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 in response to an application filed on March 7th, 2023. Claims 1-20 are pending in the current application.
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.
Claim(s) 1-10 are rejected under 35 U.S.C. 101 because the claimed invention is directed
to an abstract idea without significantly more.
Regarding claims 1-10, Under Step 1 of the Subject Matter Eligibility Test of Products
and Processes, the claims are directed towards software per se, which does not fall within at
least one of the four categories of patent eligible subject matter and so fails Step 1.
Specifically, claims 1-10 is directed to a program having a program code. As recited, the
claims amount to a computer program per se claimed as a product without any structural
recitation. In accordance with MPEP 2106.03, software expressed as code or a set of instructions detached from any medium is an idea without physical embodiment. See Microsoft Corp. v. AT&T Corp., 550 U.S. 437, 449, 82 USPQ2d 1400, 1407 (2007); see also Benson, 409 U.S. 67, 175 USPQ2d 675 (An "idea" is not patent eligible). Thus, a product claim to a software program that does not also contain at least one structural limitation has no physical or tangible
form, and thus does not fall within any statutory category.
Claim Rejections - 35 USC § 102
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1, 3-7, 11, 12, 14, 15, and 17-20 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Yamaguchi et al. (U.S. Patent Application No. US 20210334706 A1)
Regarding claim 1, Yamaguchi teaches a model training system, comprising at least one first cluster and a second cluster communicating with the at least first cluster, wherein: the at least one first cluster is configured to acquire a sample data set (“The augmentation apparatus 10 uses an outer dataset 40 to perform data augmentation of a target dataset 30 and output an augmented dataset 50.”, Paragraph 26) (The outer dataset corresponds to the claimed sample data set.) generate training data according to the sample data set and send the training data to the second cluster (According to Fig. 1, the augmented dataset (training dataset) is output by the augmentation apparatus, the augmentation apparatus corresponding to the first cluster, and then the augmented dataset is sent to the learning apparatus, corresponding to the second cluster, and so teaching the limitation.) and the second cluster is configured to train a pre-trained model according to the training data sent by the at least one first cluster. (“The augmentation apparatus 10 uses an outer dataset 40 to perform data augmentation of a target dataset 30 and output an augmented dataset 50. In addition, the learning apparatus 20 has a target model 21 to perform learning by using the augmented dataset 50.” Paragraph 26)
Regarding claim 3, Yamaguchi teaches the system of claim 1, wherein the at least one first cluster and the second cluster are heterogeneous clusters. (“The control unit 13 controls the entire augmentation apparatus 10. The control unit 13 may be an electronic circuit such as a Central Processing Unit (CPU) or a Micro Processing Unit (MPU), or an integrated circuit such as an Application Specific Integrated Circuit (ASIC) or a Field Programmable Gate Array (FPGA)… all or any part of each processing function to be performed by each apparatus can be implemented by a CPU and a program being analyzed and executed by the CPU, or can be implemented as hardware by wired logic… the augmentation apparatus 10 can be implemented as an augmentation server apparatus that has a terminal apparatus used by a user as a client and provides services regarding the above-described data augmentation to the client. For example, the augmentation server apparatus is implemented as a server apparatus that provides an augmentation service in which target data is input and augmented data is output.”, Paragraphs 35, 73, and 77) (Both the augmentation and learning apparatus contain processing, which can be performed by heterogenous processing units, such as the augmentation on a server and the learning on a host computer.)
Regarding claim 4, Yamaguchi teaches the system of claim 3, wherein a processor adopted by the at least one first cluster is different from a processor adopted by the second cluster. (“the augmentation server apparatus may be implemented as a web server or may be implemented as a cloud that provides services regarding the data augmentation through outsourcing.”, Paragraph 77) (In this embodiment, the server, acting as the first cluster, can augment the data on one processor and send the data to a personal computer, acting as our second cluster, and train a model using a different processor.)
Regarding claim 5, Yamaguchi teaches the system of claim 4, wherein the processor adopted by the at least one first cluster is a graphics processor, and the processor adopted by the second cluster is an embedded neural network processor. (“As illustrated in FIG. 2, the generative model 121 has a generator 121a and a distinguisher 121b. For example, all of the generator 121a and the distinguisher 121b are neural networks… Further, all or any part of each processing function to be performed by each apparatus can be implemented by a CPU and a program being analyzed and executed by the CPU, or can be implemented as hardware by wired logic.”, Paragraphs 32 and 73) (A processor which can run a neural network is considered to be a neural network processor, and while a GPU is never explicitly taught, a GPU is considered to be “hardware by wired logic” and so can be easily configured to work with Yamaguchi’s method.)
Regarding claim 6, Yamaguchi teaches the system of claim 1, wherein the at least one first cluster comprises a plurality of first clusters, and data types processed by the plurality of first clusters are different. (“a specific form of distribution and integration of each apparatus is not limited to the form illustrated in the drawings, and all or some of the apparatuses can be distributed or integrated functionally or physically in any units according to various loads and use situations. Further, all or any part of each processing function to be performed by each apparatus can be implemented by a CPU and a program being analyzed and executed by the CPU, or can be implemented as hardware by wired logic.”, Paragraph 73)
Regarding claim 7, Yamaguchi teaches the system of claim 1, wherein the at least one first cluster is configured to: input the sample data set into an initial generator to generate the training data and train the initial generator according to the sample data set to obtain a trained generator (“The learning unit 131 first performs pre-processing on each piece of the data. For example, the learning unit 131 changes the size of an image to a uniform size (e.g. 128×128 pixels) as pre-processing. Then, the learning unit 131 combines the datasets S.sub.target and S.sub.outer, and generates a dataset S.sub.t+o. For example, S.sub.t+o has the data and the label of S.sub.target and S.sub.outer stored in the same sequence, respectively. Then, the learning unit 131 causes the generative model 121 to learn the generated dataset S.sub.t+o as a correct dataset.”, Paragraphs 45 and 46) and wherein the second cluster, when training the pre-trained model according to the training data sent by the at least one first cluster, is configured to: train an initial discriminator according to the training data to obtain a trained discriminator. (“In addition, each dataset in FIG. 1 is data with a label to be used by the target model 21. That is, each dataset is a combination of data and a label. For example, if the target model 21 is a model for image recognition, each dataset is a combination of image data and a label. In addition, the target model 21 may be a speech recognition model or a natural language recognition model. In such a case, each dataset is speech data with a label or text data with a label.”, Paragraph 27) (The target model, located in the second cluster, is training to be a discriminator by predicting a label.)
Regarding claim 11, Yamaguchi teaches a model training method, performed by a first cluster communicatively connected to a second cluster, comprising: acquiring a sample data set; (“The augmentation apparatus 10 uses an outer dataset 40 to perform data augmentation of a target dataset 30 and output an augmented dataset 50.”, Paragraph 26) (The outer dataset corresponds to the sample data set.) generating training data according to the sample data set (According to Fig. 1, the augmented dataset (training dataset) is output by the augmentation apparatus, the augmentation apparatus corresponding to the first cluster, and then the augmented dataset is sent to the learning apparatus, corresponding to the second cluster, and so teaching the limitation.) and sending the training data to the second cluster for the second cluster to train a pre-trained model according to the training data. (“The augmentation apparatus 10 uses an outer dataset 40 to perform data augmentation of a target dataset 30 and output an augmented dataset 50. In addition, the learning apparatus 20 has a target model 21 to perform learning by using the augmented dataset 50.” Paragraph 26)
Regarding claim 12, Yamaguchi teaches the method of claim 11, wherein generating the training data according to the sample data set comprises: inputting the sample data set into an initial generator to generate the training data and training the initial generator according to the sample data set to obtain a trained generator. (“The learning unit 131 first performs pre-processing on each piece of the data. For example, the learning unit 131 changes the size of an image to a uniform size (e.g. 128×128 pixels) as pre-processing. Then, the learning unit 131 combines the datasets S.sub.target and S.sub.outer, and generates a dataset S.sub.t+o. For example, S.sub.t+o has the data and the label of S.sub.target and S.sub.outer stored in the same sequence, respectively. Then, the learning unit 131 causes the generative model 121 to learn the generated dataset S.sub.t+o as a correct dataset.”, Paragraphs 45 and 46)
Regarding claim 14, Yamaguchi teaches a model training method, performed by a second cluster communicatively connected to at least one first cluster, comprising: receiving training data sent by the at least one first cluster; (According to Fig. 1, the augmented dataset (training dataset) is output by the augmentation apparatus, the augmentation apparatus corresponding to the first cluster, and then the augmented dataset is sent to the learning apparatus, corresponding to the second cluster, and so teaching the limitation.) and training a pre-trained model according to the training data. (“the learning apparatus 20 has a target model 21 to perform learning by using the augmented dataset 50.” Paragraph 26) (The target model corresponds to a pre-trained model, with the training being performed via the learning apparatus, which corresponds to the second cluster. The training data corresponds to the augmented dataset from the first cluster.)
Regarding claim 15, Yamaguchi teaches the method of claim 14, wherein training the pre-trained model according to the training data comprises: training an initial discriminator according to the training data to obtain a trained discriminator. (“In addition, each dataset in FIG. 1 is data with a label to be used by the target model 21. That is, each dataset is a combination of data and a label. For example, if the target model 21 is a model for image recognition, each dataset is a combination of image data and a label. In addition, the target model 21 may be a speech recognition model or a natural language recognition model. In such a case, each dataset is speech data with a label or text data with a label.”, Paragraph 27) (The target model, located in the second cluster, is training to be a discriminator by predicting a label.)
Regarding claim 17, Yamaguchi teaches an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory is stored with instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to perform steps of the method of claim 11. (“FIG. 10 is a diagram illustrating an example of a computer executing an augmentation program. The computer 1000 includes, for example, a memory 1010 and a CPU 1020. The computer 1000 includes a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These units are connected by a bus 1080.”, Paragraph 78, See also Fig. 10)
Regarding claim 18, Yamaguchi teaches a non-transitory computer-readable storage medium having stored therein computer instructions that, when executed by a computer, cause the computer to perform steps of the method of claim 11. (“Note that the program module 1093 or the program data 1094 is not limited to being stored in the hard disk drive 1090, and may be stored in, for example, a removable storage medium, and read by the CPU 1020 via the disk drive 1100 or the like.”, Paragraph 82, See also Fig. 10 and Paragraph 79)
Regarding claim 19, Yamaguchi teaches an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory is stored with instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to perform steps of the method of claim 14. (“FIG. 10 is a diagram illustrating an example of a computer executing an augmentation program. The computer 1000 includes, for example, a memory 1010 and a CPU 1020. The computer 1000 includes a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These units are connected by a bus 1080.”, Paragraph 78, See also Fig. 10)
Regarding claim 20, Yamaguchi teaches a non-transitory computer-readable storage medium having stored therein computer instructions that, when executed by a computer, cause the computer to perform steps of the method of claim 14. (“Note that the program module 1093 or the program data 1094 is not limited to being stored in the hard disk drive 1090, and may be stored in, for example, a removable storage medium, and read by the CPU 1020 via the disk drive 1100 or the like.”, Paragraph 82, See also Fig. 10 and Paragraph 79)
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) 2 and 8 are rejected under 35 U.S.C. 103 as being unpatentable over Yamaguchi in view of Claude COULOMBE. (Herein referred to as Coulombe) (Text Data Augmentation Made Simple By Leveraging NLP Cloud APIs)
Regarding claim 2, Yamaguchi teaches the system of claim 1, but does not explicitly teach nodes inside the at least one first cluster communicate with each other via a first bandwidth, nodes inside the second cluster communicate with each other via a second bandwidth, and the at least one first cluster and the second cluster communicate with each other via a third bandwidth, wherein the first bandwidth is greater than the third bandwidth, and the second bandwidth is greater than the third bandwidth.
Coulombe teaches text data augmentation within a cloud API setting which has nodes inside the at least one first cluster communicate with each other via a first bandwidth, (“Although SyntaxNet is rather difficult to install on a server or local machine, Google's open-source license allows you to do that. Three ways are possible: docker image installation provided by Google [Docker, 2016], installation via a pre-configured server image (like AMI: Amazon Machine image) or installation on rented servers on a commercial cloud infrastructure from the source code of Google.”, pg. 26, fourth paragraph) (The server corresponds to the first bandwidth, as a server innately has a bandwidth associated with it.) nodes inside the second cluster communicate with each other via a second bandwidth, and the at least one first cluster and the second cluster communicate with each other via a third bandwidth wherein the first bandwidth is greater than the third bandwidth, and the second bandwidth is greater than the third bandwidth. (While not explicitly taught, in cloud-based computing, there are host PC’s which would act as our second bandwidth, with communications between the host and server acting as our third bandwidth, and it is commonly known in the art that communication within a PC and server is greater than communication BETWEEN a PC and server.)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the data augmentation of Yamaguchi, with the text augmentation of Coulombe. One would be motivated to combine the two teachings, prior to the filing date of the current application, as this allows for the use of SyntaxNet, which parses sentence for machine learning purposes as disclosed in Coulombe. (“The paraphrases generator proceeds sentence by sentence. Each sentence is parsed by SyntaxNet, an open-source library from Google that uses deep learning techniques and the TensorFlow library.”, pg. 3, third paragraph; See also the Figure on pg. 13 for how it is used)
Regarding claim 8, Yamaguchi teaches the system of claim 7, as well as the second cluster, when training the initial discriminator according to the training data to obtain the trained discriminator, is configured to: train the initial discriminator according to the second text sample data to obtain the trained discriminator. (“In addition, each dataset in FIG. 1 is data with a label to be used by the target model 21. That is, each dataset is a combination of data and a label. For example, if the target model 21 is a model for image recognition, each dataset is a combination of image data and a label. In addition, the target model 21 may be a speech recognition model or a natural language recognition model. In such a case, each dataset is speech data with a label or text data with a label.”, Paragraph 27) (The target model, located in the second cluster, is training to be a discriminator by predicting a label.)
However, Yamaguchi does not teach wherein the sample data set is a first text sample data set, and the at least one first cluster is configured to: replace a text segment in the first text sample data set with a set identifier to obtain a replaced first text sample data set, and input the replaced first text sample data set into the initial generator to obtain second text sample data;
Coulombe teaches to the sample data set is a first text sample data set, to: replace a text segment in the first text sample data set with a set identifier to obtain a replaced first text sample data set, (“Generally, there is no replacement for grammatical words. Here, in order of increasing difficulty, the types of words that are candidates for lexical substitution: adverbs, adjectives, nouns and verbs. Verbs replacement is particularly challenging because of the different arguments that accompany the verbs. In many situations, we limit to replace only adverbs and adjectives, sometimes we add nouns, more rarely verbs”, pg. 8, under “3.5 Technique 3 - Word Replacement using thesaurus”) and input the replaced first text sample data set into the initial generator to obtain second text sample data; (“From the original sentence as input, a syntactic parser builds a dependency tree. Then the paraphrases generator transforms this dependency tree to create a transformed dependency tree guided by a transformation grammar. The transformed dependency tree is then used to generate a new surface form, i.e. paraphrase.”, pg. 13, first paragraph)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the data augmentation of Yamaguchi, with the text augmentation of Coulombe. One would be motivated to combine the two teachings, prior to the filing date of the current application, as this allows for the generation of augmented text data that avoids overfitting, as disclosed in Coulombe. (“Data augmentation can also be considered as a regularization technique since that is used to avoid overfitting… This piece of work, which is more an engineering endeavor than a fundamental research, is in the vein of the methods used in computer vision which consist of applying invariant transformations to original data in order to generate augmentation data.”, pg. 2, sixth paragraph; pg. 4, third paragraph)
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Yamaguchi in view of Lin Wu et al. (Herein referred to as Wu) (Cross-Entropy Adversarial View Adaptation for Person Re-identification)
Regarding claim 10, Yamaguchi teaches the system of claim 7, wherein the second cluster is configured to: input an initial discriminant parameter into a convolutional neural network to establish the initial discriminator; (“the learning apparatus 20 has a target model 21 to perform learning by using the augmented dataset 50. The target model 21 may be a known model for performing machine learning. For example, the target model 21 is MCCNN with Triplet loss…the target dataset 30, was input into the target model 21.” Paragraph 26 and 67) (The target model takes the target dataset as parameters to establish the discriminator, with the target model being a CNN.) input the training data into the initial discriminator for pre-training to obtain a pre-trained training data (“The learning apparatus 20 performs learning of the target model 21 using the augmented dataset 50. Paragraph 64) (The augmented dataset corresponds to training data) and update a network parameter of the initial discriminator according to the pre-trained discriminant parameter to obtain the trained discriminator. (“The learning apparatus 20 performs learning of the target model 21 using the augmented dataset 50.”, Paragraph 64) (A machine learning model learning innately consists of updating network parameters.)
However, Yamaguchi does not explicitly teach to transform the pre-trained training data into a probability output according to a probability distribution function nor update the initial discriminant parameter of the initial discriminator according to a minimized cross entropy to obtain a pre-trained discriminant parameter
Wu teaches to transform the pre-trained training data into a probability output according to a probability distribution function (“In our framework, we need to minimize the distances between the probe and gallery representations through alternating the minimization between two functions. Thereby, the probe and gallery mappings should be optimized according to a constrained adversarial objective [53]… Intuitively, the loss function in Eq.(3) is a view confusion objective, under which the mapping can be trained using a cross-entropy loss function against a uniform distribution. This loss is to ensure the adversarial discriminator will view the two domains identically.”, pg. 5, left column, under “Adversarial View Adaptation with Cross-Entropy Loss”) (Cross-Entropy Loss inherently transforms data into a probability output according to a probability distribution function.) and update the initial discriminant parameter of the initial discriminator according to a minimized cross entropy to obtain a pre-trained discriminant parameter (“we introduce adversarial learning [25] into view discriminator which is optimized through cross-entropy based view confusion objective.”, pg. 2, left column, bottom paragraph; See also Algorithm 1 on pg. 5 for the updating step.) (It would be easy to configure the method of Wu which obtains a parameter via cross-entropy, with Yamaguchi’s method which obtains it via back-propagation of errors.)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the data augmentation of Yamaguchi, with the cross-entropy loss to obtain a discriminant parameter, as taught by Wu. One would be motivated to combine the two teachings, prior to the filing date of the current application, as updating the parameters with cross-entropy can optimize the model, leading to improved performance as disclosed in Wu. (“a common problem of existing domain adaptation approaches is that a principled alignment between the source and target is missing, and thus they are unable to penalize the correlated domain misalignment in practical terms. In contrast, our method explicitly minimizes the view discrepancy through the proposed view-adversarial objective. Our method is also distinct from existing methods based on adversarial losses [22]. For instance, SPGAN [22] is composed of GAN loss to update the target domain w.r.t the source, our method instead uses cross-entropy loss to optimize the view confusion objective.”, pg. 2, left column, above “B. Our Approach and Contribution”)
Claims 9, 13, 16 are rejected under 35 U.S.C. 103 as being unpatentable over Yamaguchi in view of Coulombe and in further view of Wu.
Regarding claim 9, Yamaguchi teaches the system of claim 7, wherein the at least one first cluster is configured to: update a network parameter of the initial generator according to the pre-trained network parameter to obtain the trained generator. (“in the learning of the generative model 121, parameters of the generator 121a are updated so that the error becomes smaller.”, Paragraph 33)
However, Yamaguchi does not explicitly teach to input an initial generation parameter into a recurrent neural network to establish the initial generator; (“Specifically, the storage unit 12 stores parameters used in each processing operation by the generative model 121.”, Paragraph 31) nor ) transform the pre-trained sample data set into a probability output according to a probability distribution function to obtain a pre-trained network parameter
Coulombe teaches to input an initial generation parameter into a recurrent neural network to establish the initial generator; (“Some standard deep neural network architectures were tested: the multilayer perceptron (MLP), the long short-term memory recurrent network (LSTM) and the bidirectional LSTM (biLSTM)”, pg. 1, bottom paragraph) (One would have to input a parameter to output an initial model, such as an RNN)
Therefore it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the data augmentation of Yamaguchi, with the LSTM of Coulombe. One would be motivated to combine the two teachings, prior to the filing date of the current application, as this allows for a strong initial model with compatibility with augmented data, as disclosed in Coulombe. (“It is remarkable that for a recurrent network of LSTM type, one goes from models practically incapable of predicting anything (just a little better than the 50% due to chance) to models. able to predict in more than ⅔ cases and all this by injecting only textual noise. Not surprisingly, recurrent neural networks, especially bi-directional biLSTM network, give the best results. It is especially interesting to note that they manage to make good use of the addition of the augmented data resulting from the different technique”, pg. 24, second and third paragraphs)
However, the combination still does not teach to transform the pre-trained sample data set into a probability output according to a probability distribution function to obtain a pre-trained network parameter
Wu teaches to transform the pre-trained sample data set into a probability output according to a probability distribution function to obtain a pre-trained network parameter (“In our framework, we need to minimize the distances between the probe and gallery representations through alternating the minimization between two functions. Thereby, the probe and gallery mappings should be optimized according to a constrained adversarial objective [53]… Intuitively, the loss function in Eq.(3) is a view confusion objective, under which the mapping can be trained using a cross-entropy loss function against a uniform distribution. This loss is to ensure the adversarial discriminator will view the two domains identically.”, pg. 5, left column, under “Adversarial View Adaptation with Cross-Entropy Loss”) (Cross-Entropy Loss inherently transforms data into a probability output according to a probability distribution function.)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the teachings of Yamaguchi and Coulombe with the cross-entropy loss of Wu. One would be motivated to combine the two teachings, prior to the filing date of the current application, as Wu’s method “achieves notable
improved performance in comparison to state-of-the-arts on benchmark datasets.” (pg. 1, Abstract)
Regarding claim 13, Yamaguchi teaches the method of claim 12, wherein training the initial generator according to the sample data set to obtain the trained generator comprises: inputting an initial generation parameter into a recurrent neural network to establish the initial generator (“Specifically, the storage unit 12 stores parameters used in each processing operation by the generative model 121.”, Paragraph 31) inputting the sample data set into the initial generator for pre-training to obtain a pre-trained sample data set; (“a correct dataset is input to the generative model 121. The correct dataset is a combination of correct data and a correct label added to the correct data… The generator 121a generates generative data from the correct label input with predetermined noise.”, Paragraph 32 and 33) and updating a network parameter of the initial generator according to the pre-trained network parameter to obtain the trained generator. (“in the learning of the generative model 121, parameters of the generator 121a are updated so that the error becomes smaller.”, Paragraph 33)
However, Yamaguchi does not explicitly teach transforming the pre-trained sample data set into a probability output according to a probability distribution function to obtain a pre-trained network parameter; nor the sample data set is a first text sample data set nor inputting the replaced first text sample data set into the initial generator to obtain second text sample data
Coulombe teaches the sample data set is a first text sample data set (“Generally, there is no replacement for grammatical words. Here, in order of increasing difficulty, the types of words that are candidates for lexical substitution: adverbs, adjectives, nouns and verbs. Verbs replacement is particularly challenging because of the different arguments that accompany the verbs. In many situations, we limit to replace only adverbs and adjectives, sometimes we add nouns, more rarely verbs”, pg. 8, under “3.5 Technique 3 - Word Replacement using thesaurus”) and inputting the replaced first text sample data set into the initial generator to obtain second text sample data (“From the original sentence as input, a syntactic parser builds a dependency tree. Then the paraphrases generator transforms this dependency tree to create a transformed dependency tree guided by a transformation grammar. The transformed dependency tree is then used to generate a new surface form, i.e. paraphrase.”, pg. 13, first paragraph)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the data augmentation of Yamaguchi, with the text augmentation of Coulombe. One would be motivated to combine the two teachings, prior to the filing date of the current application, as this allows for the generation of augmented text data that avoids overfitting, as disclosed in Coulombe. (“Data augmentation can also be considered as a regularization technique since that is used to avoid overfitting… This piece of work, which is more an engineering endeavor than a fundamental research, is in the vein of the methods used in computer vision which consist of applying invariant transformations to original data in order to generate augmentation data.”, pg. 2, sixth paragraph; pg. 4, third paragraph)
However, the combination still does not teach transforming the pre-trained sample data set into a probability output according to a probability distribution function to obtain a pre-trained network parameter
Wu teaches transforming the pre-trained sample data set into a probability output according to a probability distribution function to obtain a pre-trained network parameter (“In our framework, we need to minimize the distances between the probe and gallery representations through alternating the minimization between two functions. Thereby, the probe and gallery mappings should be optimized according to a constrained adversarial objective [53]… Intuitively, the loss function in Eq.(3) is a view confusion objective, under which the mapping can be trained using a cross-entropy loss function against a uniform distribution. This loss is to ensure the adversarial discriminator will view the two domains identically.”, pg. 5, left column, under “Adversarial View Adaptation with Cross-Entropy Loss”) (Cross-Entropy Loss inherently transforms data into a probability output according to a probability distribution function.)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the teachings of Yamaguchi and Coulombe with the cross-entropy loss of Wu. One would be motivated to combine the two teachings, prior to the filing date of the current application, as Wu’s method “achieves notable
improved performance in comparison to state-of-the-arts on benchmark datasets.” (pg. 1, Abstract)
Regarding claim 16, Yamaguchi teaches the method of claim 15, as well as training the initial discriminator according to the training data to obtain the trained discriminator comprises: training the initial discriminator according to the second text sample data to obtain the trained discriminator (“the distinguisher 121b is designed to be able to recognize the generative data as generative data and recognize the correct data as correct data through learning.”, Paragraph 34) (The correct data acts as our “text sample data”) training the initial discriminator according to the training data to obtain the trained discriminator comprises: inputting an initial discriminant parameter into a convolutional neural network to establish the initial discriminator (“the distinguisher 121b are neural networks… parameters of the distinguisher 121b are updated so that the error becomes larger.”, Paragraphs 32 and 33)) inputting the training data into the initial discriminator for pre-training to obtain a pre-trained training data (“the distinguisher 121b is designed to be able to recognize the generative data as generative data and recognize the correct data as correct data through learning.”, Paragraph 34) (The “correct data” is training data for the distinguisher to learn from.) and updating a network parameter of the initial discriminator according to the pre-trained discriminant parameter to obtain the trained discriminator. (“parameters of the distinguisher 121b are updated so that the error becomes larger. Note that each of the parameters for learning is updated by using a method of backward propagation of errors”, Paragraph 33)
However, Yamaguchi does not explicitly teach the training data is second text sample data, nor updating the initial discriminant parameter of the initial discriminator according to a minimized cross entropy to obtain a pre-trained discriminant parameter
Coulombe teaches the training data is second text sample data (“Generally, there is no replacement for grammatical words. Here, in order of increasing difficulty, the types of words that are candidates for lexical substitution: adverbs, adjectives, nouns and verbs. Verbs replacement is particularly challenging because of the different arguments that accompany the verbs. In many situations, we limit to replace only adverbs and adjectives, sometimes we add nouns, more rarely verbs”, pg. 8, under “3.5 Technique 3 - Word Replacement using thesaurus”)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the data augmentation of Yamaguchi, with the text augmentation of Coulombe. One would be motivated to combine the two teachings, prior to the filing date of the current application, as this allows for the generation of augmented text data that avoids overfitting, as disclosed in Coulombe. (“Data augmentation can also be considered as a regularization technique since that is used to avoid overfitting… This piece of work, which is more an engineering endeavor than a fundamental research, is in the vein of the methods used in computer vision which consist of applying invariant transformations to original data in order to generate augmentation data.”, pg. 2, sixth paragraph; pg. 4, third paragraph)
However, the combination still does not teach transforming the pre-trained training data into a probability output according to a probability distribution function nor the initial discriminant parameter of the initial discriminator according to a minimized cross entropy to obtain a pre-trained discriminant parameter.
Wu teaches transforming the pre-trained sample data set into a probability output according to a probability distribution function to obtain a pre-trained network parameter (“In our framework, we need to minimize the distances between the probe and gallery representations through alternating the minimization between two functions. Thereby, the probe and gallery mappings should be optimized according to a constrained adversarial objective [53]… Intuitively, the loss function in Eq.(3) is a view confusion objective, under which the mapping can be trained using a cross-entropy loss function against a uniform distribution. This loss is to ensure the adversarial discriminator will view the two domains identically.”, pg. 5, left column, under “Adversarial View Adaptation with Cross-Entropy Loss”) (Cross-Entropy Loss inherently transforms data into a probability output according to a probability distribution function.) and updating the initial discriminant parameter of the initial discriminator according to a minimized cross entropy to obtain a pre-trained discriminant parameter. (“we introduce adversarial learning [25] into view discriminator which is optimized through cross-entropy based view confusion objective.”, pg. 2, left column, bottom paragraph; See also Algorithm 1 on pg. 5 for the updating step.) (It would be easy to configure the method of Wu which obtains a parameter via cross-entropy, with Yamaguchi’s method which obtains it via back-propagation of errors.)
Therefore, it would have been considered obvious to one of ordinary skill in the art, prior to the current application’s filing date, to combine the data augmentation of Yamaguchi, with the cross-entropy loss to obtain a discriminant parameter, as taught by Wu. One would be motivated to combine the two teachings, prior to the filing date of the current application, as updating the parameters with cross-entropy can optimize the model, leading to improved performance as disclosed in Wu. (“a common problem of existing domain adaptation approaches is that a principled alignment between the source and target is missing, and thus they are unable to penalize the correlated domain misalignment in practical terms. In contrast, our method explicitly minimizes the view discrepancy through the proposed view-adversarial objective. Our method is also distinct from existing methods based on adversarial losses [22]. For instance, SPGAN [22] is composed of GAN loss to update the target domain w.r.t the source, our method instead uses cross-entropy loss to optimize the view confusion objective.”, pg. 2, left column, above “B. Our Approach and Contribution”)
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Tyler E Iles whose telephone number is (571)272-5442. The examiner can normally be reached 9:00am - 5:00pm.
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, Kakali Chaki can be reached at (571) 272-3719. 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.
/TYLER EDWARD ILES/ Patent Examiner, Art Unit 2122
/KAKALI CHAKI/ Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights