DETAILED ACTION
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 responsive to pending claims 1-8 filed 3/8/2024.
Claim Objections
The following claim(s) are objected to for formality issues:
In claim 2, “the class score that is low”, “the class score that is high” should be “a” class score due to lack of antecedent basis.
In claim 4, “scare” should read “score”.
Claims 4-6 last line recites calculating a function of the total, which lacks antecedent bases. Positively reciting a calculation of “a total” in the previous line is needed, (e.g., “… to generate a total; and”)
Appropriate correction(s) are required.
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) 1-8 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The 35 U.S.C. 101 subject matter eligibility analysis first asks whether the claim is directed to one of the four statutory categories (Step 1). It next asks whether the claim is directed to an abstract idea (Step 2A), via Prong 1, whether an abstract idea (e.g., mathematical concept, mental process, certain methods of organizing human activity) is recited, and Prong 2, whether it is integrated into a practical application. It finally asks whether the claim as a whole includes additional elements that amount to significantly more than the judicial exception (Step 2B). See MPEP 2106.
STEP 1: The claims falls within one of the four statutory categories:
All claims are directed to hardware devices, methods, and non-transitory computer-readable media and hence fall within one of the four statutory categories.
STEP 2A PRONG 1: The claims recite a judicial exception: The claims are directed to calculating a regularizing loss function based on the confidence class scores for an inference model. As such, it is directed to a mathematical process. In particular (additional elements are underlined and will be analyzed subsequently):
For claim 1: A learning device comprising:
a memory storing instructions; and
one or more processors configured to execute the instructions to:
perform an inference with respect to training data using an inference model, and to output a class score (Performing an inference, such as via forward propagation, is a mathematical process);
calculate each weight using a weight function which rapidly increases faster than a linear function for the class score being over-estimated or under-estimated, based on the class score being output (calculating weights using a weight function having a certain non-linearity properties is a mathematical concept);
calculate a total of weights over a mini-batch included in a predetermined number of training data (calculating sums is a mathematical concept);
calculate a regularization term by applying a rescale function, which is a monotonically increasing function gradually increasing more than a linear function, to the regularization term (applying a regularization function is a mental process); and
optimize the inference model using a total loss including the regularization term (Optimizing an inference model, such as via backpropagation, via the calculated loss is a mathematical concept).
For claim 2: The learning device according to claim 1, wherein the processor increases a value of the regularization term for the class score that is high, and decreases the value of the regularization term for the class score that is low (This describes properties of the mathematical concept and hence is a mathematical concept).
For claim 3: The learning device according to claim 1, wherein processor is further configured to calculate a loss based on the class score and a correct answer class corresponding to the training data (calculating a loss based on comparison, e.g., difference with, a label is a mathematical concept),
wherein the class score is a total of the loss and the regularization term (taking a sum is a mathematical concept).
For claim 4: The learning device according to claim 1, wherein
the class score includes a confidence score for each class with respect to one training data (calculating confidence scores via an inference model is a mathematical concept),
the weight function is a function which adds up a scare of the confidence score of each class over all classes (taking a sum is a mathematical concept), and
the rescale function is a function which calculates a square root of the total (square roots are a mathematical concept).
For claim 5: The learning device according to claim 1, wherein
the class score includes a confidence score of one training data (computing a confidence score is a mathematical concept),
the weight function is a function which adds up a natural logarithm of a square of the confidence score for each class over all classes (taking a sum of logarithms is a mathematical concept), and
the rescale function is a function which calculates a logarithm of the total (taking logarithms is a mathematical concept).
For claim 6: The learning device according to claim 1, wherein
the class score includes a confidence score for each class with respect to one training data (computing a confidence score is a mathematical concept),
the weight function is a function which adds up a natural logarithm of the confidence score for each class (taking a sum of logarithms is a mathematical concept), and
the rescale function is a function which calculates a logarithm of the total (taking a sum of logarithms is a mathematical concept).
Claim 7-8 recite methods, computer-media corresponding to claim 1 and are hence similarly analyzed.
STEP 2A PRONG 2: The claims do not integrate the exception into a practical application:
For claims 1-3, 8 the additional elements comprise implementation memory and processor. However, this is mere instructions to implement the mathematical concept on a general purpose computer and hence do not comprise an integration into a practical application.
STEP 2B: The claim as a whole do not include additional elements that amount to significantly more than the abstract idea:
For claim 1-3, 8, the additional elements comprise implementation memory and processor. However, implementation of mathematical concepts on a general purpose computer are well-understood, routine, and conventional.
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.
Claim(s) 5-6 are rejected under 35 U.S.C. 112(b) as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, or for pre-AIA the applicant regards as the invention.
Claims 5-6 recites adding up the natural logarithm or natural logarithm of the square of the confidence scores and a rescale function that calculates a logarithm of the total. However, fig.3 discloses only discloses the sum of the exponents or inverse logarithms of the scores or square of scores. Hence, a reader cannot be certain as to whether logarithm includes inverse logarithm as shown, or whether a logarithm of the scores is intended, rendering the claim indefinite.
Claims will be interpreted as reciting inverse logarithm or exponent of the scores for purposes of examination.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1-4, 6-8 are rejected under 35 U.S.C. 103 as being unpatentable over Kornblith ("What’s in a loss function for image classification", published 2020) in view of Zhang (US 20210319033 A1).
For claim 1, Kornblith discloses: a learning device comprising:
perform an inference with respect to training data using an inference model, and to output a class score (§I: training deep learning neural nets on image classification tasks, hence, output of classification score);
calculate each weight using a weight function which rapidly increases faster than a linear function for the class score being over-estimated or under-estimated, based on the class score being output (§2 gives an overview of the various loss functions being used, the loss function performing regularization over the weights of the logit layer (§2 ¶1), the logit layers being an output layer for outputting class scores, hence, the training backpropagation process including calculating weights via the various weight functions disclosed, including eq.1-3, 6 showing an sum-of-exponentials function or sum-of-squares l2 norm (being faster than linear) for each class score being over or under estimated, based on the logit class score output);
calculate a total of weights over a mini-batch included in a predetermined number of training data (p.4 ¶1, with Appendix A.1 “ImageNet”, “Cifar”, disclosing batch size details);
calculate a regularization term by applying a rescale function, which is a monotonically increasing function gradually increasing more than a linear function, to the regularization term (eq.1-3, 6: the regularization term is produced via scaling weight function by log or square root, see); and
optimize the inference model using a total loss including the regularization term (eq.1: inference model is optimized via backpropagating loss function to tune weights).
Kornblith does not disclose: a memory storing instructions; and
one or more processors configured to execute the instructions.
Zhang discloses: a memory storing instructions (fig.8:830, 0110); and
one or more processors configured to execute the instructions (fig.8-810, 0110).
It would have been obvious before the effective filing date to a person of ordinary skill in the art to modify the device of Kornblith by incorporating the hardware configuration of Zhang. Both concern the art of machine learning and entropy regularization, and the incorporation would have, according to Zhang, allow the use of exiting computing technology to perform various machine learning methodologies (0110).
For claim 2, Kornblith modified by Zhang discloses the device of claim 1, as described above. Kornblith further discloses: wherein the processor increases a value of the regularization term for the class score that is high, and decreases the value of the regularization term for the class score that is low (§2, eq.2-3: label smoothing encourages convergence towards uniform, hence, reducing class score that is high and decreasing low class scores).
For claim 3, Kornblith modified by Zhang discloses the device of claim 1, as described above. Kornblith further discloses: wherein processor is further configured to calculate a loss based on the class score and a correct answer class corresponding to the training data (§2 ¶1-3 shows loss as based on target class t and class score l),
wherein the class score is a total of the loss and the regularization term (eq.1-3 show combining of regularization term with additional target-based loss for total loss).
For claim 4, Kornblith modified by Zhang discloses the device of claim 1, as described above. Kornblith further discloses: wherein the class score includes a confidence score for each class with respect to one training data (§2 ¶1-2: logit layer class score constitutes a confidence score for each class, increasing with greater confidence, with respect to a training data),
the weight function is a function which adds up a scare of the confidence score of each class over all classes (eq.6, l-2 norm of logits l); and
the rescale function is a function which calculates a square root of the total (eq.6: shows l-2 norm of the logits as a rescaling function in order to control minimum entropy, hence, functioning as a regularization term).
For claim 6, Kornblith modified by Zhang discloses the device of claim 1, as described above. Kornblith further discloses: the class score includes a confidence score for each class with respect to one training data (§2 ¶1-2: logit layer class score constitutes a confidence score for each class, increasing with greater confidence, with respect to a training data);
the weight function is a function which adds up a natural logarithm of the confidence score for each class (eq.1), and
the rescale function is a function which calculates a logarithm of the total (eq.1).
Claim 7 recites a method analogous to the device above and hence is rejected for the same reasons.
Claim 8 modified by Zhang discloses a computer-readable medium analogous to the device of claim 1 and is hence rejected for similar reasons. Furthermore, Zhang discloses: a non-transitory computer readable recording medium storing a program, the program causing a computer to perform a process (fig.8:830, 0112).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Yeo (US 20200097813 A1) discloses the use of entropy regularization for neural networks.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to LIANG LI whose telephone number is (303)297-4263. The examiner can normally be reached Mon-Fri 9-12p, 3-11p MT (11-2p, 5-1a ET).
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. The examiner is available for interviews Mon-Fri 6-11a, 2-7p MT (8-1p, 4-9p ET).
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor Jennifer Welch can be reached on (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 an application may be obtained from Patent Center and the Private Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from Patent Center or Private PAIR. Status information for unpublished applications is available through Patent Center or Private PAIR to authorized users only. Should you have questions about access to Patent Center or the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/LIANG LI/
Primary examiner AU 2143
Prosecution Insights