DETAILED ACTION
This action is in response to the original filing on August 28, 2023, and the Remarks and Amendments filed July 9, 2026. Claims 1-20 are pending and have been considered below. Claims 1, 8, and 15 are independent claims. Claims 1-2, 8-9, and 15-16 are amended.
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
In review of Applicant’s amendments, filed July 9, 2026, the objections to the specification and drawings made in the previous office action have been withdrawn.
The rejections of claims 1-2, 8-9, and 15-16 under 112(b) set forth in the previous office action are withdrawn in view of the amendments to the claims. However, Examiner notes that Applicant has not amended claims 5-6, 12-13, and 18-19. Therefore, their rejections under 112(b) remain in place.
Response to Arguments
Applicant’s arguments and amendments filed July 9, 2026, regarding the abstract idea rejections from the previous office action made under 35 U.S.C. 101 have been fully considered but are not persuasive.
On page 18 of the Remarks, Applicant argues that “The claimed invention, given the claim amendments presented above, is eligible under 35 U.S.C. § 101.” Applicant cites as reasoning “Director Squires’ Decision on Request for Rehearing in Ex parte Desjardins, Appeal 20204-000567, Application 16/319,040 (hereinafter “Director Squires’ Decision”)” pertaining to the how the amended claims “reflect the improvements in machine learning model that are identified by the Specification” and pertaining to how paragraph 21 of the specification describes “Embodiments of the present disclosure may improve on at least some of the challenges and issues described above” for example how current available solutions “fail to give an opportunity to easily add diversity” and “suffer from limitations to train ensembles based on diversity and fail to provide well-calibrated uncertainty estimations” as described in paragraph 20 of the specification. Applicant submits that claim 1 is patent eligible because it reflects the “improvement in machine learning models, particularly ‘a DNN with a budding ensemble architecture’” identified in the specification. Examiner respectfully disagrees. Although applicant asserts that the amended features in claim 1 result in “well-calibrated uncertainty estimation” and that the described DNN architecture “may be deployed for anchor-based object detection tasks,” as supported in paragraph 21 of the specification, the specification only sets forth the improvements in a conclusory manner, while the claim itself fails to reflect the disclosed improvements of training a neural network with the “budding ensemble architecture” based on “diversity loss functions” as described in the amended claim (See MPEP 2106.04(d)(1) “if the specification explicitly sets forth an improvement but in a conclusory (i.e., 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”).
Referring to Squires’ Decision, paragraph 21 of the specification for Application 16/319,040 describes how “By training the same machine learning model on multiple tasks as described in this specification… the model can be used for each of the multiple tasks with an acceptable level of performance… systems that need to be able to achieve acceptable performance on multiple tasks can do so while using less of their storage capacity and having reduced system complexity… by maintaining a single instance of a model rather than multiple different instances of a model” while claim 1 describes “training the machine learning model on the second machine learning task by training the machine learning model on the second training data to adjust the first values of the plurality of parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task.” A person of ordinary skill in the art can reasonably discern how the described invention results in the improvement in technology as explained in paragraph 21 of the specification. For the present application, it is unclear how a person of ordinary skill in the art would find it apparent that training the neural network with a budding ensemble architecture with diversity loss as described in amended claim 1 can “facilitate well-calibrated uncertainty estimation” or “be deployed for anchor-based object detection tasks” as explained in paragraph 21 of the specification.
Applicant’s arguments, filed July 9, 2026 regarding the rejections from the previous office action made under 35 U.S.C. 103 have been fully considered but are moot as they do not apply the reference Cates being used in the current rejections of claims 1, 8, and 15 and their associated dependent claims 2-7, 9-14, and 16-20, respectively, to teach the amended claim limitations directed to the neural network head ensemble architecture.
Specifically, Cates teaches the amended claim limitation wherein the neural network has two head ensembles, each with two heads. While Cates describes each head sharing the same layer architecture, one of ordinary skill in the art may combine the architecture of Cates with the neural network architecture with diverse heads of Ro, which has been experimentally demonstrated to outperform “the state-of-the-art algorithms by margins of over 5.3%, 6.1%, 5.9%, and 1.8%” on various datasets (see Ro, Abstract).
With the addition of the reference Cates teaching the subject matter introduced in the amendments, the rejections under 35 U.S.C. 103 stand.
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 5-6, 12-13, and 18-19 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 the one or more other operations in claim 5 lacks sufficient antecedent basis as claim 2 includes both one or more other deep learning operations that compute a third detection tensor and one or more other deep learning operations that compute a fourth detection tensor and it is unclear which operations the applicant is referring to. For examination purposes this term is interpreted to mean “one or more other deep learning operations that compute a third detection tensor” in claim 2.
The term the one or more other operations in claim 6 lacks sufficient antecedent basis as claim 2 includes both one or more other deep learning operations that compute a third detection tensor and one or more other deep learning operations that compute a fourth detection tensor and claim 5 includes the one or more other operations and it is unclear which operations the applicant is referring to. For examination purposes this term is interpreted to mean “the one or more other operations” in claim 5 (see above).
Claims 12-13 recite a non-transitory computer-readable media that parallels the method claims of 5-6, respectively. Therefore, the analysis discussed above with respect to claims 5-6 applies to claims 12-13, respectively. Accordingly, claims 12-13 are rejected based on substantially the same rationale as set forth above with respect to claims 5-6, respectively.
Claims 18-19 recite an apparatus that parallels the method claims of 5-6, respectively. Therefore, the analysis discussed above with respect to claims 5-6 applies to claims 18-19, respectively. Accordingly, claims 18-19 are rejected based on substantially the same rationale as set forth above with respect to claims 5-6, respectively.
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 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1:
Step 1 – Claim 1 is directed to a method: A method of training a neural network…
Step 2A, Prong 1 – A judicial exception is recited in this claim as it recites mathematical concepts and a mental process (see MPEP 2106.04(a)(2)(I) and MPEP 2106.04(a)(2)(III)):
selecting a layer in a backbone of the neural network, the layer outputting an intermediate tensor based on the training dataset… A person can reasonably perform “selecting a layer in a backbone of the neural network” given a small enough number of layers in a neural network. Hence, “selecting a layer in a backbone of the neural network, the layer outputting an intermediate tensor based on the training dataset” is a mental process.
determining a loss for the neural network, the loss comprising a diversity loss, the diversity loss indicating a measurement of similarity between the first detection tensor and the second detection tensor… To determine a “loss comprising a diversity loss… indicating a measurement of similarity” involves using a formula to calculate a similarity between two tensors, which is a mathematical concept. Hence, “determining a loss for the neural network, the loss comprising a diversity loss, the diversity loss indicating a measurement of similarity between the first detection tensor and the second detection tensor” is a mathematical concept.
training the neural network by adjusting one or more weights in the backbone, first head, or second head based on the loss… To train a neural network “by adjusting one or more weights” involves following an algorithm, for example gradient descent, to modify the weights of a neural network, which is a mathematical concept. Hence, “training the neural network by adjusting one or more weights in the backbone, first head, or second head based on the loss” is a mathematical concept.
Step 2A, Prong 2 – The following limitations are additional elements that fail to implement the abstract ideas into a practical application:
inputting a training dataset into the neural network, wherein the neural network comprises a backbone, a first head ensemble comprising a first head and a second head, and a second head ensemble comprising a third head and a fourth head, and wherein the first head has a same layer as the second head and a different layer from the third head… “inputting a training dataset” amounts to insignificant extra-solution activity of necessary data gathering that does not add a meaningful limitation to the “method” (see MPEP 2106.05(g)).
inputting the intermediate tensor into the first head, the first head comprising one or more deep learning operations that compute a first detection tensor… “inputting the intermediate tensor” amounts to insignificant extra-solution activity of necessary data gathering that does not add a meaningful limitation to the “method” (see MPEP 2106.05(g)).
inputting the intermediate tensor into the second head, the second head comprising one or more deep learning operations that compute a second detection tensor that is different from the first detection tensor… “inputting the intermediate tensor” amounts to insignificant extra-solution activity of necessary data gathering that does not add a meaningful limitation to the “method” (see MPEP 2106.05(g)).
Step 2B – These elements are recited at such a high level of generality that they fail to integrate the abstract idea into a practical application, since they only amount to data gathering or outputting (MPEP 2106.05(g)) without significantly more. These limitations, either taken alone or in combination, fail to provide an inventive concept. Thus, the claim is not patent eligible.
Claims 2-7 recite limitations which further narrow the abstract ideas of claim 1 by specifying more details of the mathematical concepts that occur:
Regarding claim 2, this claim further limits the abstract ideas of claim 1 to be based on a mathematical concept and mental process: selecting another layer in the backbone of the neural network, the another layer generating another intermediate tensor based on the training dataset… Given a small enough number of layers in the neural network, selecting another layer can reasonably be performed in the human mind, which is a mental process. Furthermore, wherein the loss further comprises another diversity loss that indicates a measurement of similarity between the third detection tensor and the fourth detection tensor… Another diversity loss that indicates a measurement of similarity between two tensors involves using a mathematical formula to calculate a measurement of similarity, which is a mathematical concept. Furthermore, specifying inputting the another intermediate tensor into the third head, the third head comprising one or more other deep learning operations that compute a third detection tensor… and inputting the another intermediate tensor into the fourth head, the fourth head comprising one or more other deep learning operations that compute a fourth detection tensor that is different from the third detection tensor… is still insignificant extra-solution activity of necessary data gathering (see MPEP 2106.05(g)).
Regarding claim 3, specifying wherein the another intermediate tensor has a different size from the intermediate tensor in this manner does not overcome the rejection of claim 1 as modifying the another intermediate tensor does not make the abstract ideas of claim 1 to not be mathematical concepts or mental processes.
Regarding claim 4, specifying inputting the first detection tensor into the third head, the third detection tensor computed based on the first detection tensor and the another intermediate tensor… and inputting the second detection tensor into the fourth head, the fourth detection tensor computed based on the second detection tensor and the another intermediate tensor… is still insignificant extra-solution activity of necessary data gathering (see MPEP 2106.05(g)).
Regarding claim 5, this claim further limits the abstract ideas of claim 4 to be based on a mathematical concept: wherein the one or more other operations comprise an upsampling operation that computes an upsampled tensor by increasing one or more dimensions of the first detection tensor… An upsampling operation that computes an upsampled tensor involves padding the values in a tensor with additional zeroes or values, which is a mathematical concept.
Regarding claim 6, this claim further limits the abstract ideas of claim 5 to be based on a mathematical concept: wherein the one or more other operations further comprise a concatenation operation that computes a tensor by concatenating the upsampled tensor with the another intermediate tensor… A concatenation operation that computes a tensor involves appending two tensors together to create a single, larger tensor, which is a mathematical concept.
Regarding claim 7, this claim further limits the abstract ideas of claim 1 to be based on a mathematical concept: wherein the measurement of similarity comprises a measurement of centered kernel alignment similarity between the first detection tensor and the second detection tensor or a measurement of cosine similarity between the first detection tensor and the second detection tensor… A measurement of centered kernel alignment similarity or a measure of cosine similarity both involve calculating the similarity between two sets of data, which is a mathematical concept.
Regarding claim 8:
Step 1 – Claim 8 is directed to a product: one or more non-transitory computer-readable media…
Step 2A, Prong 1 – A judicial exception is recited in this claim as it recites mathematical concepts and a mental process (see MPEP 2106.04(a)(2)(I) and MPEP 2106.04(a)(2)(III)): see claim 1 above.
Step 2A, Prong 2 – The following limitations are additional elements that fail to implement the abstract ideas into a practical application:
one or more non-transitory computer-readable media storing instructions executable to perform operations for training a neural network… one or more non-transitory computer-readable media used as mere tools to apply an exception are generic elements for performing or applying the abstract idea using a generic computing environment (see MPEP 2106.05(f)).
inputting a training dataset into the neural network, wherein the neural network comprises a backbone, a first head ensemble comprising a first head and a second head, and a second head ensemble comprising a third head and a fourth head, and wherein the first head has a same layer as the second head and a different layer from the third head… “inputting a training dataset” amounts to insignificant extra-solution activity of necessary data gathering that does not add a meaningful limitation to the “non-transitory computer-readable media” (see MPEP 2106.05(g)).
inputting the intermediate tensor into a first head of the neural network, the first head comprising one or more deep learning operations that compute a first detection tensor… “inputting the intermediate tensor” amounts to insignificant extra-solution activity of necessary data gathering that does not add a meaningful limitation to the “non-transitory computer-readable media” (see MPEP 2106.05(g)).
inputting the intermediate tensor into a second head of the neural network, the second head comprising the one or more deep learning operations that compute a second detection tensor that is different from the first detection tensor… “inputting the intermediate tensor” amounts to insignificant extra-solution activity of necessary data gathering that does not add a meaningful limitation to the “non-transitory computer-readable media” (see MPEP 2106.05(g)).
Step 2B – These elements are recited at such a high level of generality that they fail to integrate the abstract idea into a practical application, since they provide nothing more than mere instructions to implement an abstract idea on a generic computer (MPEP 2106.05(f)) or only amount to data gathering or outputting (MPEP 2106.05(g)) without significantly more. These limitations, either taken alone or in combination, fail to provide an inventive concept. Thus, the claim is not patent eligible.
Claims 9-14 recite a non-transitory computer-readable media that parallels the method claims of 2-7, respectively. Therefore, the analysis discussed above with respect to claims 2-7 applies to claims 9-14, respectively. Accordingly, claims 9-14 are rejected based on substantially the same rationale as set forth above with respect to claims 2-7, respectively.
Regarding claim 15:
Step 1 – Claim 15 is directed to an apparatus: an apparatus…
Step 2A, Prong 1 – A judicial exception is recited in this claim as it recites mathematical concepts and a mental process (see MPEP 2106.04(a)(2)(I) and MPEP 2106.04(a)(2)(III)): see claims 1 and 8 above.
Step 2A, Prong 2 – The following limitations are additional elements without significantly more than the abstract idea:
a computer processor for executing computer program instructions… a computer processor used as a mere tool to apply an exception is a generic element for performing or applying the abstract idea using a generic computing environment (see MPEP 2106.05(f)).
a non-transitory computer-readable memory storing computer program instructions executable by the computer processor to perform operations for training a neural network… a non-transitory computer-readable memory used as a mere tool to apply an exception is a generic element for performing or applying the abstract idea using a generic computing environment (see MPEP 2106.05(f)).
Claims 15-20 recite an apparatus that parallels the computer-readable medium claims of 8-9 and 11-14, respectively. Therefore, the analysis discussed above with respect to claims 8-9 and 11-14 applies to claims 15-20, respectively. Accordingly, claims 15-20 are rejected based on substantially the same rationale as set forth above with respect to claims 8-9 and 11-14, respectively.
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 1-3, and 7 are rejected under 35 U.S.C. 103 as being unpatentable over Ro et al. (“Heterogeneous Double-Head Ensemble for Deep Metric Learning,” 2020, hereinafter Ro) in view of Cates et al. (“More Is More – Narrowing the Generalization Gap by Adding Classification Heads,” 2021, hereinafter Cates) and further in view of Li et al. (“On the diversity of multi-head attention, 2021, hereinafter Li).
Regarding claim 1:
Regarding the limitation a method of training a neural network, comprising: inputting a training dataset into the neural network, wherein the neural network comprises a backbone, a first head ensemble comprising a first head and a second head, and a second head ensemble comprising a third head and a fourth head, and wherein the first head has a same layer as the second head and a different layer from the third head, Ro teaches a method of training a neural network, comprising: inputting a training dataset into the neural network, wherein the neural network comprises a backbone, a first head ensemble comprising a first head and a second head (Page 1, Col. 1, Section 1, ¶1 “Deep metric learning refers to the design of feature extracting functions with deep neural networks so that the features of semantically similar images are close to others. Ensemble is a method of ensuring robust performance by training diverse models and aggregating their prediction results,” Page 2, Col. 2, Section 3, ¶1 “The multi-head structure for deep metric learning,” Page 5, Col. 2, Section IV, ¶1 “we provide experimental results for evaluating the proposed structure, especially in deep metric learning for image retrieval tasks,” Page 3, Col. 1, ¶1 “For the baseline network, we adopt the ResNet-50 network, most popular backbone model. All experiments covered in this section are conducted on validating the CUB-200–2011 dataset. The CUB-200–2011 training dataset is divided into training and validation sets equally,” Page 7, Col. 2, Section D, ¶1 “The proposed HDhE structure trained by only softmax loss with a random sampling manner… a simple and fast way to train the model,” Page 5, Col. 2, Fig. 3 depicts wherein the neural network comprises a backbone, a first head ensemble comprising a first head and a second head) and a third head and a fourth head (Page 7, Col. 2, ¶1 “triple-head or four-head structure”), and wherein the first head has… a different layer from the third head (Page 7, Col. 1, Section 4, ¶1 “The triple-heads were designed as Res4_3, Res4_2, and Res4_1 of Last layer,” Page 5, Col. 1, Table 5 and Page 7, Table 8, Caption: “the last layer of the head”). However, Ro fails to teach and a second head ensemble comprising a third head and a fourth head, and wherein the first head has a same layer as the second head…
Cates, in the same field of endeavor, teaches and a second head ensemble comprising a third and a fourth head (Page 3, Col. 1, ¶2 “Given an existing NN model h, henceforth called the base model, we can split it to two components: all the layers except for the last one denoted f, and the last layer g assumed to be the fully-connected layer… we enhance the model h by replacing g with |T| = m heads, where each head is an independent fully connected layer gt,” Page 4, Col. 1, Section 3.4, ¶1 “Once trained, the full TransNet model can be viewed as an ensemble of m shared classifiers,” wherein a “base model” with, for example, m = 2 heads, encompasses a first ensemble with a first and second head, Pages 5-6, Col. 2, Section 5, ¶3 “Notations… “Tm-CNN” – a full TransNet model trained with m heads, where all are used for prediction… To denote an ensemble of the models above, we add a suffix of a number in parentheses, e.g. T2-CNN(3) is an ensemble of 3 T2-CNN models,” wherein an exemplary “T2-CNN(3)” model has an ensemble of 3 CNN models each with m = 2 heads or “an ensemble of m shared classifiers,” which is implied to contain a second head ensemble comprising a third and a fourth head), and wherein the first head has a same layer as the second head (Page 3, Col. 1, ¶2 “we enhance the model h by replacing g with |T| = m heads, where each head is an independent fully connected layer gt associated with a specific transformation t… each head is defined by ht = gt ◦ f,” ¶3 “the resulting model… essentially represents m models, which share via f all the weights up to the last fully-connected layer,” wherein it is implicit that all m heads branch from the same point f in the neural network, hence the first head has a same layer as the second head).
Ro and Cates are analogous to the claimed invention as both are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the neural network architecture of Cates with the neural network architecture and training methodology of Ro. The motivation to do so is to design a multi-head architecture to “encourage the model to capture meaningful orientation information” (Cates, Page 9, Col. 1, ¶2).
Ro further teaches selecting a layer in a backbone of the neural network, the layer outputting an intermediate tensor based on the training dataset (Page 2, Col. 2, Section III, ¶1 “the feature vector should be of a small dimension… the structure that shares low-level layers should save memory and not degrade performance,” Page 3, Col. 1, ¶1 “For the baseline network, we adopt the ResNet-50 network, most popular backbone model,” Page 1, Fig. 1 depicts “Shared layers,” or a backbone of the neural network, Page 3, Col. 1, Section A, ¶1 “we set up multiple variants of the last feature block in the direction in order to reduce feature dimension… Then, we select one of the variants,” Page 3, Col. 2, Section B, ¶2 “The layers of ResNet-50 network can be divided into… blocks… We have simply redefined it as {B1, B2, B3, B4},” one of ordinary skill in the art would recognize that individual blocks of a “ResNet-50 network” produce tensors as their outputs, hence outputting an intermediate tensor based on the training dataset is implicit when generating a tensor from an intermediate block, for example B3, in a shared “ResNet-50 network,” or backbone, that receives a training dataset, Page 5, Fig. 3 – (b), Page 5, Col. 2, Section D, ¶2 “we can customize the multi-head structure in various views. For example in Figure 3-(b), a memory-saving version of HDhE (Ms-HDhE) can be suggested as follows. In Ms-HDhE, the shared body is enlarged Until B3,” wherein “customizing” the last block in the shared backbone, for example B3, encompasses selecting a layer in a backbone of the neural network, as the last layer in a selected block is implied to be the layer outputting an intermediate tensor based on the training dataset).
Ro further teaches inputting the intermediate tensor into the first head, the first head comprising one or more deep learning operations that compute a first detection tensor (Page 1, Col. 1, Section I, ¶2 “A variety of feature vectors can be obtained by semantically diverse attention of the input image,” wherein “feature vectors” generated from an “input image” encompasses detection tensors, Page 2, Col. 1, ¶2 “we search for an effective multi-head ensemble structure referred to as Heterogeneous Double-head Ensemble (HDhE), as shown in Figure 1… we consider… designing a dimension-reduced feature vector, determining the shared body part, and designing diverse heads to generate diverse feature vectors,” Page 1, Fig. 1 depicts the first head computing a first detection tensor, “feature vector f1,” after receiving an input from a preceding block; one of ordinary skill in the art would recognize that inputting the intermediate tensor into the first head is implicit when using a multi-head architecture that receives an input from an intermediate block of a ResNet-50 backbone, as explained above, Page 5, Fig. 3 – (b) and Col. 2, Section D, ¶2 “the heads can be chosen by Res4_2 and Res4_1 with their learning rates of 0.5× and 4×, respectively,” wherein a first head with a unique block architecture and “learning rate” encompasses one or more deep learning operations that compute a first detection tensor).
Ro further teaches inputting the intermediate tensor into the second head, the second head comprising one or more deep learning operations that compute a second detection tensor that is different from the first detection tensor (Page 2, Col. 1, ¶2 “designing diverse heads to generate diverse feature vectors,” Page 1, Fig. 1 depicts the second head computing a second detection tensor that is different from the first detection tensor, “feature vector f2,” after receiving an input from a preceding block; one of ordinary skill in the art would recognize that inputting the intermediate tensor into the second head is implicit when using a multi-head architecture that receives an input from an intermediate block of a ResNet-50 backbone, as explained above, Page 5, Fig. 3 – (b) and Col. 2, Section D, ¶2 “the heads can be chosen by Res4_2 and Res4_1 with their learning rates of 0.5× and 4×, respectively,” wherein a second head with a different block architecture and “learning rate” from the first head encompasses one or more deep learning operations that compute a second detection tensor that is different from the first detection tensor).
Regarding the limitation determining a loss for the neural network, the loss comprising a diversity loss, the diversity loss indicating a measurement of similarity between the first detection tensor and the second detection tensor, Ro teaches determining a loss for the neural network (Page 2, Col. 1, ¶3 “To show the advantages of our HDhE structure… we show that the HDhE is valid for other types of loss”) and the first detection tensor and the second detection tensor (Page 1, Fig. 1 – f1-f2, Page 2, Col. 1, ¶2 “designing diverse heads to generate diverse feature vectors”). However, Ro fails to teach the loss comprising a diversity loss, the diversity loss indicating a measurement of similarity between the first detection tensor and the second detection tensor.
Li, in the same field of endeavor, teaches the loss comprising a diversity loss, the diversity loss indicating a measurement of similarity between outputs of a first head and a second head (Page 3, Col. 1, Section 3.1, ¶1 “To further guarantee the diversity, we enlarge the distances among multiple attention heads with disagreement regularization. To this end, we introduce an auxiliary regularization term in order to encourage the diversity among multiple attention heads,” Page 3, Col. 2, Equation (7) and ¶2 “Disagreement on Outputs. This disagreement directly applies regularization on the outputs of each attention head, by maximizing the difference among them… we employ negative cosine similarity to measure the distance,” Page 5, Col. 1, Section 3.3, ¶1 “disagreement regularization focuses on adjusting the training objective, i.e. the loss function”).
Ro further teaches training the neural network by adjusting one or more weights in the backbone, first head, or second head based on the loss (Page 6, Col. 1, Section B, ¶1 “our HDhE was trained along with classifier by softmax (cross-entropy) loss. The leaning rates were initially set to 0.001 for convolutional blocks (B1, B2, B3, and B4) and 0.01 for the classifier… The optimizer used in this paper was stochastic gradient descent (SGD) with nesterov momentum. The initial momentum rate and the weight decay were set to 0.9 and 5×10−4, respectively”).
Ro, Cates, and Li are analogous to the claimed invention as all are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the neural network architecture of Cates and the diversity loss of Li with the neural network architecture, loss, and training methodology of Ro. The motivation to do so is to design a multi-head architecture to “encourage the model to capture meaningful orientation information” (Cates, Page 9, Col. 1, ¶2) and a training method to “further improve… model performance” (Li, Abstract).
Regarding claim 2, Ro in view of Cates and further in view of Li teaches the method of claim 1 (and thus the rejection of claim 1 is incorporated).
Ro further teaches selecting another layer in the backbone of the neural network, the another layer generating another intermediate tensor based on the training dataset (Page 3, Col. 1, ¶1 “For the baseline network, we adopt the ResNet-50 network, most popular backbone model… All experiments covered in this section are conducted on validating the CUB-200–2011 dataset. The CUB-200–2011 training dataset is divided into training and validation sets equally,” Page 1, Fig. 1, Page 3, Col. 1, Section A, ¶1 “we set up multiple variants of the last feature block in the direction in order to reduce feature dimension… Then, we select one of the variants,” Page 3, Col. 2, Section B, ¶2 “The layers of ResNet-50 network can be divided into five blocks… We have simply redefined it as {B1, B2, B3, B4},” Page 5, Fig. 3 – (a), Page 4, Col. 1, ¶3 “the Until B2 is selected for our proposed structure,” wherein selecting another block “B2” encompasses selecting another layer in the backbone of the neural network, as explained above with respect to claim 1; one of ordinary skill in the art would recognize that the another layer generating another intermediate tensor based on the training dataset is implicit when generating an output from an intermediate block, for example B2, in a shared “ResNet-50 network,” or backbone, that receives a training dataset).
Ro further teaches inputting the another intermediate tensor into the third head, the third head comprising one or more other deep learning operations that compute a third detection tensor (Page 7, Table 8 and Col. 1, Section 4, ¶1 “Table 8 shows the evaluation results of the model using triple-heads with Until B2 shared structure,” one of ordinary skill in the art would recognize that inputting the another intermediate tensor into the third head is implicit when using a multi-head architecture that receives an output from an intermediate block “B2” of a ResNet-50 backbone, as explained above with respect to claim 1, Page 1, Fig. 1 depicts multiple heads of the neural network outputting diverse vectors after receiving an input, Page 4, Col. 1, Section C, ¶1 “The third design concept of the proposed multi-head structure is to produce diverse feature vectors from multi-heads… the diverse designs of multi-heads can yield diverse feature vectors,” Page 7, Col. 1, Section 4, ¶1 “The triple-heads were designed as Res4_3, Res4_2, and Res4_1 of Last layer and 0.5× , 2× , and 4× of the default learning rate,” wherein a third head with a different block architecture and “learning rate” from the other two heads encompasses the third head comprising one or more other deep learning operations that compute a third detection tensor).
Ro teaches inputting the another intermediate tensor into the fourth head, the fourth head comprising one or more other deep learning operations that compute a fourth detection tensor that is different from the third detection tensor (Page 7, Col. 2, ¶1 “four-head structure,” one of ordinary skill in the art would recognize that inputting the another intermediate tensor into the fourth head is implicit when using a multi-head architecture that receives an output from an intermediate block “B2” of a ResNet-50 backbone, as explained above with respect to claim 1, Page 1, Fig. 1 depicts multiple heads of the neural network outputting diverse vectors after receiving an input, Page 4, Col. 1, Section C, ¶1 “The third design concept of the proposed multi-head structure is to produce diverse feature vectors from multi-heads… the diverse designs of multi-heads can yield diverse feature vectors,” wherein the fourth head, the fourth head comprising one or more other deep learning operations that compute a fourth detection tensor that is different from the third detection tensor is implicit when describing a “four-head structure” with “diverse designs” for each head to “yield” a “diverse feature vector” when compared to the outputs from the other three heads).
Regarding the limitation wherein the loss further comprises another diversity loss that indicates a measurement of similarity between the third detection tensor and the fourth detection tensor, Ro teaches the third detection tensor and the fourth detection tensor (Page 2, Col. 1, ¶2 “we consider… designing diverse heads to generate diverse feature vectors,” Page 7, Col. 2, ¶1 “the triple-head or four-head structure,” wherein the third detection and the fourth detection tensor are implicit when generating “diverse feature vectors” from “the triple-head or four-head structure”). However, Ro fails to teach wherein the loss further comprises another diversity loss that indicates a measurement of similarity between the third detection tensor and the fourth detection tensor.
Li teaches wherein the loss further comprises another diversity loss that indicates a measurement of similarity between outputs of a third and fourth head (Page 3, Col. 1, Section 3.1, ¶1 “To further guarantee the diversity, we enlarge the distances among multiple attention heads with disagreement regularization. To this end, we introduce an auxiliary regularization term in order to encourage the diversity among multiple attention heads,” Page 3, Col. 2, Equation (7) and ¶2 “Disagreement on Outputs. This disagreement directly applies regularization on the outputs of each attention head,” Page 5, Col. 1, Section 3.3, ¶1 “disagreement regularization focuses on adjusting the training objective, i.e. the loss function,” Page 5, Col. 2, ¶1 “the number of attention head is 8,” wherein the calculation of the “disagreement regularization” or another diversity loss that indicates a measurement of similarity is implied to occur between “the outputs of each attention head,” or between any of the 8 heads, which includes a third and fourth head and their respective outputs).
Ro and Li are analogous to the claimed invention as all are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the another diversity loss of Li with the loss and training methodology of Ro. The motivation to do so is to design a training method to “further improve… model performance” (Li, Abstract).
Regarding claim 3, Ro in view of Cates and further in view of Li teaches the method of claim 2 (and thus the rejection of claim 2 is incorporated).
Ro further teaches wherein the another intermediate tensor has a different size from the intermediate tensor (Page 3, Col. 1, ¶1 “we adopt the ResNet-50 network, most popular backbone model,” Col. 2, Section B, ¶2 “The layers of ResNet-50 network can be divided into five blocks as {conv1, Res1_x, Res2_x, Res3_x, Res4_x}. We have simply redefined it as {B1, B2, B3, B4} by combining conv1 into Res1_x,” Page 1, Fig. 1 depicts block “B2” producing an output to two heads; one of ordinary skill in the art would recognize that it is implicit that the output is the another intermediate tensor, as explained above with respect to claim 1, Page 5, Fig. 3 – (b) depicts block “B3” producing an output to two heads; one of ordinary skill in the art would recognize that it is implicit that the output is the intermediate tensor, as explained above with respect to claim 1; although Ro does not explicitly state wherein the another intermediate tensor has a different size from the intermediate tensor, one of ordinary skill in the art would recognize that different “blocks” in a “ResNet-50 network” such as “Res2_x, Res3_x,” which has been redefined as blocks “B2, B3,” output tensors of different sizes, hence wherein the another intermediate tensor has a different size from the intermediate tensor is implicit).
Regarding claim 7, Ro in view of Cates and further in view of Li teaches the method of claim 1 (and thus the rejection of claim 1 is incorporated).
Regarding the limitation wherein the measurement of similarity comprises a measurement of centered kernel alignment similarity between the first detection tensor and the second detection tensor or a measurement of cosine similarity between the first detection tensor and the second detection tensor, Ro teaches the first detection tensor and the second detection tensor (Page 1, Fig. 1 – f1-f2, Page 2, Col. 1, ¶2 “designing diverse heads to generate diverse feature vectors”). However, Ro fails to teach wherein the measurement of similarity comprises a measurement of centered kernel alignment similarity between the first detection tensor and the second detection tensor or a measurement of cosine similarity between the first detection tensor and the second detection tensor.
Li teaches wherein the measurement of similarity comprises a measurement of centered kernel alignment similarity between the first detection tensor and the second detection tensor or a measurement of cosine similarity between outputs of a first head and a second head (Page 3, Col. 2, Equation (7) and ¶2 “Disagreement on Outputs. This disagreement directly applies regularization on the outputs of each attention head, by maximizing the difference among them… we employ negative cosine similarity to measure the distance”).
Ro and Li are analogous to the claimed invention as all are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the measurement of cosine similarity of Li with the loss and training methodology of Ro. The motivation to do so is to design a training method to “further improve… model performance” (Li, Abstract).
Claims 4-6 are rejected under 35 U.S.C. 103 as being unpatentable over Ro in view of Cates and further in view of Li, and further in view of Lai et al. (“DCPNet: A Densely Connected Pyramid Network for Monocular Depth Estimation,” 2021, hereinafter Lai).
Regarding claim 4, Ro in view of Cates and further in view of Li teaches the method of claim 2 (and thus the rejection of claim 2 is incorporated).
Regarding the limitation inputting the first detection tensor into the third head, the third detection tensor computed based on the first detection tensor and the another intermediate tensor, Ro teaches the another intermediate tensor (Page 3, Col. 1, ¶1 “ResNet-50 network… backbone model,” Page 3, Col. 2, Section B, ¶2 “The layers of ResNet-50 network can be divided into… blocks… {B1, B2, B3, B4},” Page 1, Fig. 1 depicts multiple heads receiving an output from block “B2,” one of ordinary skill in the art would recognize that the output from “B2” is the another intermediate tensor, as explained above with respect to claim 1). However, Ro fails to teach inputting the first detection tensor into the third head, the third detection tensor computed based on the first detection tensor and the another intermediate tensor.
Lai, in the same field of endeavor, teaches inputting the first detection tensor into the third head, the third detection tensor computed based on the first detection tensor and another tensor (Page 5, Section 3.1, ¶1 “The encoder part is a traditional pre-trained backbone such as ResNet… The input color image is highly compressed as a very dense feature (i.e., 𝑆/32 feature block in the illustration) through the encoder part, which contains a large amount of deeply stacked convolution blocks. In the encoder process, the intermediate features with the size of 𝑆/2, 𝑆/4, 𝑆/8 and 𝑆/16 are preserved to be connected with the decoder part,” ¶2 “The proposed decoder is a pyramid-like structure which contains six floors (i.e., 𝐹1˜𝐹6 from top to bottom). Each floor contains six layers (i.e., 𝐿1˜𝐿6 from left to right),” one of ordinary skill in the art would recognize that a “floor” that produces “features” at the end of a “backbone” functions in substantially the same manner as a “head,” Page 6, ¶1 “The features generated… are preserved to be fused with features in lower floors,” ¶2 “a dense connection module (DCM) is used to fuse features on higher floors with features on the current floor,” Page 6, Fig. 2, Caption: “The resolution of the features on the last layer of every floor is S… the same as the input image,” one of ordinary skill in the art would recognize that “features on the last layer of every floor” generated after receiving an “input color image” encompass detection tensors; Figure 2 depicts inputting the first detection tensor “D1” from the first head “F1” into the third head “F3,” which computes the third detection tensor “D3” based on the first detection tensor “D1” and other tensors from previous layers in third head “F3”).
Regarding the limitation and inputting the second detection tensor into the fourth head, the fourth detection tensor computed based on the second detection tensor and the another intermediate tensor, Ro teaches the another intermediate tensor (Page 3, Col. 1, ¶1, Page 3, Col. 2, Section B, ¶2, Page 1, Fig. 1 as explained above). However, Ro fails to teach and inputting the second detection tensor into the fourth head, the fourth detection tensor computed based on the second detection tensor and the another intermediate tensor.
Lai teaches and inputting the second detection tensor into the fourth head, the fourth detection tensor computed based on the second detection tensor and another tensor (Page 5, Section 3.1, ¶1 “The encoder part is a… backbone such as ResNet… which contains a large amount of deeply stacked convolution blocks… the intermediate features… are preserved to be connected with the decoder part,” ¶2 “The… decoder is a pyramid-like structure which contains six floors… Each floor contains six layers,” Page 6, ¶1 “The features generated… are preserved to be fused with features in lower floors,” ¶2 “a dense connection module (DCM) is used to fuse features on higher floors with features on the current floor,” Page 6, Fig. 2, Caption: “The resolution of the features on the last layer of every floor is S… the same as the input image,” Figure 2 depicts and inputting the second detection tensor “D2” from the second head “F2” into the fourth head “F4,” which computes the fourth detection tensor “D4” based on the second detection tensor “D2” and other tensors from previous layers in fourth head “F4”).
Ro and Lai are analogous to the claimed invention as all are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the multi-head neural network architecture of Lai with the another intermediate tensor and neural network architecture of Ro. The motivation to do so is to offer “a more efficient way to fuse features from multiple scales” (Lai, Abstract).
Regarding claim 5, Ro in view of Cates and further in view of Li, and further in view of Lai teaches the method of claim 4 (and thus the rejection of claim 4 is incorporated).
Regarding the limitation wherein the one or more other operations comprise an upsampling operation that computes an upsampled tensor by increasing one or more dimensions of the first detection tensor, Ro teaches the first detection tensor (Page 1, Fig. 1 – “feature vector” f1). However, Ro fails to teach wherein the one or more other operations comprise an upsampling operation that computes an upsampled tensor by increasing one or more dimensions of the first detection tensor.
Lai teaches wherein the one or more other operations comprise an upsampling operation that computes an upsampled tensor by increasing one or more dimensions of a tensor (Page 5, Section 3.1, ¶2 “The upscale block is a sequence of a convolution operation and a nearest upsample operation… the latter enlarges the size of the feature… The upscale blocks enlarge the size of the features by a ratio of 2,” Page 6, Fig. 2 depicts a third head “F3” receiving a “feature” from the “Encoder backbone” that is passed through an “Upscale block,” wherein the one or more other operations comprises an upsampling operation that computes an upsampled tensor by increasing one or more dimensions of a tensor).
Ro and Lai are analogous to the claimed invention as all are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the upsampling operation of Lai with the first detection tensor of Ro. The motivation to do so is to offer “a more efficient way to fuse features from multiple scales” (Lai, Abstract).
Regarding claim 6, Ro in view of Cates and further in view of Li, and further in view of Lai teaches the method of claim 5 (and thus the rejection of claim 5 is incorporated).
Regarding the limitation wherein the one or more other operations further comprise a concatenation operation that computes a tensor by concatenating the upsampled tensor with the another intermediate tensor, Ro teaches the another intermediate tensor (Page 3, Col. 1, ¶1, Page 3, Col. 2, Section B, ¶2, Page 1, Fig. 1 as explained above with respect to claim 4). However, Ro fails to teach wherein the one or more other operations further comprise a concatenation operation that computes a tensor by concatenating the upsampled tensor with the another intermediate tensor.
Lai teaches wherein the one or more other operations further comprise a concatenation operation that computes a tensor by concatenating the upsampled tensor with another tensor (Page 7, Fig. 3, Caption: “the symbols ©… indicate concatenation”, Page 6, Section 3.2, ¶1 “Figure 3 shows the dense connection module… 𝑓𝑖,𝑗 denotes one of the features in floor i as well as layer j, which is obtained from an adjacent upscale block and will be sent to next dense connection module. In the DCM, feature 𝑓𝑖,𝑗 is concatenated with features obtained from DCMs at the same layer j from higher floors,” Page 5, Section 3.1, ¶2 “The upscale block… enlarges the size of the feature,” Page 6, Fig. 2 depicts third head “F3” with multiple “DCM” wherein the one or more other operations further comprise a concatenation operation, the upsampled tensor calculated after the first “Upscale block” in third head “F3” is concatenated with other tensors at each “DCM” in third head “F3” to compute a tensor).
Ro and Lai are analogous to the claimed invention as all are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the concatenation operation and upsampled tensor of Lai with the another intermediate tensor of Ro. The motivation to do so is to offer “a more efficient way to fuse features from multiple scales” (Lai, Abstract).
Claims 8 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Ro in view of Rosewarne (US 20240242481 A1, hereinafter Rosewarne) and further in view of Cates, and further in view of Li.
Regarding claim 8:
Regarding the limitation one or more non-transitory computer-readable media storing instructions executable to perform operations for training a neural network, the operations comprising: inputting a training dataset into the neural network, wherein the neural network comprises a backbone, a first head ensemble comprising a first head and a second head, and a second head ensemble comprising a third head and a fourth head, and wherein the first head has a same layer as the second head and a different layer from the third head, Ro teaches operations for training a neural network, the operations comprising: inputting a training dataset into the neural network, wherein the neural network comprises a backbone, a first head ensemble comprising a first head and a second head (Page 1, Col. 1, Section 1, ¶1, Page 2, Col. 2, Section 3, ¶1, Page 5, Col. 2, Section IV, ¶1, Page 3, Col. 1, ¶1, Page 7, Col. 2, Section D, ¶1, Page 5, Col. 2, Fig. 3 as explained above with respect to claim 1) and a third head and a fourth head (Page 7, Col. 2, ¶1 “triple-head or four-head structure”), and wherein the first head has… a different layer from the third head (Page 7, Col. 1, Section 4, ¶1, Page 5, Col. 1, Table 5 and Page 7, Table 8, Caption: “the last layer of the head”). However, Ro fails to teach one or more non-transitory computer-readable media storing instructions executable to perform operations for training a neural network, the operations comprising: inputting a training dataset into the neural network, wherein the neural network comprises… a second head ensemble comprising a third head and a fourth head, and wherein the first head has a same layer as the second head…
Rosewarne, in the same field of endeavor, teaches one or more non-transitory computer-readable media storing instructions executable to perform operations (Fig. 2A – 200, ¶81 “the software can also be loaded into the computer system 200 from other computer readable media. Computer readable storage media refers to any non-transitory tangible storage medium that provides recorded instructions and/or data to the computer system 200 for execution and/or processing”). However, Rosewarne fails to teach wherein the neural network comprises… a second head ensemble comprising a third head and a fourth head, and wherein the first head has a same layer as the second head…
Cates teaches and a second head ensemble comprising a third and a fourth head (Page 3, Col. 1, ¶2, Page 4, Col. 1, Section 3.4, ¶1, Pages 5-6, Col. 2, Section 5, ¶3 all as explained above with respect to claim 1), and wherein the first head has a same layer as the second head (Page 3, Col. 1, ¶¶2-3 as explained above with respect to claim 1).
Ro, Rosewarne, and Cates are analogous to the claimed invention as all are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the non-transitory computer-readable media of Rosewarne and the neural network architecture of Cates with the teachings of Ro. The motivation to do so is to design a multi-head architecture to “encourage the model to capture meaningful orientation information” (Cates, Page 9, Col. 1, ¶2) and to achieve high compression efficiency when encoding and decoding image signals (Rosewarne, ¶215 “The arrangements described are applicable to the computer and data processing industries and particularly for the digital signal processing for the encoding and decoding of signals such as video and image signals, achieving high compression efficiency”).
Claims 8-14 recite a non-transitory computer-readable media that parallels the method claims of 1-7, respectively. Therefore, the analysis discussed above with respect to claims 1-7 applies to claims 8-14, respectively. Accordingly, claims 8-14 are rejected based on substantially the same rationale as set forth above with respect to claims 1-7, respectively.
Regarding claim 15:
Ro teaches an apparatus, comprising: a computer processor for executing computer program instructions (Page 1, Col. 1, Section I, ¶1 “Deep metric learning has been successfully used in various applications related to computer vision, such as image retrieval, person re-identification, and face verification,” while Ro does not explicitly describe an apparatus, comprising: a computer processor for executing computer program instructions, one of ordinary skill in the art would recognize that an apparatus, comprising: a computer processor for executing computer program instructions, for example a CPU or GPU, is implicit when implementing “various applications related to computer vision”).
Regarding the limitation a non-transitory computer-readable memory storing computer program instructions executable by the computer processor to perform operations for training a neural network, Ro teaches operations for training a neural network (Page 1, Col. 1, Section 1, ¶1, Page 2, Col. 2, Section 3, ¶1, Page 5, Col. 2, Section IV, ¶1, Page 3, Col. 1, ¶1, Page 7, Col. 2, Section D, ¶1). However, Ro fails to teach a non-transitory computer-readable memory storing computer program instructions executable by the computer processor to perform operations…
Rosewarne, in the same field of endeavor, teaches a non-transitory computer-readable memory storing computer program instructions executable by the computer processor to perform operations (Fig. 2A – 200-201, 205, ¶75 “The computer module 201 typically includes at least one processor unit 205,” ¶81 “the software can also be loaded into the computer system 200 from other computer readable media. Computer readable storage media refers to any non-transitory tangible storage medium that provides recorded instructions and/or data to the computer system 200 for execution and/or processing,” ¶88 “the processor 205 is given a set of instructions which are executed therein”) …
Ro and Rosewarne are analogous to the claimed invention as both are from the same field of endeavor of multiheaded architecture. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the non-transitory computer-readable media of Rosewarne with the processor and teachings of Ro. The motivation to do so is to achieve high compression efficiency when encoding and decoding image signals (Rosewarne, ¶215 “The arrangements described are applicable to the computer and data processing industries and particularly for the digital signal processing for the encoding and decoding of signals such as video and image signals, achieving high compression efficiency”).
Claims 15-20 recite an apparatus that parallels the computer-readable medium claims of 8-9 and 11-14, respectively. Therefore, the analysis discussed above with respect to claims 8-9 and 11-14 applies to claims 15-20, respectively. Accordingly, claims 15-20 are rejected based on substantially the same rationale as set forth above with respect to claims 8-9 and 11-14, respectively.
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 WILLIAM M LEE whose telephone number is (571)272-4761. The examiner can normally be reached Mon-Fri. 8am-5pm.
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, Cesar Paula can be reached at (571)272-4128. 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.
/WILLIAM M LEE/
Examiner, Art Unit 2145
/CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights