). Basically, the above process is equivalent to genetically training the neural network with the labeled dataset to get a subset of the labeled dataset. Figure 20 also shows the training process with genetic algorithm.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the to modify the training of machine learning models taught by Liu to include genetic training based on the teachings of Pasupalak. The motivation for doing so would have been to determine areas to perform additional training to increase the accuracy of the trained neural network (Pasupalak, Para [0119]).
Claims 5, 6, 15 and 16 are rejected under 35 U.S.C 103 as being unpatentable over Liu et al. United States Patent Application Publication US 2021/0326751 A1 in view of Turkelson et al. United States Patent Application Publication US 2020/0193206 A1 in view of Lanzi “Fast Feature Selection with Genetic Algorithms: A Filter Approach” in view of Delvin et al. “Pre-Training of Deep Bidirectional Transformers for Language Understanding”.
Regarding Claim 5,
The combination of Liu in view of Turkelson and Lanzi does not disclose:
wherein the pre-training comprises bidirectional training by applying a missing words mask to the unlabeled dataset
Delvin discloses wherein the pre-training comprises bidirectional training by applying a missing words mask to the unlabeled dataset. (Page 4174 Col 1 Line 34-36 Section 3.1 Pre-training BERT talks about pre-train BERT using two unsupervised tasks while the pre training process. Page 4174 Col 2 Line 1-5 Masked Language Model (MLM). Bidirectional training is done by masking some percentage of the input tokens at random and then predict those masked tokens which is equivalent to applying a missing words mask to the unlabeled dataset. Page 4173 Figure. 1 shows pre-training comprising bidirectional training by applying missing works mask to the unlabeled dataset.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified trained neural network model taught by Liu to include bidirectional training by applying a missing words mask based on the teachings of Delvin. The motivation for doing so would have been to improve the performance of the neural network model, make more accurate predictions and help in a more comprehensive understanding of the language. (Delvin, Page 4174).
Regarding Claim 6,
Liu additionally discloses wherein the pre-training comprises training through sentence prediction (Para [0094] Line 12-16 The pretraining stage 302 can involve performing next sentence prediction on unsupervised learning data.)
Regarding Claim 15,
The combination of Liu in view of Turkelson and Lanzi does not disclose:
wherein the pre-training step comprises bidirectional training by applying a missing words mask to the unlabeled dataset.
Delvin discloses wherein the pre-training step comprises bidirectional training by applying a missing words mask to the unlabeled dataset. (Page 4174 Col 1 Line 34-36 Section 3.1 Pre-training BERT talks about pre-train BERT using two unsupervised tasks while the pre training process. Page 4174 Col 2 Line 1-5 Masked Language Model (MLM). Bidirectional training is done by masking some percentage of the input tokens at random and then predict those masked tokens which is equivalent to applying a missing words mask to the unlabeled dataset. Page 4173 Figure. 1 shows pre-training comprising bidirectional training by applying missing works mask to the unlabeled dataset.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified trained neural network model taught by Liu to include bidirectional training by applying a missing words mask based on the teachings of Delvin. The motivation for doing so would have been to improve the performance of the neural network model, make more accurate predictions and help in a more comprehensive understanding of the language. (Delvin, Page 4174).
Regarding Claim 16,
Liu additionally discloses wherein the pre-training step further comprises training through sentence prediction (Para [0094] Line 12-16 The pretraining stage 302 can involve performing next sentence prediction on unsupervised learning data.)
Claims 7 and 17 are rejected under 35 U.S.C 103 as being unpatentable over Liu et al. United States Patent Application Publication US 2021/0326751 A1 in view of Turkelson et al. United States Patent Application Publication US 2020/0193206 A1 in view of Lanzi “Fast Feature Selection with Genetic Algorithms: A Filter Approach” in view of Husain et al. United States Patent Application Publication US 2021/0117803 A1.
Regarding Claim 7,
The combination of Liu in view of Turkelson and Lanzi does not disclose:
wherein the fine-tuning comprises training through back-propagation.
Husain discloses wherein the fine-tuning comprises training through back-propagation. (Para [0005] Line 1-7 The present application describes training a neural network with systems and methods that utilize backpropagation. Para [0027] Line 10-14 FIG. 1A The system 100 includes a genetic algorithm 110 and a backpropagation trainer 180. The output of the genetic algorithm serves as an input to the backpropagation trainer which is equivalent to fine tuning training through backpropagation.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the trained neural network model taught by Liu to include backpropagation method based on the teachings of Husain. The motivation for doing so would have been to achieve high accuracy in the multi-layer neural network model, efficient work on large data and improve prediction accuracy on unseen examples specially in image classification applications. (Husain, Para [0005]).
Regarding Claim 17,
The combination of Liu in view of Turkelson and Lanzi does not disclose:
wherein the fine-tuning of the neural network model comprises training through back-propagation.
Husain discloses wherein the fine-tuning of the neural network model comprises training through back-propagation. (Para [0005] Line 1-7 The present application describes training a neural network with systems and methods that utilize backpropagation. Para [0027] Line 10-14 FIG. 1A The system 100 includes a genetic algorithm 110 and a backpropagation trainer 180. The output of the genetic algorithm serves as an input to the backpropagation trainer which is equivalent to fine tuning training through backpropagation.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the trained neural network model taught by Liu to include backpropagation method based on the teachings of Husain. The motivation for doing so would have been to achieve high accuracy in the multi-layer neural network model, efficient work on large data and improve prediction accuracy on unseen examples specially in image classification applications. (Husain, Para [0005]).
Claims 8 and 18 are rejected under 35 U.S.C 103 as being unpatentable over Liu et al. United States Patent Application Publication US 2021/0326751 A1 in view of Turkelson et al. United States Patent Application Publication US 2020/0193206 A1 in view of Lanzi “Fast Feature Selection with Genetic Algorithms: A Filter Approach” in view of Koumchatzky et al. United States Patent Application Publication US 2020/0021866 A1.
Regarding Claim 8,
The combination of Liu in view of Turkelson and Lanzi does not disclose:
wherein the one or more media objects comprise video segments that are selected by a multi-class media object selector and combined into a dynamic video response for delivery to the user.
Koumchatzky discloses wherein the one or more media objects comprise video segments that are selected by a multi-class media object selector and combined into a dynamic video response for delivery to the user. (FIG. 1 is a schematic diagram of a system 100 for identifying and displaying live video broadcast previews. Para [0036] Line 1-7 The interactive streaming application 132 may be configured to use various components of the computing device 102 or components in communication with the computing device 102 to capture and send real-time video stream and to display interactions from viewers which is equivalent to dynamic video response to the user. Para [0043] Line 9-20 The classification data may be generated by one or more classifiers 176. Examples are possible classes are specified. Para [0045] Line 6-23 The video sharing engine 170 includes a video streaming engine 172 and a video discovery engine 174. The video streaming engine 172 may be configured to provide the transcoded video stream, included any added engagement indications, to client viewing devices in a format appropriate for a particular client viewing. Para [0048] Line 1-17 The video discovery engine 174 may enable the interactive video sharing engine 170 to suggest real-time video streams, stored video streams, or pre-recorded video files to a viewer. The video discovery engine 174 may use the broadcast metadata 16 and the classifiers 176 to provide real time video streams to a user. Basically, function of the interactive video sharing engine 170 with video streaming engine 172 and video discovery engine 174 is equivalent to selecting one or more media objects comprising video segments by a multi-class media object selector and combining into a dynamic video response for delivery to the user.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the trained neural network model taught by Liu to include multi-class media object selector and dynamic video response to the user based on the teachings of Koumchatzky. The motivation for doing so would have been to efficiently provide the user with selected video material that suits their interests. (Koumchatzky, Para [0003]).
Regarding Claim 18,
The combination of Liu in view of Turkelson and Lanzi does not disclose:
wherein the media objects comprise video segments that are combinable into a dynamic video response for delivery to the user.
Koumchatzky discloses wherein the media objects comprise video segments that are combinable into a dynamic video response for delivery to the user. (FIG. 1 is a schematic diagram of a system 100 for identifying and displaying live video broadcast previews. Para [0036] Line 1-7 The interactive streaming application 132 may be configured to use various components of the computing device 102 or components in communication with the computing device 102 to capture and send real-time video stream and to display interactions from viewers which is equivalent to dynamic video response to the user. Para [0043] Line 9-20 The classification data may be generated by one or more classifiers 176. Examples are possible classes are specified. Para [0045] Line 6-23 The video sharing engine 170 includes a video streaming engine 172 and a video discovery engine 174. The video streaming engine 172 may be configured to provide the transcoded video stream, included any added engagement indications, to client viewing devices in a format appropriate for a particular client viewing. Para [0048] Line 1-17 The video discovery engine 174 may enable the interactive video sharing engine 170 to suggest real-time video streams, stored video streams, or pre-recorded video files to a viewer. The video discovery engine 174 may use the broadcast metadata 16 and the classifiers 176 to provide real time video streams to a user. Basically, function of the interactive video sharing engine 170 with video streaming engine 172 and video discovery engine 174 is equivalent to the media objects comprising video segments that are combinable into a dynamic video response for delivery to the user.)
Before the time of the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art to have modified the trained neural network model taught by Liu to include multi-class media object selector and dynamic video response to the user based on the teachings of Koumchatzky. The motivation for doing so would have been to efficiently provide the user with selected video material that suits their interests. (Koumchatzky, Para [0003]).
Allowable Subject Matter
Claim 3, 13, 21 and 22 include allowable subject matter, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims and overcoming the 35 USC 101 rejection.
Response to Arguments
Claim Rejections - 35 USC § 101
Applicant argues “It can be seen that the claimed subject matter provides technical improvements over conventional systems”
Examiner Respectfully disagrees, benefits described only in the specification are not sufficient; the improvement must be reflected in the claims themselves. The independent claims do not recite how the genetic filtering is performed in a manner that improves the functioning of the computer (e.g., no claimed memory usage reduction technique, no new data structure, no specialized hardware pipeline). The recited steps describe desired results (subset selection; fine-tuning; classification; selection of media) using generic computing. This is insufficient under MPEP 2106.05(a) and (d). Examiner would also like to point out that “It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements. (MPEP 2106.05(a)” In this case applicant argues that the improvement lies on the genetically filtering a labeled dataset, this limitation has been analyzed above as an abstract idea therefore the improvement is currently being claimed as part of the judicial exception.
Claim Rejections - 35 USC § 103
Applicant’s arguments with respect to the claim(s) 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 MARIELA D REYES whose telephone number is (571)270-1006. The examiner can normally be reached Monday-Friday, 7:30 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, David Wiley can be reached at 571-272-4150. 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.
/Mariela Reyes/Supervisory Patent Examiner, Art Unit 2142
Prosecution Insights