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 amendment filed on 09/01/2026 has been entered. Claims 1, 6, 11, 13, & 16 are amended. Claims 1-16 are pending in the application.
Claim Objections
Claims 1, 11, & 16 are objected to because of the following informality:
Claims 1, 11, & 16 recite ‘a HET classifier’; however, it should recite - - a heteroscedastic (HET) classifier - -.
Appropriate correction is required.
Claim Rejections - 35 USC § 103
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 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 1-5, 9, 10-15, 16-17, & 19 are rejected under 35 U.S.C. 103 as being unpatentable over Chang et al. ("Data Uncertainty Learning in Face Recognition", Megvii Inc., arXiv) (Year: 2020), hereafter Chang, in view of Collier et al., ("Correlated Input-Dependent Label Noise in Large-Scale Image Classification", arXiv:2105.10305v1, May 19, 2021, 15 pages), hereafter Collier 21.
Collier 21 was cited in IDS filed 10/07/2024.
Regarding independent claim 1, Chang teaches a system comprising:
generating a first vector based on an embedding model operating on an input object, wherein the first vector is embedded within a first vector space that has a first number of dimensions ([Sec. 3.2 & 4.1] discusses generating a first vector based on an embedding model; the first vector space has a first number of dimensions);
generating a noise vector, wherein generating the noise vector comprises: sampling a random vector from a normal distribution, wherein the random vector is based on a covariance ([Sec. 3.2] discusses generating a noise vector by sampling a random vector from a normal distribution; the random vector is based on a covariance);
generating a second vector based on combining the first vector with the noise vector, wherein the second vector and the noise vector are embedded in the first vector space ([Sec. 3.2 & Fig. 3] discusses generating a second vector based on combining the first vector and noise vector);
generating a third vector based on the second vector and a logit function, wherein a linear classification projection layer of a HET classifier model projects and embeds the second vector into a second vector space that has a second number of dimensions that is greater than the first number of dimensions and is equivalent to a number of classes of the HET classifier model ([Sec. 3.2, 4.1-4.2, & Fig. 3] discusses generating a third vector based on a the second vector and a logit function; a linear projection layer embeds the second vector into a space that has a greater number of dimensions than the first number and is equivalent to the number of classes of the classifier);
and training the HET classifier model based on the third vector ([Sec. 3.2, 4.1-4.2, & Fig. 3] discusses training the model end-to-end.
Chang does not explicitly teach wherein the random vector is based on a covariance associated with a set of components of the first vector.
However, in a similar field of endeavor, Collier 21 teaches a system that employs a HET classifier model wherein a generated random vector is based on a non-diagonal, input-dependent covariance ([Sec. 2, 3.1, & 4.1] discusses the use of a full covariance that is input-dependent over the embedding components and thus, associated with a set of components of the first vector).
Because Chang teaches generating a first vector based on an embedding model, generating a noise vector by sampling a random vector from a normal distribution wherein the random vector is based on a covariance, generating a second vector based on combining the first vector with the noise vector, generating a third vector based on the second vector and a logit function, and a projection layer projects and embeds the second vector into a second vector space that has more dimensions than the first number of dimensions and equal number to the classes of the HET classifier model; and Collier 21 teaches a covariance associated with a set of components of the first vector, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate a covariance associated with a set of components of the first vector as taught by Collier 21 into Chang’s computer-implemented system, with a reasonable expectation of success, to teach generating a first vector based on an embedding model operating on an input object, wherein the first vector is embedded within a first vector space that has a first number of dimensions; generating a noise vector, wherein generating the noise vector comprises: sampling a random vector from a normal distribution, wherein the random vector is based on a covariance associated with a set of components of the first vector; generating a second vector based on combining the first vector with the noise vector, wherein the second vector and the noise vector are embedded in the first vector space; generating a third vector based on the second vector and a logit function, wherein a linear classification projection layer of a HET classifier model projects and embeds the second vector into a second vector space that has a second number of dimensions that is greater than the first number of dimensions and is equivalent to a number of classes of the HET classifier model; and training the HET classifier model based on the third vector. This combination would have been motivated by the desire to reduce parameter and computational requirements by implementing a covariance matrix associated with a set of components of the first vector (Collier 21 [Sec. 2]).
Regarding dependent claim 2, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 1 including wherein the covariance associated with the set of components of the first vector is based on a multivariate Gaussian distribution that has a separate random variable associated with each component of the set of components of the first vector and the covariance is a covariance between the separate random variables (Collier 21 [Sec. 2, Pg. 2] discusses computing a covariance matrix based on a multivariate Gaussian distribution; and the covariance is input dependent and thus, has a separate random variable associated with each component).
Regarding dependent claim 3, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 1 including wherein each component of the third vector corresponds to a separate class of the number of classes of the HET classifier model and the operations further comprise: generating a fourth vector based on an activation function, wherein the fourth vector is embedded in the second vector space and the activation function transforms each component of the third vector into a probability value such that each component of the fourth vector corresponds to a separate class of the number of classes and is equivalent to the probability value corresponding to the class (Collier 21 [Sec. 2, Pg. 2] discusses that K represents the number of classes associated with x; Collier 21 [Alg. 1] discusses generating a fourth vector where the activation function embedded transforms each component into a probability value pc and each component of the vector corresponds to their respective class).
Regarding dependent claim 4, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 3 including wherein the activation function is a softmax function (Collier 21 [Alg. 1] discusses the use of an activation function using a softmax function).
Regarding dependent claim 5, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 3 including wherein the activation function is a sigmoid function (Collier 21 [Sec. 2] discusses that the method operates when more than one class is in an image and the activation function will apply a sigmoid smoothing function).
Regarding dependent claim 9, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 1 including wherein the covariance associated with the set of components of the first vector is encoded in a covariance matrix that has a number of rows equivalent to the first number of dimensions and a number of columns that is equivalent to the first number of dimensions (Chang [Sec. 3] discusses the use of a diagonal covariance matrix; a covariance matrix is square and will have an equivalent amount of rows and columns to dimensions; Collier 21 [Sec. 2-4 & Alg. 1] discusses a covariance associated with the set of components of the first vector is encoded in a covariance matrix that has a number of rows equivalent to the first number of dimensions and a number of columns that is equivalent to the first number of dimensions).
Regarding dependent claim 10, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 9 including wherein the covariance matrix is generated via a Monte Carlo process (Collier 21 [Sec. 2] discusses use of Monte Carlo estimation in the covariance computation).
Regarding claims 11-15, claims 11-15 are method claims that are substantially the same as the system of claims 1, 3-5, & 9. Therefore, claims 11-15 are rejected for the same reasons as claims 1, 3-5, & 9.
Regarding claims 16-17, claims 16-17 are non-transitory computer-readable storage medium claims that are substantially the same as the system of claims 1 & 3. Therefore, claims 16-18 are rejected for the same reasons as claims 1 & 3.
Regarding dependent claim 19, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 19 including wherein the operations further comprise: deploying the HET classifier model for an image classification task (Collier 21 [Sec. 4] discusses deploying the HET classifier model on four large-scale image classification datasets).
Claims 6 & 18 are rejected under 35 U.S.C. 103 as being unpatentable over Collier 21, in view of Chang, as applied in claims 3 and 17, and further in view of Collier et al., ("A Simple Probabilistic Method for Deep Classification under Input-Dependent Label Noise", arXiv:2003.06778v3, November 12, 2020, 19 pages), hereafter Collier 20.
Collier 20 was cited in the IDS submitted 10/07/2024.
Regarding dependent claim 6, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 3 including a temperature parameter of the HET classifier model (Collier 21 [Sec. 2] discusses that the system uses a temperature parameterized softmax).
The combination of Chang and Collier 21 does not explicitly teach wherein the operations further comprise: transforming each component of the third vector based on a temperature parameter of the HET classifier model; and generating the fourth vector transforming each component of the third vector based on a temperature parameter of the HET classifier model.
However, in the same field of endeavor, Collier 20 teaches transforming components of an approximation based on a temperature parameter ([Sec. 3.3, Eq. 5] discusses each individual component uc being divided by the temperature parameter and thus, is transformed based on the temperature parameter; [Sec. 3.3] also discusses that as the temperature parameter gets close to zero, the bias goes down while the MC estimate of the gradients of the objective increase and thus, the fourth vector is generated by transforming the third vector directly based on the temperature parameter).
Because the combination of Chang and Collier 21 teaches the use of a temperature parameter and Collier 20 teaches transforming the third vector based on a temperature parameter and generating a fourth vector based on that transformation, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate the transformation of the third vector by a temperature parameter and further generating the fourth vector based on the transformation as taught by Collier 20 into the combination of Chang and Collier 21’s computer-implemented system, with a reasonable expectation of success, to teach wherein the operations further comprise: transforming each component of the third vector based on a temperature parameter of the HET classifier model; and generating the fourth vector transforming each component of the third vector based on a temperature parameter of the HET classifier model. This combination would have been motivated by the desire to improve accuracy, log-likelihood and calibration on both image classification benchmarks with controlled label noise as well as Imagenet-21k which has naturally occurring label noise (Collier 20 [Abstract]).
Regarding claim 18, claim 18 is a non-transitory computer-readable storage medium claim that is substantially the same as the system of claim 6. Therefore, claim 18 is rejected for the same reasons as claim 6.
Claims 7 & 8 are rejected under 35 U.S.C. 103 as being unpatentable over Collier 21, in view of Chang, as applied in claim 1, and further in view of Guo et al. ("On Calibration of Modern Neural Networks", Proceedings of the 34th International Conference on Machine Learning, arXiv) (Year: 2017), hereafter Guo.
Regarding dependent claim 7, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 1 including training the HET classifier model includes a value for a temperature parameter (Collier [Sec. 1] discusses the use of a temperature parameter to control the bias-variance tradeoff).
The combination of Chang and Collier 21 does not explicitly teach wherein training the classifier model includes learning a value for a temperature parameter.
However, Guo teaches temperature optimization during training ([Sec. 4.2 & Eq. 9] discusses T being optimized during the process of computing the softmax and thus, the temperature parameter’s value is learned during training).
Because the combination of Chang and Collier 21 teaches training the HET classifier model includes a value for a temperature parameter and Guo teaches training the classifier model includes learning a value for a temperature parameter, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate learning a value for the temperature parameter during training as taught by Guan into the combination of Chang and Collier 21’s computer-implemented system, with a reasonable expectation of success, to teach wherein training the HET classifier model includes learning a value for a temperature parameter. This combination would have been motivated by the desire that temperature scaling is by far the fastest method, as it amounts to a one-dimensional convex optimization problem. Using a conjugate gradient solver, the optimal temperature can be found in 10 iterations, or a fraction of a second on most modern hardware (Guo [Sec. 5]).
Regarding dependent claim 8, the combination of Chang, Collier 21, and Guo teaches the claimed invention as claimed in claim 7 including wherein the value for the temperature parameter is learned by a gradient descent algorithm (Guo [Sec. 4.2] discusses parameter T being optimized with a softmax problem and thus, is learned by a type of gradient descent algorithm).
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Collier 21, in view of Chang, as applied in claim 16, and further in view of Guan et al. ("Nonparametric Stochastic Contextual Bandits", Proceedings of the AAAI Conference on Artificial Intelligence, 32(1)) (Year: 2018), hereafter Guan.
Regarding dependent claim 20, the combination of Chang and Collier 21 teaches the claimed invention as claimed in claim 16 including wherein the operations further comprise: deploying the HET classifier model (Collier 21 [Sec. 4] discusses deploying the HET classifier model on four large-scale image classification datasets).
The combination of Chang and Collier 21 does not explicitly teach wherein the operations further comprise: deploying a classifier model for a multi-armed bandit task.
However, Guan teaches deploying an image classifier for a bandit task ([Sec. Experiments, Pg. 3124] discusses extending bandit tasks to image classification and the bandit is an infinite-armed; thus, deploys a classifier for a multi-armed bandit task).
Because the combination of Chang and Collier 21 teaches deploying the HET classifier and Guan teaches deploying a classifier model for a multi-armed bandit task, accordingly, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate deploying a classifier for a multi-armed bandit task as taught by Guan into the combination of Chang and Collier 21’s computer-implemented system, with a reasonable expectation of success, to teach wherein the operations further comprise: deploying the HET classifier model for a multi-armed bandit task. This combination would have been motivated by the desire to learn over the entire context space with strong guarantees, despite the difficulty that arises with allowing a wide variety of reward functions (Guan [Introduction]).
Response to Arguments
Applicant’s amendments and remarks filed 09/01/2026, on pages 7-11, traversing the 35 U.S.C. 101 rejections set forth in the Office Action dated 06/01/2026, are persuasive and hereby withdrawn.
Applicant’s arguments filed 09/01/2026, on pages 11-16, with respect to claims 1-20 under 35 U.S.C. 103 have been considered but are moot because of the new ground of rejection (see rejection above).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's
disclosure.
Ramachandran (US 11481636 B2) ([Detailed Description] Specifically, a sparsification vector is generated based on Gaussian distribution (or other probabilistic distribution) and is used to multiply with the higher dimension output to reduce the number of feature dimensions. The pruned output may be then used for the neural network to learn the sparsification vector. In this way, OOD detection accuracy can be improved).
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 RILEY S ACOSTA whose telephone number is (571)272-8714. The examiner can normally be reached Monday-Thursday 6am-4pm.
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, Jennifer N Welch can be reached at (571)272-7212. 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.
/RILEY S ACOSTA/Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143
