DETAILED ACTION
This final action is in response to the amendment and remarks filed on 06/25/2026 for application 18/170,632.
Claims 1, 4, 8, and 11 have been amended.
Claims 1-14 remain pending in the application. Claims 1 and 8 are independent claims.
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 .
Claim Interpretation
As recited in MPEP § 2111, during patent examination, “the pending claims must
be given their broadest reasonable interpretation consistent with the specification”.
Under a broadest reasonable interpretation (BRI), claim terms must be given their plain
and ordinary meaning (i.e., the meaning that the term would have to a person of
ordinary skill in the art), unless applicant sets forth a special definition of a claim term
within the specification. The plain and ordinary meaning of a term “may be evidenced by
a variety of sources, including the words of the claims themselves, the specification,
drawings, and prior art”.
Independent claims 1 and 8 recite the limitation “identifying, by the task generation system, one or more secondary features by mapping the
first set of features with the feature set utilizing a reverse singular value decomposition (R-SVD)”.
Based on the known definition of singular value decomposition (SVD) in the art as a linear algebra technique, and the utilization of “reverse SVD” as described in the specification (see [¶ 0026]), which does not appear to define “reverse SVD” as a particular inverse algorithm or require a particular ordering, the examiner has interpreted “identifying…one or more secondary features by mapping the first set of features with the feature set utilizing a reverse singular value decomposition (R-SVD” to broadly encompass any technique which utilizes SVD-based decomposition/projection within a procedure of distinguishing secondary feature information from primary feature information.
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 1-14 are rejected under 35 U.S.C. 103 as being unpatentable over Li al., (“Multi-task learning with Attention: Constructing auxiliary tasks for learning to learn” , available conference 03 Nov 2021), hereinafter Li, in view of Thopalli et al. (“SALT : Subspace Alignment as an Auxiliary Learning Task for Domain Adaptation”, available arXiv 19 Dec 2019), hereinafter Thopalli.
Regarding claim 1, Li teaches A method for generating secondary tasks for neural networks (“With the development of deep learning in various fields, a deep learning method that can optimize multiple target tasks at the same time, that is, deep multi-task learning (MTL), has attracted more and more attention. We aim to find a multitask learning method to optimize a single goal. For image classification tasks, it is difficult to find multiple tasks with task relevance. We use an unsupervised clustering algorithm to construct multiple related auxiliary tasks in the dataset to solve this problem in order to achieve a kind of data enhancement. The purpose is to improve the accuracy of the main task. While these newly constructed auxiliary tasks may exhibit semantic features that are not relevant to the main task, in order to reduce the impact of such non-ideal auxiliary tasks on the main task, we use a multi-task learning based on learning to learn (MTL-LTL) approach with a spatially dependent attention function embedded in an underlying joint model with hard parameter sharing that allows our model to have pixelated modeling capabilities. In addition, we also use the method of learning to learn to randomly sample multiple auxiliary tasks. It can train these tasks on the shared hidden layer, and at the same time minimize the loss of the main task, and ensure that the optimization direction leads to the improvement of the main task.” [Li Abstract]), the method comprising:
receiving, by a task generation system, a feature set comprising one or more features of each of a plurality of data items, (see Fig. 1 including Input in (b) being images (i.e., data items) – “Fig. 1 Illustration of the training process of our proposed method,…(b) In the learning stage using MTL-LTL, our model samples a batch of images in each episode to update the task-specific decoder, and uses the shared model containing attention to learn the potential information of the task-specific. Our ultimate goal is to use the shared layer to minimize the loss of the main task” [Li page 148];) wherein the one or more features are generated for a primary task; (“1) MTL framework based on hard parameter sharing: Our work is aimed at the supervised classification task, which is a traditional STL-based task. We use the k-means unsupervised clustering method to construct multiple auxiliary tasks {Tt}T t=1 related to the main task on the unlabeled dataset Daux obtained from the original data, and use these auxiliary tasks to improve the generalization performance of MTL” [Li page 147 MTL with Attention]; see Fig. 1 including Intput [sic] in (a) being data (i.e., data items) {Xi} each having features NxN; “In order to learn the semantic meaning of the original features gathered in the space, we use an unsupervised learning method to generate a useful embedding space. Specifically, first run an unsupervised embedding learning algorithm E on Daux. E is a process that takes an unlabeled dataset Daux = {xi} as input, and then maps {Xi} to a low-latitude embedding space Z to generate {Zi}” [Li page 148 Construction of auxiliary tasks]; The MTL-LTL framework takes a set of images (i.e., data items) as input data, wherein the images are sampled for the purpose of supervised classification (i.e., generated for the primary task), and then obtains unlabeled dataset Daux = {Xi} (i.e. feature set) from the original data)
determining, by the task generation system, an association score between each of the one or more features of a data item from the plurality of data items with each of the one or more features of other data items from the plurality of data items; (“Similarly, for the tasks we need to construct, we can first construct different partitions Pn on Daux using the similar method described above…. we use the k-means clustering algorithm to treat the learned partition P = {Cl}L l=1 as a simplified Gaussian mixture model p(x | c)p(c)…E is a process that takes an unlabeled dataset Daux = {xi} as input, and then maps {Xi} to a low-latitude embedding space Z to generate {Zi}. In order to generate a different taskset, we generate T partitions {Pt}T t=1 by running the standard clustering algorithm kmeans, and apply random scaling to the dimensionality of the embedding space to induce different metrics. The clustering operation is summarized as:
PNG
media_image1.png
106
528
media_image1.png
Greyscale
. Equ. 6 learns a d × k central diagonal matrix U and the cluster assignment y∗ i of each vector Zi together, where y∗ l 1k = 1 and U represent the centroid of the learned cluster.” [Li page 148 Construction of auxiliary tasks]; The MTL-LTL framework utilizes a k-means clustering algorithm to partition feature set Daux into clusters – by definition, the k-means algorithm utilizes distance measure
PNG
media_image2.png
44
109
media_image2.png
Greyscale
to compare features of data item Zi to each cluster centroid, wherein features of each centroid are representative of the features of data items belonging to that cluster (i.e., other data items). The inverse of the distance measure can thereby be implicitly understood as an association score)
identifying, by the task generation system, a first set of features from the feature set, based on a comparison of the association score related to the one or more features of the plurality of data items with a threshold value, wherein the association score is greater than the threshold value for the first set of features; (“Equ. 6 learns a d × k central diagonal matrix U and the cluster assignment y∗ i of each vector Zi together, where y∗ l 1k = 1 and U represent the centroid of the learned cluster. We iteratively cluster the depth features to obtain a set of best distributions {y∗--- l}. Then, we use these y∗ i as pseudo-labels to allocate T partitions to construct T auxiliary tasks” [Li pages 148-149 Construction of auxiliary tasks]; As explained above, the inverse of distance measure
PNG
media_image2.png
44
109
media_image2.png
Greyscale
for the closest cluster can be implicitly understood as an association score, and the “threshold” value being implicitly understood as the inverse of distance to the second-nearest cluster. By definition, the k-means algorithm assigns a set of data items (and their associated features) from dataset {Zi} to a particular cluster (i.e., first set of features) based on identifying the centroid to which they have minimum distance (
PNG
media_image3.png
48
180
media_image3.png
Greyscale
))
identifying, by the task generation system, one or more secondary features by mapping the first set of features with the feature set; (“In our method, a similar classic MTL method is added. Specifically, the network is shared by using the bottom part of the joint model as the backbone, and then task-specific decoders are used for learning according to different tasks…The obtained useful features h = F(x) can improve the performance of its main task, and then assign a task-specific decoder to each task according to its uniqueness, and obtain the output ˆyt = Dt( h) of the task-specific.” [Li page 147 MTL with Attention]; “Based on the above unsupervised embedding algorithm E, we have obtained a taskset including the main task and multiple auxiliary tasks. For this taskset, we use the MTL-LTL framework with hard parameter sharing embedded with attention to learn the multi-task neural classification function ˆyt i = Dt (F (xi; θF) , θDt ), and use the learned prior knowledge and shared information of multiple tasks to learn the main task... For related auxiliary tasks, we further use an MTL method based on learning to learn, which can randomly sample a batch of tasks T from multiple auxiliary tasks and main tasks {Tt}T t=0 in the taskset. Then combine the data of multiple auxiliary tasks to train Dt {F (xi; θF) , θDt } to ensure that the main task is optimized in the correct direction. MTL-LTL with attention makes our model highly generalized and robust. Specifically, we use a hard parameter sharing MTL framework that combines spatially-dependent Attention. It will use the bottom layer of the combined model to let all tasks share the same hidden space, while retaining a specific decoder Dt for a task-specific. At the same time, the shared layer F on the auxiliary task is trained to generalize to the main task…After these two stages of learning, we can use the prior knowledge and initialization representation that are beneficial to the main task learned from each task-specific decoder Dt, thereby helping to improve the robustness and accuracy of the main task.” [Li page 149 MTL based on learning to learn]; see Fig. 1. including Shared Layers –> Auxiliary Task Decoder 1..3 in (b) [Li page 148]; Upon identifying T potential tasks through partitioned sets of features (including, e.g., a first set of features), the MTL-LTL framework can further learn, through the auxiliary task-specific decoders that are based on the constructed tasks (and partitioned sets of features therein), secondary features that contribute towards optimization of the main (i.e., primary) task)
generating, by the task generation system, one or more secondary tasks (104) based on the one or more secondary features, for a neural network; ([Li page 147 MTL with Attention] and [Li page 149 MTL based on learning to learn] and Fig. 1. including Shared Layers –> Auxiliary Task Decoder 1..3 in (b) [Li page 148]; MTL-LTL utilizes the constructed auxiliary task-specific decoders to thereby generate secondary training tasks for the neural framework based on them each learning secondary features that contribute towards optimization of the primary task) and
training the neural network based on at least the one or more secondary tasks, wherein training the neural network comprises training an untrained neural network to perform the primary task and the one or more secondary tasks, and wherein the one or more secondary tasks are distinct from the primary task (“We use the idea of model-agnostic meta-learning (MAML) [22], which is compatible with the MTL model learned through gradient descent, and uses knowledge in auxiliary tasks to improve the performance of the main task [24]. Specifically, the proposed method can jointly train these auxiliary tasks on the shared hidden layer, update the parameters of the shared layer with a method similar to transfer learning, and use a gradient descent strategy to minimize the loss of the main task to ensure the optimization direction leads to the improvement of the main task” [Li page 147 Learning to Learn]; “Based on the above unsupervised embedding algorithm E, we have obtained a taskset including the main task and multiple auxiliary tasks. For this taskset, we use the MTL framework with hard parameter sharing embedded with attention to learn the multi-task neural classification function ˆyt i = Dt (F (xi; θF) , θDt ), and use the learned prior knowledge and shared information of multiple tasks to learn the main task” [Li page 149 MTL based on learning to learn]; The MTL model is first trained on a taskset comprising multiple auxiliary tasks to learn prior knowledge (i.e., starts with an untrained model) for subsequent main task optimization)
However, Li does not expressly teach identifying one or more secondary features by mapping the first set of features with the feature set utilizing a reverse singular value decomposition (R-SVD).
In the same field of endeavor, Thopalli teaches a means of primary task optimization through auxiliary learning (“This paper represents a hybrid approach, where we assume simplified data geometry in the form of subspaces, and consider alignment as an auxiliary task to the primary task of maximizing performance on the source. The alignment is made rather simple by leveraging tractable data geometry in the form of subspaces. We synergistically allow certain parameters derived from the closed-form auxiliary solution, to be affected by gradients from the primary task. The proposed approach represents a unique fusion of geometric and model-based alignment with gradients from a data-driven primary task. Our approach termed SALT, is a simple framework that achieves comparable or sometimes outperforms state-of-the-art on multiple standard benchmarks” [Thopalli Abstract]) that identif[ies] one or more secondary features by mapping [a] first set of features with [a] feature set utilizing a reverse singular value decomposition (R-SVD) (“Let us denote the basis vectors for the d-dimensional subspaces inferred from source and target domains as {Zs} and {Zt} respectively and they satisfy ZTs Zs = I, ZTt Zt = I, where I denotes the identity matrix. The subspaces are inferred using singular value decomposition of source/target domain latent features {Xs, Xt}” [Thopalli page 4 Closed form subspace alignment]; see also equations 3-6 – feature subspace Zt is aligned/mapped with source feature subspace Zs to produce Zat, and Xt is re-projected through aligned subspace to obtain modified target features X^*t [Thopalli page 4])
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have incorporated identifying one or more secondary features by mapping [a] first set of features with [a] feature set utilizing a reverse singular value decomposition (R-SVD as taught by Thopalli into Li because they are both directed towards primary task optimization through auxiliary learning. Given that Li already seeks to construct auxiliary tasks from feature information related to the primary task, incorporating the teachings of Thopalli would further provide an auxiliary-learning subspace alignment mechanism whose interaction with the primary task (e.g., classification) would improve its performance (“A natural candidate is subspace alignment [5, 11, 42, 47], which utilizes simplified data representations, i.e., low-dimensional linear subspaces, and poses the problem of achieving domain invariance as learning a mapping between those representations… To this end, we develop SALT, an unsupervised domain adaptation algorithm based on simple subspace-based alignment, which is capable of producing highly effective classifiers through synergistic optimization between improving classifier performance and minimizing domain mismatch” [Thopalli pages 1-2 Introduction]).
Regarding claim 2, the combination of Li and Thopalli teaches the limitations of parent claim 1, and Li further teaches wherein generating the one or more secondary tasks comprises:
generating one or more secondary task groups of the one or more secondary features; (“Based on the above unsupervised embedding algorithm E, we have obtained a taskset including the main task and multiple auxiliary tasks. For this taskset, we use the MTL-LTL framework with hard parameter sharing embedded with attention to learn the multi-task neural classification function ˆyt i = Dt (F (xi; θF) , θDt ), and use the learned prior knowledge and shared information of multiple tasks to learn the main task... For related auxiliary tasks, we further use an MTL method based on learning to learn, which can randomly sample a batch of tasks T from multiple auxiliary tasks and main tasks {Tt}T t=0 in the taskset” [Li page 149 MTL based on learning to learn]; As explained above, a number of potential auxiliary tasks T are first identified through partitioned sets of features, from which a batch of tasks (i.e., secondary task group) is sampled for learning task-specific decoders (and associated secondary features))
labelling the one or more secondary task groups based on the feature set; (“We iteratively cluster the depth features to obtain a set of best distributions {y∗ l }. Then, we use these y∗ i as pseudo-labels to allocate T partitions to construct T auxiliary tasks” [Li page 149 Construction of auxiliary tasks]) and
generating the one or more secondary tasks corresponding to the one or more secondary task groups (“The multi-layer feedforward neural network proposed by [8] uses the hidden layer as the underlying model F shared among all tasks, and selects a specific task decoder Dt according to different tasks, and uses each decoder as a different output layer. The output of the output layer is regarded as the prediction result of the corresponding data sample. The model of task T is defined as follows:
PNG
media_image4.png
48
308
media_image4.png
Greyscale
Where θF and θDt are the parameters of the underlying shared layer F and the fixed task decoder Dt, respectively. Our model can find the optimal parameter
PNG
media_image5.png
47
256
media_image5.png
Greyscale
for the learning-based MTL based on these two parameters, such that
PNG
media_image6.png
85
435
media_image6.png
Greyscale
” [Li page 146 Multi-task learning]; Each task-specific decoder uses its respective pseudo-label y∗ l (of the overall set of pseudo-labels of the sampled task group) for optimizing model parameters θDt (i.e., obtaining secondary features and generating secondary training tasks))
Regarding claim 3, the combination of Li and Thopalli teaches the limitations of parent claim 1, and Li further teaches identifying a second set of features having the association score below the threshold value ([Li pages 148-149 Construction of auxiliary tasks] as detailed in claim 1 above; As explained above, the inverse of distance measure
PNG
media_image2.png
44
109
media_image2.png
Greyscale
for the closest cluster can be implicitly understood as an association score, and the “threshold” value being implicitly understood as the inverse of distance to the second-nearest cluster. By definition, the k-means algorithm assigns a set of data items (and their associated features) from dataset {Zi} to a particular cluster (i.e., first set of features) based on identifying the centroid to which they have minimum distance (
PNG
media_image3.png
48
180
media_image3.png
Greyscale
) – any other clusters thereby can be implicitly understood as separate from the first set of features (i.e., second set of features))
Regarding claim 4, the combination of Li and Thopalli teaches the limitations of parent claim 3, and Li further teaches selecting a first set of desired features from the second set of features, based on a predefined selection technique; ([Li pages 148-149 Construction of auxiliary tasks] as detailed in claim 1 above; Features of each centroid are representative of the features of data items belonging to that cluster (e.g., a second set of features) – features of centroids can thereby be implicitly understood as “ideal” features representative of that cluster)
selecting a second set of task groups comprising a second set of desired features, based on a similarity between the second set of desired features of each of the first set of task groups; ([Li pages 148-149 Construction of auxiliary tasks] as detailed in claim 1 above; Through the k-means algorithm, determining partitions of features from which tasks are constructed (i.e. task groups) involves comparing features of each data item Zi to features of each cluster centroid (i.e., comparing based on distance to ideal features for each task group (incl. e.g., first set of task groups)) to identify minimum distance and thereby the task groups (incl. e.g., second set of task groups) with closest similarity), and
generating the one or more secondary tasks corresponding to the second set of task groups with the second set of desired features ([Li page 147 MTL with Attention] and [Li page 149 MTL based on learning to learn] and Fig. 1. including Shared Layers –> Auxiliary Task Decoder 1..3 in (b) [Li page 148] as detailed in claim 1 above; Upon identifying T potential tasks through partitioned sets of features (including, e.g., a second set of task groups), the MTL-LTL framework can further learn, through the auxiliary task-specific decoders that are based on the constructed tasks (and partitioned sets of features therein), secondary features that contribute towards optimization of the main (i.e., primary) task)
Regarding claim 5, the combination of Li and Thopalli teaches the limitations of parent claim 1, and Li further teaches wherein the plurality of data items comprises one of, images, videos, audio inputs, text inputs, and speech inputs (see Fig. 1 including Input in (b) being images (i.e., data items) [Li page 148])
Regarding claim 6, the combination of Li and Thopalli teaches the limitations of parent claim 1, and Li further teaches wherein the feature set is received from one of, a trained neural network and an untrained neural network (see CIFAR-10 Dataset in Experiments – “CIFAR-10 [31] is a lightweight natural image dataset. The dataset includes 60,000 32×32 RGB color pictures, divided into 10 categories, each containing 5000 training images and 1000 test images. For this dataset, we use the two layers of the CNN architecture as the underlying shared layer of the joint model. The first convolutional layer has 64 filters with a size of 3×3, followed by a 2×2 maximum pooling layer. The second convolutional layer has 128 filters of size 3×3 and a maximum pooling layer of 2×2, as shown in [29]. At the same time, an activation function layer with a spatially dependent attention mechanism is embedded between each convolutional layer and the maximum pooling layer. Each task-specific decoder has two fully connected layers.” [Li pages 149-150 Experiments]; The images (from which the embedding algorithm obtains features) are received prior to training of the CNN architecture (i.e., untrained neural network))
Regarding claim 7, the combination of Li and Thopalli teaches the limitations of parent claim 1, and Li further teaches wherein the neural network is one of, a trained neural network and an untrained neural network ([Li pages 149-150 Experiments] as detailed in claim 6 above; As is typical of neural networks, the neural architecture is initially untrained, and then is trained over iterations).
Regarding claims 8-10 and 12-14, they are apparatus claims that largely correspond to the method of claims 1-3 and 5-7, which are already disclosed by Li as detailed above. Li further discloses A task generation system comprising: one or more processors; and a memory storing processor-executable instructions, which, on execution, cause the one or more processors to: perform the claimed functions ([Li pages 149-150 Experiments] as detailed in claim 6 above; Execution of the MTL-LTL framework on the disclosed datasets implicitly requires a computer with adequate processing capabilities). Consequently, claims 8-10 and 12-14 are rejected for the same reasons as claims 1-3 and 5-7.
Regarding claims 11, it is an apparatus claim that largely corresponds to the method of claim 4, which is already disclosed by Li as detailed above. Li further discloses identifying a first set of task groups of the first set of desired features, based on one or more inputs, wherein each task group from the first set of task groups comprises a plurality of ideal features from the set of desired features (Li pages 148-149 Construction of auxiliary tasks] as detailed in claim 1 above; Through the k-means algorithm, determining partitions of features from which tasks are constructed (i.e. task groups) involves comparing features of each data item Zi to features of each cluster centroid (i.e., comparing based on distance to ideal features for each task group (incl. e.g., first set of task groups and their associated ideal features)). Consequently, claim 11 is rejected for the same reasons as claim 4.
Response to Amendment and Arguments
The amendment filed 06/25/2026 has been entered.
Applicant’s amendment to the claims with respect to resolving objections and indefiniteness rejections under 35 U.S.C. 112(b) has been considered, and the previous objections and rejections are consequently withdrawn.
The remarks filed 06/25/2026 have been fully considered.
Applicant’s remarks traversing the non-eligible subject matter rejections under 35 U.S.C. 101 set forth in the office action mailed 03/18/2026, in view of claims 1-14 as amended, have been considered and are persuasive in part.
The examiner agrees that the amended claims at least recite a procedure/apparatus that conveys a technical improvement detailed in the specification (see page 8-10 of Remarks, from “Furthermore, claims 1 and 8…” through “…the functioning of a computer and computing technology.”), and therefore adequately integrate recited abstract ideas into a practical application.
Consequently, the rejections are withdrawn.
Applicant’s remarks traversing the anticipation rejections under 35 U.S.C. 102 set forth in the office action mailed 03/18/2026, in view of claims 1-14 as amended, have been considered, but are moot because the new grounds of rejection set forth above does not rely on the reference(s) applied in the prior rejection of record for the subject matter being challenged in applicant's argument.
Conclusion
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 VIJAY M BALAKRISHNAN whose telephone number is (571) 272-0455. The examiner can normally be reached 10am-5pm EST Mon-Thurs.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, JENNIFER WELCH can be reached on (571) 272-7212. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of 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.
/V.M.B./
Examiner, Art Unit 2143
/JENNIFER N WELCH/Supervisory Patent Examiner, Art Unit 2143
Prosecution Insights