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 .
Status of Claims
Claims 1-24 are pending.
Claims 11-12, 14-17, 19 and 22 are withdrawn.
Claims 13, 18 and 20 are cancelled.
Claim Rejections - 35 USC § 101
Previous 101 rejections to claims have been withdrawn in view of amendments made by the applicant.
Response to Arguments
Applicant’s amendments filed 04/27/2026 have been considered and entered, however, in view of applicant’s amendments, the previous rejection(s) have been withdrawn and upon further consideration, a new ground(s) of rejection(s) are now made in view of Zhang.
Applicant argues, see remarks, page 11, that amendments have been made to claim 23, thereby previously made interpretation of claims under 112(f) shall be withdrawn.
In reply, examiner asserts that applicant has still failed to address the limitations of claim 23 pertaining to different modules, therefore, the interpretation of claim 23 under 112(f) has been yet again repeated below.
Applicant argues, see remarks, pages 17-18, that cited references do not teach all the newly amended limitations of claim 1.
In reply, Examiner disagrees because the new 103 combination made in view of Zhang, thereby teaching all the newly amended limitations of claim 1.
For instance, Li teaches determining, for respective object, an importance based on respective loss function values for the respective object (importance of the first training sample to the verification loss is evaluated based on the impact function value. In other words, the importance of the first training sample is determined by measuring impact of data disturbance on a verification loss of the first recommendation model on the second training sample, paragraph 458);
adjusting, for respective object in the sample image, a weight of a corresponding task loss term used to obtain the respective loss function value for the respective object, based on the determined importance (the weight of the first training sample consisting of respective object(s) is adjusted based on determined importance of second training sample, in other words, the importance of the first training sample is determined by measuring impact of data disturbance on a verification loss of the first recommendation model on a second training sample consisting of said respective object(s) in second sample image, so that the recommendation model can more accurately fit data distribution without a bias resulting in retraining based on the first training sample with a weight that can better fit data distribution of the second training sample, to improve accuracy of predicting the second training sample by the recommendation model, paragraphs 208, 458).
And the newly cited reference, Zhang teaches obtaining, for each respective object in a sample image, respective processing results of each of at least two tasks for detecting the respective object, and respective loss function values corresponding to the respective processing results of the at least two tasks for the respective object after performing processing in a neural network on a sample image (obtaining a face data set, where the face data set includes a plurality of face images with image tag information, and the image tag information includes: first face frame classification information labeled, first person The position information of the face frame and the coordinate information of the first face key point and using the face data set as the input of the preset face detection and alignment network model for model training to output face label information predicted by the network, wherein the face label information includes the first face image Two face frame position information, second face frame classification information, and second face key point coordinate information. In an embodiment, preferably, the preset face detection and alignment network model uses ResNet50 as a basic skeleton and has an image feature pyramid structure to calculate the first loss function corresponding to the face frame position information, the second loss function corresponding to the face frame classification information, and the third loss corresponding to the face key point coordinate information according to the image tag information and the face tag information function, paragraphs 65-67, also paragraphs 30-40);
wherein the neural network outputs the respective processing results of the at least two tasks for each respective object in the sample image (output the first weighting factor corresponding to the first loss function, the second weighting factor corresponding to the second loss function, and the third weight corresponding to the third loss function through the fully connected layer of the network model factor such that the face detection results and aligned face images can be output through a network end-to-end training, which saves time during engineering deployment, improves detection efficiency, and can adaptively learn weight factors through the network training process to balance Loss between different branch tasks, while improving the effect of detection and key point positioning, paragraphs 68, 75);
determining, for each respective object in the sample image, a factor for each of the at least two tasks based on the respective loss function values of the at least two tasks for the respective object (determining the total loss function according to each loss function and its corresponding weight factor, paragraph 69);
adjusting, for each respective object in the sample image, a weight of a corresponding task loss term used to obtain the respective loss function value of each of the at least two tasks for the respective object (for each object such as based on first and second face object detection and corresponding first and second frame information tasks, weight of each task is updated and adjusted according to the total loss function back propagation to determine the target face detection and alignment network model preventing the weight factor for each object from being too large, paragraphs 73-74);
and updating network based on the corresponding task loss terms whose weights are adjusted (updating for each first frame and second frame tasks whose weights are adjusted according to the total loss function back propagation to determine the target face detection and alignment network model to improve the effect of detection and key point positioning, paragraphs 74-75).
Applicant’s rest of the arguments have been rendered moot because they are based on same repeated assertions arguing the same limitations over and over and which have been successfully taught by cited references as explained above.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are: “storage module”, “receiving module” and “processing module” in claim 23.
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
A review of the specification, shows that the following appears to be the corresponding structure described in the specification, according to PG-Pub, for the 35 U.S.C. 112(f) or pre- AIA 35 U.S.C. 112, sixth paragraph limitation:
storage module - CPU executes various application programs stored in the ROM 130 or the hard disk 140, such as a memory, paragraph 35
receiving module - input device 150 receives a neural network, paragraph 43
processing module - CPU 110 as a processor, and may perform various functions to be described hereinafter by executing various application programs stored, paragraph 35
If applicant wishes to provide further explanation or dispute the examiner’s interpretation of the corresponding structure, applicant must identify the corresponding structure with reference to the specification by page and line number, and to the drawing, if any, by reference characters in response to this Office action.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
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 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, 6, 9, 21 and 23-24 are rejected under 35 U.S.C. 103 as being unpatentable over Tang et al., US 2023/0401446 in view of Li et al., US 2023/0153857 further in view of Zhang, CN 111738077A.
Regarding claim 1, Tang discloses a method of training a neural network (training device may first select, from a plurality of existing types of convolutional neural networks, a convolutional neural network that is suitable for processing the target task, paragraph 113), comprising:
obtaining, for respective object in a sample image, respective processing results of at least two tasks, and respective loss function values corresponding to the at least two tasks after performing processing in a neural network on a sample image (trained target convolutional neural network is obtained through training based on the objective loss function constructed and the objective loss function may be specifically obtained based on a first sub-loss function and a second sub-loss function, wherein, the first sub-loss function is determined based on target task, and the second sub-loss function is determined based on a channel importance function and a dynamic weight. The first sub-loss function represents a difference between a training sample input to the target convolutional neural network and an output prediction result, paragraph 180, and note that target task could be based on plurality of task such as i.e., at least two tasks, for example, target task can be a classification task making the first sub-loss function a cross-entropy loss function. Wherein, the target task can be another specific task (for example, a regression task), where the first sub-loss function is be a common loss function such as a perceptual loss function or a hinge loss function. Selection of the first sub-loss function is specifically related to a type of the target task, and processing’s related to plurality of tasks could be obtained, paragraph 114);
wherein the neural network outputs the respective processing results of the at least two tasks for respective object in the sample image (results for each type (both types) of target task is outputted, for example, when the target task is classification task, it classifies the input image by using the sub-network, and outputs a classification result, that is, output data is the classification result of the image, paragraphs 34-35, else when the target task a regression task, it outputs different result, paragraph 114, for detecting/predicting the target task to obtain a detection result which is presented on a display interface, not that, target detection task is usually for detecting a target object in an image. In this case, the input data is usually an input image. The execution device prunes the trained target convolutional neural network based on the input image to obtain a pruned sub-network, and performs target detection on the input image by using the sub-network to obtain a detection result, that is, output data is the detection result, paragraphs 33, 105);
determining, for respective object in the sample image, an importance of the at least two tasks based on the respective loss function values of the at least two tasks for the respective object (training device further obtains the second sub-loss function based on the channel importance function and the dynamic weight which is determined based on the first sub-loss function belonging to the type of associated target tasks (i.e., at least two tasks) such as classification and regression task as explained above to identify similarity between different training samples corresponding to different target tasks which are input to different sub-networks, paragraphs 118, 152).
Tang fails to explicitly disclose obtaining, for each respective object in a sample image, respective processing results of each of at least two tasks for detecting the respective object, and respective loss function values corresponding to the respective processing results of the at least two tasks for the respective object; outputting respective processing results of the at least two tasks for each respective object in the sample image; determining, for each respective object, an importance of each of the at least two tasks based on the respective loss function values; adjusting, for each respective object in the sample image, a weight of a corresponding task loss term used to obtain the respective loss function value of each of the at least two tasks for the respective object, based on the determined importance; and updating the neural network based on the corresponding task loss terms whose weights are adjusted.
However, Li teaches determining, for respective object, an importance based on respective loss function values for the respective object (importance of the first training sample to the verification loss is evaluated based on the impact function value. In other words, the importance of the first training sample is determined by measuring impact of data disturbance on a verification loss of the first recommendation model on the second training sample, paragraph 458);
adjusting, for respective object in the sample image, a weight of a corresponding task loss term used to obtain the respective loss function value for the respective object, based on the determined importance (the weight of the first training sample consisting of respective object(s) is adjusted based on determined importance of second training sample, in other words, the importance of the first training sample is determined by measuring impact of data disturbance on a verification loss of the first recommendation model on a second training sample consisting of said respective object(s) in second sample image, so that the recommendation model can more accurately fit data distribution without a bias resulting in retraining based on the first training sample with a weight that can better fit data distribution of the second training sample, to improve accuracy of predicting the second training sample by the recommendation model, paragraphs 208, 458);
and updating the neural network based on the corresponding task loss terms whose weights are adjusted (recommendation model based on neural network is trained and continuously updated based on reverse gradient of the value of the loss function whose weight is adjusted, to improve a prediction effect of the recommendation model, paragraphs 114, 457).
Tang and Li are combinable because they both are in the same field of endeavor dealing with neural networks involving loss functions.
Therefore, it would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Tang to incorporate the teachings of Li to provide adjusting of a weight of a loss function based on determined importance for the benefit of improving accuracy of results of the recommendation model, which provides improved user experience as taught by Li at paragraph 9.
Tang with Li fails to explicitly teach obtaining, for each respective object in a sample image, respective processing results of each of at least two tasks for detecting the respective object, and respective loss function values corresponding to the respective processing results of the at least two tasks for the respective object; outputting respective processing results of the at least two tasks for each respective object in the sample image; determining, for each respective object, a factor for each of the at least two tasks based on the respective loss function values; adjusting, for each respective object in the sample image, a weight of a corresponding task loss.
However, Zhang teaches obtaining, for each respective object in a sample image, respective processing results of each of at least two tasks for detecting the respective object, and respective loss function values corresponding to the respective processing results of the at least two tasks for the respective object after performing processing in a neural network on a sample image (obtaining a face data set, where the face data set includes a plurality of face images with image tag information, and the image tag information includes: first face frame classification information labeled, first person The position information of the face frame and the coordinate information of the first face key point and using the face data set as the input of the preset face detection and alignment network model for model training to output face label information predicted by the network, wherein the face label information includes the first face image Two face frame position information, second face frame classification information, and second face key point coordinate information. In an embodiment, preferably, the preset face detection and alignment network model uses ResNet50 as a basic skeleton and has an image feature pyramid structure to calculate the first loss function corresponding to the face frame position information, the second loss function corresponding to the face frame classification information, and the third loss corresponding to the face key point coordinate information according to the image tag information and the face tag information function, paragraphs 65-67, also paragraphs 30-40);
wherein the neural network outputs the respective processing results of the at least two tasks for each respective object in the sample image (output the first weighting factor corresponding to the first loss function, the second weighting factor corresponding to the second loss function, and the third weight corresponding to the third loss function through the fully connected layer of the network model factor such that the face detection results and aligned face images can be output through a network end-to-end training, which saves time during engineering deployment, improves detection efficiency, and can adaptively learn weight factors through the network training process to balance Loss between different branch tasks, while improving the effect of detection and key point positioning, paragraphs 68, 75);
determining, for each respective object in the sample image, a factor for each of the at least two tasks based on the respective loss function values of the at least two tasks for the respective object (determining the total loss function according to each loss function and its corresponding weight factor, paragraph 69);
adjusting, for each respective object in the sample image, a weight of a corresponding task loss term used to obtain the respective loss function value of each of the at least two tasks for the respective object (for each object such as based on first and second face object detection and corresponding first and second frame information tasks, weight of each task is updated and adjusted according to the total loss function back propagation to determine the target face detection and alignment network model preventing the weight factor for each object from being too large, paragraphs 73-74);
and updating network based on the corresponding task loss terms whose weights are adjusted (updating for each first frame and second frame tasks whose weights are adjusted according to the total loss function back propagation to determine the target face detection and alignment network model to improve the effect of detection and key point positioning, paragraphs 74-75).
Tang and Li are combinable with Zhang because they all are in the same field of endeavor dealing with detecting objects in image and obtaining corresponding loss functions.
Therefore, it would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the teachings of Tang and Li to incorporate the teachings of Zhang to provide adaptive adjusting of learned weight factors during the network training process to balance the loss between different branch tasks in order to improve face detection and alignment as taught by Zhang at paragraph 10.
Regarding claim 6, Combination of Tang with Li and Zhang teaches wherein, in the determining, the processing results are sorted according to the loss function values, and the importance of the processing result is determined based on a sorted order thereof (Tang, average importance of the convolution kernels (that is, the channels) in the trained target convolutional neural network for different input data are sorted and only a channel whose value of the channel importance function is larger than the threshold needs to be retained with a channel whose value of the channel importance function is smaller than the threshold is pruned, paragraph 184).
Regarding claim 9, Combination of Tang with Li and Zhang teaches wherein, the at least one network structure in the neural network can process one or more tasks (Tang, constructing a network structure of a convolutional neural network that meets the target task, paragraph 113).
Regarding claim 21, Combination of Tang with Li and Zhang teaches a device of training a neural network (Tang, training device may first select, from a plurality of existing types of convolutional neural networks, a convolutional neural network that is suitable for processing the target task, paragraph 113), comprising:
one or more memories; and one or more processors to execute instruction stored in the memories to function (Tang, as described in paragraphs 217, 235) as:
an obtaining unit (Tang, CPU, paragraph 209); a determining unit (Li, processing unit 3020, paragraph 610); an adjustment unit (Li, processing unit 3020, paragraph 610); and an updating unit (Li, processing unit 3020, paragraph 610);
Rest of the claim recites similar features as claim 1 and thus is rejected on the same rationale.
Regarding claim 23, Combination of Tang with Li and Zhang teaches a neural network application device comprising: a storage module configured to store a neural network trained based on a training method (Tang, store the training set into the database 230, paragraph 103); a receiving module configured to receive a data set corresponding to requirements of a task that the neural network can perform (Tang, training device may first select, from a plurality of existing types of convolutional neural networks, a convolutional neural network that is suitable for processing the target task, paragraph 113); a processing module (Tang, CPU, paragraph 209) configured to process the data set in each layer from top to bottom in the neural network, and output a result of the process (paragraphs 74, 154-161, all the layers from first to last of the neural network are processed and result as a representation form of a similarity between feature maps is outputted by the last layer of the target convolutional neural network for different training samples);
Rest of the claim recites similar features as claim 1 and thus is rejected on the same rationale.
Regarding claim 24, which recites a non-transitory computer readable storage medium version of claim 1, see rationale as applied above. Note that non-transitory computer readable storage medium is taught by Tang in paragraph 235.
Claims 2-3 are rejected under 35 U.S.C. 103 as being unpatentable over Tang et al., US 2023/0401446 in view of Li et al., US 2023/0153857 further in view of Zhang, CN 111738077A as applied in claim 1 above and further in view of Stent et al., US 2020/0074589.
Regarding claim 2, Tang further discloses wherein, in the determining, regarding tasks within same object in the sample image, determining importance (paragraph 209, determine a first sub-loss function based on a target task, where the first sub-loss function represents a difference between a training sample input to a convolutional neural network (that is, a target convolutional neural network) on which pruning processing is to be performed and an output prediction result. For example, when the target task is a classification task, the first sub-loss function may be a cross-entropy loss function. In addition to determining the first sub-loss function based on the target task, the processor may further obtain a second sub-loss function based on a channel importance function and a dynamic weight).
Combination of Tang with Li and Zhang fails to explicitly teach regarding different tasks, determining importance of processing result of each task is determined.
However, Stent teaches regarding different tasks within same object in sample image, determining importance of processing result of each task is determined (sampling process may be considered in two stages. In the first stage, a CNN may be used to produce a saliency map from the original high-resolution image. This map may be task specific, as for each different task different regions of the original high-resolution image may be relevant. In the second stage, the original high-resolution image may be sampled in proportion to their importance, paragraph 41).
Tang with Li and Zhang are combinable with Stent because they all are in the same field of endeavor dealing with neural networks involving tasks.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang with Li and Zhang with the teachings of Stent for the benefit of providing improved systems and methods for sampling datasets to reduce the loss of relevant details as taught by Stent at paragraphs 3-4.
Regarding claim 3, Combination of Tang with Li, Zhang and Stent further teaches wherein, in the determining, regarding same tasks across different objects in the sample image, determining importance of the processing result of each task is determined (Stent, “system may achieve this by analyzing the original high-resolution image 520 and then sampling areas of the original high-resolution image 520 in proportion to their perceived importance. That is, the CNN model may capture the benefit of increased resolution without significant and/or additional computational burdens or the risk of overfitting. For example and without limitation, the sampling process may be considered in two stages. In the first stage, a CNN may be used to produce a saliency map 530 from the original high-resolution image 520. This map may be task specific, as for each different task different regions of the original high-resolution image 520 may be relevant. In the second stage, the original high-resolution image 520 may be sampled in proportion to their importance as indicated by the saliency map 530. In some embodiments, a SoftMax operation may be applied to the saliency map to normalize the output map”, paragraph 41).
Tang with Li and Zhang are combinable with Stent because they all are in the same field of endeavor dealing with neural networks involving tasks.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang with Li and Zhang with the teachings of Stent for the benefit of providing improved systems and methods for sampling datasets to reduce the loss of relevant details as taught by Stent at paragraphs 3-4.
Claims 4-5 and 7-8 are rejected under 35 U.S.C. 103 as being unpatentable over Tang et al., US 2023/0401446 in view of Li et al., US 2023/0153857 further in view of Zhang, CN 111738077A as applied in claim 1 above and further in view of Turner et al., US 2022/0414469.
Regarding claim 4, Combination of Tang with Li and Zhang fails to explicitly teach the greater the loss function value of the processing result is, the higher the importance of the processing result is.
However, Turner teaches the greater the loss function value of the processing result is, the higher the importance of the processing result is (Lagrangian multiplier λ can adjust the relative importance between enforcing the constraint and minimizing the loss function L(w). A higher value of λ would indicate enforcing the constraints has higher weight and the value of L(w) might not be optimized properly. A lower value of λ would indicate that optimizing the loss function is more important and the constraints might not be satisfied, paragraph 58).
Tang, Li and Zhang are combinable with Turner because they all are in the same field of endeavor dealing with neural networks with loss function.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang, Li and Zhang with the teachings of Turner for the benefit of providing machine learning using artificial neural networks for emulating intelligence that are trained for assessing risks or performing other operations and for providing explainable outcomes associated with these outputs as taught by Turner at paragraph 2.
Regarding claim 5, Combination of Tang, Li and Zhang fails to explicitly teach the greater the loss function value of processing result is, the lower the importance of the processing result is.
However, Turner teaches the greater the loss function value of processing result is, the lower the importance of the processing result is (Lagrangian multiplier λ can adjust the relative importance between enforcing the constraint and minimizing the loss function L(w). A higher value of λ would indicate enforcing the constraints has higher weight and the value of L(w) might not be optimized properly. A lower value of λ would indicate that optimizing the loss function is more important and the constraints might not be satisfied, paragraph 58).
Tang, Li and Zhang are combinable with Turner because they all are in the same field of endeavor dealing with neural networks with loss function.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang, Li and Zhang with the teachings of Turner for the benefit of providing machine learning using artificial neural networks for emulating intelligence that are trained for assessing risks or performing other operations and for providing explainable outcomes associated with these outputs as taught by Turner at paragraph 2.
Regarding claim 7, Combination of Tang with Li further teaches wherein, in the determining, in a case where the loss function is a regression loss function or an intersection-over-union loss function (Li, first recommendation model is a logistic regression model, and a loss function of the first recommendation model is a loss function with a weight, paragraph 455), the importance of the processing result is determined based on a likelihood value of the loss function value (Li, importance of the first training sample to the verification loss is evaluated based on the impact function value. In other words, the importance of the first training sample is determined by measuring impact of data disturbance on a verification loss of the first recommendation model on the second training sample, paragraph 458).
Tang and Li are combinable because they both are in the same field of endeavor dealing with neural networks involving loss functions.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang with the teachings of Li for the benefit of improving accuracy of results of the recommendation model, which provides improved user experience as taught by Li at paragraph 9.
Combination of Tang with Li and Zhang fails to explicitly teach wherein, the greater the likelihood value is, the lower the importance of the processing result is.
However, Turner teaches wherein, the greater the likelihood value is, the lower the importance of the processing result is (Lagrangian multiplier λ can adjust the relative importance between enforcing the constraint and minimizing the loss function L(w). A higher value of λ would indicate enforcing the constraints has higher weight and the value of L(w) might not be optimized properly. A lower value of λ would indicate that optimizing the loss function is more important and the constraints might not be satisfied, paragraph 58).
Tang, Li and Zhang are combinable with Turner because they all are in the same field of endeavor dealing with neural networks with loss function.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang, Li and Zhang with the teachings of Turner for the benefit of providing machine learning using artificial neural networks for emulating intelligence that are trained for assessing risks or performing other operations and for providing explainable outcomes associated with these outputs as taught by Turner at paragraph 2.
Regarding claim 8, Combination of Tang with Li further teaches wherein, in the determining, in a case where the loss function is a regression loss function or an intersection-over-union loss function (Li, first recommendation model is a logistic regression model, and a loss function of the first recommendation model is a loss function with a weight, paragraph 455), the importance of the processing result is determined based on a likelihood value of the loss function value (Li, importance of the first training sample to the verification loss is evaluated based on the impact function value. In other words, the importance of the first training sample is determined by measuring impact of data disturbance on a verification loss of the first recommendation model on the second training sample, paragraph 458).
Tang and Li are combinable because they both are in the same field of endeavor dealing with neural networks involving loss functions.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang with the teachings of Li for the benefit of improving accuracy of results of the recommendation model, which provides improved user experience as taught by Li at paragraph 9.
Combination of Tang with Li and Zhang fails to explicitly teach wherein, the greater the likelihood value is, the higher the importance of the processing result is.
However, Turner teaches wherein, the greater the likelihood value is, the higher the importance of the processing result is (Lagrangian multiplier λ can adjust the relative importance between enforcing the constraint and minimizing the loss function L(w). A higher value of λ would indicate enforcing the constraints has higher weight and the value of L(w) might not be optimized properly. A lower value of λ would indicate that optimizing the loss function is more important and the constraints might not be satisfied, paragraph 58).
Tang, Li and Zhang are combinable with Turner because they all are in the same field of endeavor dealing with neural networks with loss function.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang, Li and Zhang with the teachings of Turner for the benefit of providing machine learning using artificial neural networks for emulating intelligence that are trained for assessing risks or performing other operations and for providing explainable outcomes associated with these outputs as taught by Turner at paragraph 2.
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Tang et al., US 2023/0401446 in view of Li et al., US 2023/0153857 further in view of Zhang, CN 111738077A as applied in claim 1 above and further in view of Yin et al., US 2021/0056293.
Regarding claim 10, Combination of Tang with Li and Zhang fails to explicitly teach wherein, in a case where the neural network is a network where tasks are cascaded, the processing result of a latter task is adjusted and obtained based on the processing result of a previous task.
However, Yin teaches wherein, in a case where the neural network is a network where tasks are cascaded (structure of a cascaded convolutional neural network, paragraph 22), the processing result of a latter task is adjusted and obtained based on the processing result of a previous task (“results of the cascaded network structure is also cascaded, which will cause previous wrong results to be irrecoverable at a later stage. In the present method, the angle classification tasks of the preceding two networks are completely identical, both performing classification and prediction in four direction ranges, but is also different in that the input samples of the second-level network contain more positive samples and therefore have more credible prediction results. The angle arbitration mechanism combines the preceding two angle prediction results by providing a predefined threshold value. Specifically, when the prediction result of the second-level network is higher than the threshold, or when the two prediction results having the highest confidence respectively in the two networks are identical, the prediction of the second-level network is taken as the final prediction result”, paragraph 36).
Tang with Li and Zhang are combinable with Yin because they all are in the same field of endeavor dealing with neural networks.
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the application to combine the teachings of Tang with Li and Zhang with the teachings of Yin for the benefit of providing effective and efficient face detection method by applying regression and cascading features as taught by Yin at paragraphs 7 and 22.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Metzler Jr. et al., US 2021/0125108
Kim et al., US 2021/0125056
Zhang et al., US 2021/0027081
Sikka et al., US 20190325243
Alakuijala et al., US 2019/0251444
Tieg et al., US 11,531,879
Learning a Unified Sample Weighting Network for Object Detection, Qi Cai et al., IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, pp. 1-11, December 31, 2020.
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 PAWANDEEP DHINGRA whose telephone number is (571) 270-1231. The examiner can normally be reached 9:00-5:00.
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, Abderrahim Merouan can be reached at (571) 270-5254. 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.
/PAWAN DHINGRA/Examiner, Art Unit 2683
/ABDERRAHIM MEROUAN/Supervisory Patent Examiner, Art Unit 2683
Prosecution Insights