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 Arguments
Applicant’s argument filed 07/21/2026 have been fully considered but they are not persuasive. The amendments have corrected the informalities of the objection of claim 10.
Applicant’s Argument: On page 7-9 of Applicant’s response to rejections under 35 U.S.C. § 112(b), applicant states that (a) the claimed “first/second intensity” corresponds to “weak/strong augmentation” consistently disclosed in the Specification, (b) “weak augmentation” and “strong augmentation” are standard terms of art, (c) “larger” is definite, and (d) claim construction must be performed in light of the Specification.
Examiner’s Response: Applicant’s argument is not persuasive. The claims do not explicitly corresponds the claimed “first intensity” to “weak augmentation” and the claimed “second intensity” with “strong augmentation”. The Specification also does not disclose such correlation between the terms. Under the broadest reasonable interpretation, the claimed “first intensity” can be interpreted to be weak augmentation or strong augmentation in light of the Specification. In addition, the Specification does not provide a standard for ascertaining the requisite degree as to what constitutes as weak and strong augmentation. Applicant argues that weak augmentation is a standard term and is defined by Sohn (FixMatch) as flipping and translating images. The Sohn reference does not provide a standard definition for the terms “weak augmentation” and “strong augmentation” that is recognized as an industry standard. The Applicant fails to provide any evidence to support how these term definitions from the Sohn reference are the industry standard. In Sohn, there is disclosure as what constitutes to be weak and strong augmentation in the context of the reference. However, in the Specification of the claimed invention, there is no such disclosure as to what constitutes as weak augmentation and strong augmentation. The Specification of the claimed invention also does not disclose any relations between the weak and strong augmentation with the claimed first and second intensity.
Under the broadest reasonable interpretation, the claimed “first data” and “second data” may be generated using the same data augmentation. Thus, the claimed “first intensity” may be interpreted as flipping the first data one time and the claimed “second intensity” may be interpreted as flipping the data two times. Therefore, it directly contradicts with the Applicant’s argument that that claimed “second intensity” corresponds with the “strong augmentation” because the standard definition of strong augmentation consists of CTAugment as proposed by the Applicant. It is not clear what constitutes as weak and strong augmentation.
Claim language may not be "ambiguous, vague, incoherent, opaque, or otherwise unclear in describing and defining the claimed invention." In re Packard, 751 F.3d 1307, 1311, 110 USPQ2d 1785, 1787 (Fed. Cir. 2014). Applicants need not confine themselves to the terminology used in the prior art, but are required to make clear and precise the terms that are used to define the invention whereby the metes and bounds of the claimed invention can be ascertained. During patent examination, the pending claims must be given the broadest reasonable interpretation consistent with the specification. In re Morris, 127 F.3d 1048, 1054, 44 USPQ2d 1023, 1027 (Fed. Cir. 1997); In re Prater, 415 F.2d 1393, 162 USPQ 541 (CCPA 1969).
The claims recites “the second intensity being larger than the first intensity” and the claim limitation is indefinite because the specification does not provide a standard for ascertaining the requisite degree of much larger does the second intensity needs to be in relation with the first intensity. Applicant arguments (a) and (b) discloses that weak and strong augmentation are standard terms of art as confirmed by the cited reference Sohn. Thus, weak augmentation corresponds to flipping and translating images and strong augmentation consists of CTAugment, which may include transformation that are different from the weak augmentation. The Specification does not disclose the claimed invention with clarity and precision to allow a POSITA to determine what constitutes as a larger intensity when the data augmentations are different.
Applicant’s Argument: On page 9-10 of Applicant’s response to rejections under 35 U.S.C. § 101, applicant states that the claims integrate the alleged judicial exception into a practical application. The claims recite a specific technical solution for machine learning training of selecting unlabeled training data based on a pseudo-distance between a model output distribution and a gain matrix-based probability distribution. The Specification discloses a specific improvement to the training performance and generalization performance of the machine learning model.
Examiner’s Response: Applicant’s argument is not persuasive. An important consideration in determining whether a claim improves technology is the extent to which the claim covers a particular solution to a problem or a particular way to achieve a desired outcome, as opposed to merely claiming the idea of a solution or outcome (see MPEP 2106.05(a)). The amended claims do not provide sufficient details to describe any technological improvement. If the specifications explicitly set forth an improvement but in a conclusory manner (see MPEP 2106.04(d)(1): a bare assertion of an improvement without the detail necessary to be apparent to a person of ordinary skill in the art), the examiner should not determine the claim improves technology.
During examination, the examiner should analyze the "improvements" consideration by evaluating the specification and the claims to ensure that a technical explanation of the asserted improvement is present in the specification, and that the claim reflects the asserted improvement (see MPEP §2106.05(a)). The MPEP (§2106.05(a)(II)) also warns, “it is important to keep in mind that an improvement in the abstract idea itself (e.g. a recited fundamental economic concept) is not an improvement in technology.” Here, the alleged improvement in the form of “selecting unlabeled training data based on a pseudo-distance between a model output distribution and a gain matrix-based probability distribution” is an improvement to the abstract idea of a mental process that can be performed in the human mind.
Applicant’s Argument: On page 10-13 of Applicant’s response to rejections under 35 U.S.C. § 103, applicant states that the Narasimhan reference does not describe the claimed unlabeled data and predicted label. Huang discloses that the selected instances are used for training only after its true label has been obtained by querying and it is not the same as the recited claims. Applicant argues that there is a lack of motivation to combine Huang’s query-based instance-selection technique with Narasimhan’s supervised framework.
Examiner’s Response: Applicant’s argument is not persuasive. In the previous Office Action (dated 04/21/2026), Examiner states that the Narasimhan reference discloses selecting, based on a gain matrix, first training data to be used for training a machine learning model from a plurality of training data. Narasimhan does not explicitly disclose the plurality of data is unlabeled. Narasimhan discloses assigning importance weights to training samples to allow the model to focus on information samples and it is directed to optimizing data training to improve model training. Narasimhan (pg. 8-9, Section 6, par. 4-7) discloses soft predictions from a teacher model to teach “a predicted label that is predicted from the first training data”. Under the broadest reasonable interpretation, the claimed “predicted label that is predicted from the first training data” does not specify what entity performs the prediction and can encompass a prediction by a model or a human expert.
Huang (pg. 1-2, Section II, par. 1) discloses selection of data from an unlabeled pool, queries its label, and trains a model based on the data instance and its queried label. Under the broadest reasonable interpretation, Huang teaches selecting training data from a plurality of unlabeled training data to be used for training a machine learning model. The recited claims do not further limit the scope of the definition of the training of the machine learning model. The recited claims do not explicitly define the training of the machine learning model as trained with unlabeled data without any queried labels and the claim does not disclose the details of how the predicted label is obtained from the first training data.
Huang describes a query-based instance selection technique to select training instances for a machine learning model. Narasimhan (abstract) discloses a cost-sensitive loss optimization technique for assigning importance weights to training samples to allow the model to focus on informative samples. The cost-sensitive loss optimization technique from Narasimhan can be implemented into Huang to improve the training data selection process because the optimization technique can improve the performance of the model on under-represented samples and it is directed to selecting more informative samples.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 3-4, 7-8, and 11-12 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
The term “a first intensity” in claims 3-4, 7-8, and 11-12 is a relative term which renders the claims indefinite. The term “a first intensity” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. The Specification does not disclose the term “intensity”. It is not clear what constitutes as a first intensity for a data augmentation. Examiner interprets data augmentation with a first intensity as applying a number of variations to data.
The term “the second intensity being larger than the first intensity” in claims 4, 8, and 12 is a relative term which renders the claims indefinite. The term “a second intensity” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. It is not clear what constitutes as a second intensity for a data augmentation. Additionally, the term “larger” is a relative term which renders the claims indefinite. Examiner interprets data augmentation with a second intensity as applying a number of variations to data. The second intensity refers to applying a higher number of variations to the data compared to the first intensity.
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 1-12 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Subject Matter Eligibility Analysis Step 1:
Claim 1 recites “An information processing apparatus comprising” and is thus a machine, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
“decide a gain matrix based on an input metric” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. judgement)
“perform selection of first training data from a plurality of unlabeled training data, to be used for training a machine learning model, based on the gain matrix” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. judgement)
“perform training of the machine learning model based on the first training data, a predicted label that is predicted from the first training data, and a loss function including the gain matrix” (a mathematical calculation, see par. 72-78 in Specification)
Claim 1 therefore recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
“one or more memories” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“one or more processors coupled to the one or more memories, the one or more processors being configured to” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
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. Therefore, Claim 1 is directed to the abstract idea.
Subject Matter Eligibility Analysis Step 2B:
“one or more memories” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“one or more processors coupled to the one or more memories, the one or more processors being configured to” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
The additional elements as disclosed above alone or in combination do not recite significantly more than the abstract idea itself 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. Therefore, Claim 1 is subject-matter ineligible.
Regarding Claim 5:
The claim recites a method that performs the process as described in claim 1. Therefore, claim 5 is rejected for the same reasons as disclosed in claim 1.
Regarding Claim 9:
The claim recites an article of manufacture that performs the method as described in claim 1. Therefore, claim 9 is rejected for the same reasons as disclosed for claim 1. The limitations for additional elements of claim 9 are analyzed below.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Please see Step 2A Prong 1 analysis of claim 1
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“A non-transitory computer-readable recording medium having stored therein machine learning program that causes a computer to execute a process comprising” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 2, 6, and 10:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“wherein the selection includes selecting the first training data in a case where a pseudo-distance between a probability distribution output from the machine learning model in response to inputting data obtained by augmenting the first training data into the machine learning model and a probability distribution that is based on the gain matrix is equal to or less than a threshold” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B: None
Regarding Claim 3:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“wherein the processors are further configured to generate the predicted label based on an output result from the machine learning model in response to inputting first data into the machine learning model, the first data being generated by executing data augmentation with a first intensity on the first training data” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 7 and 11:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“executing data augmentation with a first intensity on the first training data” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“generating the predicted label based on an output result from the machine learning model in response to inputting first data into the machine learning model, the first data being generated by executing data augmentation with the first intensity on the first training data” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 4, 8, and 12:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“wherein the training is executed by inputting a first value and a second value to the loss function, the first value being an output result from the machine learning model in response to inputting second data generated by executing data augmentation with a second intensity on the first training data into the machine learning model, the second intensity being larger than the first intensity, the second value being obtained by vectorizing an output result from the machine learning model in response to inputting the first data into the machine learning model” (a mathematical calculation; see Specification par. 51)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B: None
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, 3, 5, 7, 9, and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Narasimhan, “Training Over-parameterized Models with Non-decomposable Objectives” in view of Huang, “A Novel Uncertainty Sampling Algorithm for Cost-sensitive Multiclass Active Learning”.
Regarding claim 1, Narasimhan teaches:
“An information processing apparatus comprising: one or more memories; and one or more processors coupled to the one or more memories, the one or more processors being configured to” ([abstract, pg. 21, Section E, par. 1-5], Experiments are conducted on several training datasets. It is implied that the experiments are conducted on a computer that consists of processors and memories.)
“decide a gain matrix based on an input metric” ([pg. 2-3, Section 2, par. 7-9; pg. 17, Section A, par. 1-2; pg. 4, Table 1], Table 1 shows the different forms of the gain matrix for various constraints. Algorithm 1 shows the problem of maximizing the worst-case recall and Algorithm 2 shows the problem of maximizing the average recall subject to coverage constraints. Each cost-sensitive loss is associated with a different gain matrix.)
“perform selection of first training [parameters] ” ([pg. 4, Section 3, par. 2-5; pg. 5-6, Section 4, par. 7-8], A re-weighted loss is used to train models and the weighted loss is calibrated for the gain matrix. In one embodiment, a hybrid approach is disclosed that combines logit adjustment with an outer weighting. The outer diagonal matrix can be selected to reflect the relative importance of the classes of the training samples. Narasimhan does not explicitly teach selection of training samples based on a gain matrix. Narasimhan discloses assigning importance weights to training samples to allow the model to focus on informative samples.)
“perform training of the machine learning model based on the first training data, a predicted label that is predicted from the first training data, and a loss function including the gain matrix” ([pg. 8-9. Section 6, par. 4-7], Knowledge distillation can be combined with the proposed cost-sensitive losses based on logit adjustment. Soft predictions from a teacher model is used as labels to train a student model. The student model is trained on the teacher’s predicted labels of the same dataset and the loss is based on the gain matrix as disclosed in Equation 11.)
Narasimhan does not explicitly disclose an implementation of “perform selection of first training data from a plurality of unlabeled training data, to be used for training a machine learning model, based on the gain matrix”. However, Huang discloses in the same field of endeavor:
“perform selection of first training data from a plurality of unlabeled training data, to be used for training a machine learning model, based on the ” ([pg. 1-2, Section II, par. 1; pg. 2, Section A, par. 1; pg. 4, Section E, par. 2; pg. 4, Section A, par. 1], Active learning with cost embedding is disclosed and the framework is based on pool-based multiclass active learning, which selects one instance from the unlabeled pool to query its label in every iteration. The algorithm selects the instance with respect to the cost.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “perform selection of first training data from a plurality of unlabeled training data, to be used for training a machine learning model, based on the gain matrix” from Huang into the teaching of Narasimhan. Doing so can improve cost-sensitive learning algorithms by selecting useful instances by taking the cost information into account (Huang, abstract).
Regarding claim 5:
Claim 5 recites a method that performs the same process as described in Claim 1. Therefore claim 5 is rejected under the same reasons mentioned for claim 1.
Regarding claim 9:
Claim 9 recites an article of manufacture that performs the same process as described in Claim 1. Therefore claim 9 is rejected under the same reasons mentioned for claim 1. The additional elements of claim 9 are addressed below by Narasimhan:
“A non-transitory computer-readable recording medium having stored therein machine learning program that causes a computer to execute a process comprising” ([abstract, pg. 21, Section E, par. 1-5], Experiments are conducted on several training datasets. It is implied that the experiments are conducted on a computer that consists of processors and memories.)
Regarding claim 3, Narasimhan in view of Huang teaches:
“wherein the processors are further configured to generate the predicted label based on an output result from the machine learning model in response to inputting first data into the machine learning model, the first data being generated by executing data augmentation with a first intensity on the first training data” ([Huang, pg. 2-3, Section A, par. 1], For a K number of classes, the framework determines K corresponding hidden points in a M-dimensional hidden space. For any testing instance sample, the predicted hidden point is obtained. Embedding input data into a hidden space is a type of data augmentation. The final prediction is the class corresponding to the nearest hidden point.)
Regarding claims 7 and 11, Narasimhan in view of Huang teaches:
“executing data augmentation with a first intensity on the first training data” ([Huang, pg. 2-3, Section A, par. 1], For any testing instance sample, the predicted hidden point is obtained by applying a function g to the sample. Embedding input data into a hidden space is a type of data augmentation. Performing the embedding into the hidden space and mapping data into a lower-dimensional representation represents an intensity of the data augmentation.)
“generating the predicted label based on an output result from the machine learning model in response to inputting first data into the machine learning model, the first data being generated by executing data augmentation with the first intensity on the first training data” ([Huang, pg. 2-3, Section A, par. 1], For a K number of classes, the framework determines K corresponding hidden points in a M-dimensional hidden space. For any testing instance sample, the predicted hidden point is obtained. Embedding input data into a hidden space is a type of data augmentation. The final prediction is the class corresponding to the nearest hidden point.)
Claims 2, 6, and 10 are rejected under 35 U.S.C. 103 as being unpatentable over Narasimhan, “Training Over-parameterized Models with Non-decomposable Objectives” in view of Huang, “A Novel Uncertainty Sampling Algorithm for Cost-sensitive Multiclass Active Learning” and Karim, “UNICON: Combating Label Noise Through Uniform Selection and Contrastive Learning”.
Regarding claims 2, 6, and 10, Narasimhan teaches:
“wherein the selection includes selecting the first training ” ([pg.2-3, Section 2, par. 7-9; pg. 4, Section 3, par. 2-5; pg. 5-6, Section 4, par. 7-8], A gain matrix are the rewards associated with predicting class j when the true class is i. Thus, it represents the probability of the predicted class label from the true label. The outer diagonal matrix can be selected to reflect the relative importance of the classes of the training samples. Narasimhan discloses assigning importance weights to training samples to allow the model to focus on informative samples.)
Narasimhan does not explicitly disclose an implementation of “wherein the selection includes selecting the first training data in a case where a pseudo-distance between a probability distribution output from the machine learning model in response to inputting data obtained by augmenting the first training data into the machine learning model and a probability distribution that is based on the gain matrix is equal to or less than a threshold”. However, Huang discloses in the same field of endeavor:
“” ([pg. 2-3, Section A, par. 1; pg. 3, Section B, par. 1; ], For a K number of classes, the framework determines K corresponding hidden points in a M-dimensional hidden space. For any testing instance sample, the predicted hidden point is obtained. Embedding input data into a hidden space is a type of data augmentation. The final prediction is the class corresponding to the nearest hidden point and it is determined with a nearest neighbor function to calculate the distance between the predicted hidden point and the class with embedded cost information.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “wherein the selection includes selecting the first training data in a case where a pseudo-distance between a probability distribution output from the machine learning model in response to inputting data obtained by augmenting the first training data into the machine learning model and a probability distribution that is based on the gain matrix is equal to or less than a threshold” from Huang into the teaching of Narasimhan. Doing so can improve cost-sensitive learning algorithms by selecting useful instances by taking the cost information into account (Huang, abstract).
Narasimhan in view of Huang does not explicitly disclose an implementation of “wherein the selection includes selecting the first training data in a case where a pseudo-distance between a probability distribution output from the machine learning model in response to inputting data obtained by augmenting the first training data into the machine learning model and a probability distribution that is based on the gain matrix is equal to or less than a threshold”. However, Karim discloses in the same field of endeavor:
“wherein the selection includes selecting the first training data in a case where a pseudo-distance between a probability distribution output from the machine learning model in response to inputting data obtained by ” ([pg. 3-4, Section 4.1, par. 1-5; pg. 4, Algorithm 1], The divergence is computed between the ground-truth labels and the predicted probabilities of the training dataset using KL divergence function. A cutoff divergence value is determined and a selection criterion is determined to filter R portion of samples from each class based on the cutoff divergence value.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “wherein the selection includes selecting the first training data in a case where a pseudo-distance between a probability distribution output from the machine learning model in response to inputting data obtained by augmenting the first training data into the machine learning model and a probability distribution that is based on the gain matrix is equal to or less than a threshold” from Karim into the teaching of Narasimhan in view of Huang. Doing so can improve class imbalance by implementing a sample selection method which is robust to high label noise (Karim, abstract).
Claims 4, 8, and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Narasimhan, “Training Over-parameterized Models with Non-decomposable Objectives” in view of Huang, “A Novel Uncertainty Sampling Algorithm for Cost-sensitive Multiclass Active Learning” and Sohn, “FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence”. The Sohn reference is included in the IDS filed 05/19/2023.
Regarding claims 4, 8, and 12, Narasimhan in view of Huang does not explicitly disclose an implementation of “wherein the training is executed by inputting a first value and a second value to the loss function, the first value being an output result from the machine learning model in response to inputting second data generated by executing data augmentation with a second intensity on the first training data into the machine learning model, the second intensity being larger than the first intensity, the second value being obtained by vectorizing an output result from the machine learning model in response to inputting the first data into the machine learning model”. However, Sohn discloses in the same field of endeavor:
“wherein the training is executed by inputting a first value and a second value to the loss function, the first value being an output result from the machine learning model in response to inputting second data generated by executing data augmentation with a second intensity on the first training data into the machine learning model, the second intensity being larger than the first intensity, the second value being obtained by vectorizing an output result from the machine learning model in response to inputting the first data into the machine learning model” ([pg. 3, Section 2.1, par. 3; pg. 3-4, Section 2.2, par. 1-3; pg. 4, Section 2.3, par. 1-2; pg. 2, Figure 1], The loss function consists of the artificial label computed based on a weakly-augment image and the loss is enforced against the model’s output for a strongly-augmented image. Equation 4 shows the loss function and it consists of the output of a weakly-augmented version and a strongly-augmented version of a given unlabeled image. Weak augmentation consists of a flipping and translating images. Strong augmentation consists of CTAugment, which randomly selects transformations for each sample. In the experiment, the function arg max applied to a probability distribution produces a valid “one-hot” probability distribution (vectorizing an output result).)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “wherein the training is executed by inputting a first value and a second value to the loss function, the first value being an output result from the machine learning model in response to inputting second data generated by executing data augmentation with a second intensity on the first training data into the machine learning model, the second intensity being larger than the first intensity, the second value being obtained by vectorizing an output result from the machine learning model in response to inputting the first data into the machine learning model” from Sohn into the teaching of Narasimhan in view of Huang. Doing so can improve semi-supervised learning by leveraging unlabeled data to improve a model’s performance (Sohn, abstract).
Conclusion
THIS ACTION IS MADE FINAL. 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 GARY MAC whose telephone number is (703)756-1517. The examiner can normally be reached Monday - Friday 8:00 AM - 5:00 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Kawsar can be reached at (571) 270-3169. 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.
/GARY MAC/Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
Prosecution Insights