DETAILED ACTION
This office action is in response to amendments filed on 02/24/2026.
Claims 1, 8, and 15 have been amended. Claims 3 and 16 have been canceled. Claims 21-22 have been added. Claims 1-2, 4-15, and 17-22 are pending.
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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 03/16/2026 has been entered.
Response to Arguments
35 U.S.C. 112(b) Rejection:
In light of applicant’s amendments to the claims (pg. 2-10), the rejection of claims 8-14 under 35 USC § 112(b) has been withdrawn.
Prior Art Rejections:
Applicant's arguments regarding the prior art rejections (pg. 11-12) have been fully considered but they are not persuasive.
Applicant argues that the cited references do not disclose "flipping the tuples to generate flipped tuples, and processing the flipped tuples, with a linear fit model, to predict unnormalized weights for the output data," as recited in amended independent claim 1, which incorporates the subject matter of previously-presented dependent claim 3. Applicant specifically argues that Khanfor fails to teach “flipping the tuples”, as the softmax function taught by Khanfor does not involve flipping tuples as recited in the claim. Khanfor teaches generating a vector representation for each node in a graph, thereby generating a node-representation tuple. The representation is then transformed using a softmax function (Khanfor, pg. 3, section II). Examiner respectfully notes that flipping tuples is not a term or concept with a known definition in the art, nor is it further defined in the specification, and thus transforming the node-representation tuple via a softmax function, as taught by Khanfor, falls within the broadest reasonable interpretation of flipping a tuple.
The prior art rejections have been updated to include the amended limitations and to clarify the reasoning given for the limitations that were not amended.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over
Khanfor et al. (hereinafter Khanfor), “Graph Neural Networks-based Clustering for Social Internet of Things” in view of
Singla et al. (hereinafter Singla), “IoT2Vec: Identification of Similar IoT Devices via
Activity Footprints”,
Sugiyama et al. (hereinafter Sugiyama), “Direct importance estimation for covariate shift adaptation”, and
Kapoor et al. (hereinafter Kapoor), U.S. Patent Application Publication US 20130173621 A1.
Regarding Claim 1,
Khanfor teaches A method, comprising:
receiving, by a device, device data identifying activation and usage of different types of Internet of things (IoT) devices; (Pg. 2, section II: “In Fig. 2, we present the different steps of our proposed GNN-based clustering framework for social IoT systems.” Figure 2 (pg. 2) shows the IoT device data which is input to the framework, including device features such as ‘type’, ‘brand’, ‘model’, ‘mobility’, and ‘battery’, as well as device locations, and connections based on relations such as ‘co-location’, ‘social friendship and ownership’, and ‘social object’. Pg. 2, section II.A(1): “Social object relation (SOR): The SOR relation is created when two devices collaborate in a continuous or sporadic form.” Social object relation data, indicating collaboration between devices, corresponds to activation and usage data of the devices.)
identifying, by the device, a device activation sequence based on the device data, (Pg. 2, section II: “a pre-processing step is executed to create multiple weighted graphs of social relations connecting the devices.” A weighted graph of collaboration data between devices amounts to a device activation sequence.)
training, by the device, a model with the device activation sequence; (Pg. 3, section II.B: “…we tend to use the GNN model as an embedder where we extract the feature representations of IoT devices in a forward pass using the propagation rule. We then label few nodes of our data and we train the model in a semi-supervised way in order to learn better representations.” The GNN model is trained by inputting the graph (i.e. the device activation sequence).)
generating, by the device, output data based on training the model with the device activation sequence; (Pg. 3, section II.C: “…a vector representation for each node in a social relation graph is determined using GNN…” The graph (i.e. the device activation sequence) is fed into the trained GNN model to generated a vector representation of each node (i.e. output data).)
creating tuples based on the output data; (Pg. 3, section II.C: “…a vector representation for each node in a social relation graph is determined using GNN…” Each node and its corresponding vector representation from the second message passing hidden layer represent a tuple of input-output data.)
flipping the tuples to generate flipped tuples; (Pg. 3, section II.B: “The two message passing layers are followed by a fully-connected layer which makes use of the softmax function to produce a probability distribution over the class labels.” The softmax function transforms or ‘flips’ the vector representation of each node, generating a flipped tuple of input-output data.)
generating, by the device, device representations of the different types of IoT devices using the retrained model; (Pg. 3, section II.C: “Once a vector representation for each node in a social relation graph is determined using GNN…” Each node of the graph represents an IoT device, and each node is passed through the GNN model to generate a vector representation of the device.)
determining, by the device, groups of similar IoT devices based on the device representations of the different types of IoT devices; (Pg. 3, section II.C: “…an unsupervised machine learning technique can be utilized to group the IoT devices with common features and attributes into clusters or communities.” Clustering is performed on the vector representations of the devices to determine groups of similar devices.)
Khanfor does not appear to explicitly disclose wherein the device activation sequence is based on respective periods of time when the IoT devices are powered on and powered off;
However, Singla teaches wherein the device activation sequence is based on respective periods of time when the IoT devices are powered on and powered off; (Pg. 2, section IV.E: “Using the previously mentioned postulates, we define some steps to generate and analyze the word embeddings from IoT device sensor logs. Our method includes the following steps: 1. Filter out the IoT sensors whose data is not meaningful or we cannot make sense of the data. 2. Examine the activity data of the selected sensors to see whether it shows meaningful activity or actions. 3. Extract only the values where the sensor state is in transition (e.g. ON to OFF or OFF to ON). 4. Build a session of the sensor values (similar to sentence in NLP domain) by choosing a session gap. Session gap is the gap of time where we construct the boundaries of each session.” A session (i.e. device activation sequence) is determined based on IoT sensor state values (e.g. transitions between ON and OFF) (i.e. when the devices are powered on and off) within a time period.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Khanfor and Singla. Khanfor teaches a method for clustering similar Internet-of-Things devices based on latent representations generated by a graph neural network. Singla teaches generating embeddings for IoT devices based on sequences of sensor state values. One of ordinary skill would have motivation to combine Khanfor and Singla because “[o]ur approach has the advantage of not needing any prior information or assumptions about the IoT devices. So it can be used in a wider variety of potential use cases… One obvious application can be to make a search function to search for similar devices in the vicinity” (Singla, pg. 2, section II-III).
Khanfor and Singla do not appear to explicitly disclose
calculating, by the device, weights for the output data with a weight detection model,
wherein calculating the weights comprises:
processing the flipped tuples, with a linear fit model, to predict unnormalized weights for the output data; and
normalizing the unnormalized weights to calculate the weights for the output data;
calculating, by the device, a loss function for the output data, wherein the loss function provides a number measuring how the model fits the output data;
retraining, by the device, the model based on the weights and the loss function to generate a retrained model;
However, Sugiyama teaches A method, comprising:
calculating, by the device, weights for the output data with a weight detection model, (Pg. 702, section 2.1: “The goal of this paper is to develop a method of estimating the importance w(x) from
x
i
t
r
i
=
1
n
t
r
and
x
j
t
e
j
=
1
n
t
e
… Let us model the importance w(x) by the following linear model:
w
^
x
=
∑
l
=
1
b
α
l
φ
l
(
x
)
, where
α
l
l
=
1
b
are parameters to be learned from data samples…” Importance
w
^
x
is the weight given to input sample x and its corresponding output y by the importance model (i.e. weight detection model).)
wherein calculating the weights comprises:
processing the flipped tuples, with a linear fit model, to predict unnormalized weights for the output data; and (Pg. 702, section 2.1: “Let us model the importance w(x) by the following linear model:
w
^
x
=
∑
l
=
1
b
α
l
φ
l
(
x
)
, where
α
l
l
=
1
b
are parameters to be learned from data samples…” Importance
w
^
x
is the weight given to the tuple comprising input x and its corresponding output y (i.e. flipped tuple) by the linear importance model (i.e. linear fit model).)
normalizing the unnormalized weights to calculate the weights for the output data. (Pg. 703, section 2.2: “In addition to the non-negativity,
w
^
x
should be properly normalized…”)
calculating, by the device, a loss function for the output data, wherein the loss function provides a number measuring how the model fits the output data; (Pg. 717-718, section 4.3: “The parameter vector θ is learned by importance-weighted least-squares (IWLS):
θ
^
I
W
L
S
:
=
argmin
θ
∑
i
=
1
n
t
r
w
^
(
x
i
t
r
)
(
f
^
x
i
t
r
;
θ
-
y
i
t
r
)
2
.” The loss function for IWLS given here includes the term
f
^
x
i
t
r
;
θ
-
y
i
t
r
, which represents the difference between the model output
f
^
x
i
t
r
;
θ
and expected output
y
i
t
r
for each example
i
(i.e. a measure of how the model fits the output data). Examiner notes that a loss function provides a scalar measuring how well the model fits the output data.)
retraining, by the device, the model based on the weights and the loss function to generate a retrained model; (Pg. 717-718, section 4.3: “The parameter vector θ is learned by importance-weighted least-squares (IWLS):
θ
^
I
W
L
S
:
=
argmin
θ
∑
i
=
1
n
t
r
w
^
(
x
i
t
r
)
(
f
^
x
i
t
r
;
θ
-
y
i
t
r
)
2
.” The parameters of the model, represented by parameter vector θ, are iteratively trained (i.e. retrained) by the IWLS loss function, which is based on the importance weights.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Khanfor, Singla, and Sugiyama. Khanfor teaches a method for clustering similar Internet-of-Things devices based on latent representations generated by a graph neural network. Singla teaches generating embeddings for IoT devices based on sequences of sensor state values. Sugiyama teaches a method for improving the accuracy of a machine learning model by importance weighting of training data. One of ordinary skill would have motivation to combine Khanfor, Singla, and Sugiyama because in Khanfor, “We then label few nodes of our data and we train the model in a semi-supervised way” (Khanfor, pg. 3, section II.B), and these few nodes may not comprehensively represent the distribution of IoT devices. According to Sugiyama, “A situation where training and test samples follow different input distributions is called covariate shift” (Sugiyama, pg. 699, abstract), and importance weighting “contributes to improving the prediction performance in covariate shift scenarios” (Sugiyama, pg. 702, section 1).
Khanfor, Singla, and Sugiyama do not appear to explicitly disclose generating, by the device, configuration data for each of the groups of the similar IoT devices; and causing, by the device, the different types of IoT devices to be configured based on the configuration data.
However, Kapoor teaches generating, by the device, configuration data for each of the groups of the similar IoT devices; and causing, by the device, the different types of IoT devices to be configured based on the configuration data. ([0015]: “Clustering similar Internet connected devices within a particular cluster group can enable a larger system to better manage a collection of heterogeneous devices by, for example, imposing usage rules and policies on the set of similar devices that are included in a cluster group, providing access control and security restrictions to the set of similar devices that are included in a cluster group, performing device configuration operations on the set of similar devices that are included in a cluster group, and so on.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Khanfor, Singla, Sugiyama, and Kapoor. Khanfor teaches a method for clustering similar Internet-of-Things devices based on latent representations generated by a graph neural network. Singla teaches generating embeddings for IoT devices based on sequences of sensor state values. Sugiyama teaches a method for improving the accuracy of a machine learning model by importance weighting of training data. Kapoor teaches a method for clustering similar Internet-of-Things devices based on device attributes to facilitate efficient device management. One of ordinary skill would have motivation to combine Khanfor, Singla, Sugiyama, and Kapoor because Kapoor’s device cluster configuration “can enable a larger system to better manage a collection of heterogeneous devices” (Kapoor, para. 0015).
Regarding Claim 2, Khanfor, Singla, Sugiyama, and Kapoor teach The method of claim 1, as shown above.
Khanfor also teaches further comprising:
preprocessing the device activation sequence prior to training the model based on the device activation sequence. (Pg. 2-3, section II.B: “Given the weight matrix A of a social relation graph of n nodes, we first normalize it to obtain the matrix Ã… This normalization addresses numerical instabilities which may lead to exploding/vanishing gradients when used in a deep neural network model.” The social relation graph is a device activation sequence, and normalizing its weight matrix before input into the model is a preprocessing step.)
Regarding Claim 4, Khanfor, Singla, Sugiyama, and Kapoor teach The method of claim 1, as shown above.
Khanfor also teaches further comprising:
preparing the model with an input layer of one and an output layer based on a window size, prior to training the model; (Pg. 3, section II.B: “Zt is the learned embeddings of the graph in the tth layer. As an initialization, Z0 = X where X ∈ Rn×d is a matrix whose jth row contains the feature vector Xj of the IoT node j.” Z0 represents the input layer of the model, taking a single input: the graph. “…we tend to only make use of the nodes hidden representations Z2 ∈ Rn×h2 that are produced from the second message passing layer.” The output of the second hidden message passing layer is the output layer, with a window size of 32 units as shown below.)
defining hidden internal layers of the model prior to training the model; and (pg. 3, section II.B: “Our GNN model consists of two message passing hidden layers, where the first hidden layer has h1 = 64 units and the second hidden layer has h2 = 32 units…”)
generating the model based on the input layer, the output layer, and the hidden internal layers. (Pg. 3, section II.B: “We train the model over 100 epochs.” Training the model with the described structure amounts to generating the model.)
Regarding Claim 5, Khanfor, Singla, Sugiyama, and Kapoor teach The method of claim 1, as shown above.
Khanfor also teaches wherein the model is a neural network model. (Pg. 3, section II.B: “…we tend to use the GNN [Graph Neural Network] model as an embedder where we extract the feature representations of IoT devices in a forward pass using the propagation rule.” A graph neural network (GNN) is a type of neural network model.)
Regarding Claim 6, Khanfor, Singla, Sugiyama, and Kapoor teach The method of claim 1, as shown above.
Sugiyama also teaches wherein the loss function provides a measure of how the model fits the output data. (Pg. 717-718, section 4.3: “The parameter vector θ is learned by importance-weighted least-squares (IWLS):
θ
^
I
W
L
S
:
=
argmin
θ
∑
i
=
1
n
t
r
w
^
(
x
i
t
r
)
(
f
^
x
i
t
r
;
θ
-
y
i
t
r
)
2
.” The loss function for IWLS given here includes the term
f
^
x
i
t
r
;
θ
-
y
i
t
r
, which represents the difference between the model’s predicted output and the actual output data.)
Regarding Claim 7, Khanfor, Singla, Sugiyama, and Kapoor teach The method of claim 1, as shown above.
Kapoor also teaches further comprising: providing the retrained model to another device for implementation. ([0023]: “The operating system (154) and device clustering module (304) in the example of FIG. 1 are shown in RAM (168), but many components of such software typically are stored in non-volatile memory also, such as, for example, on a disk drive (170).” [0065]: “The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified…” The device clustering module (i.e. trained model) can be stored on a disk drive and loaded onto another computer for implementation.)
Claim 8 is a system claim containing substantially the same elements as method claim 1. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 1, as shown above.
Kapoor also teaches A device, comprising: one or more processors ([0014]: “FIG. 1 sets forth a block diagram of automated computing machinery comprising an example computer (152) useful in clustering devices in an IoT according to embodiments of the present invention. The computer (152) of FIG. 1 includes at least one computer processor…”)
Claim 9 is a system claim containing substantially the same elements as method claim 2. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 2, as shown above.
Regarding Claim 10, Khanfor, Singla, Sugiyama, and Kapoor teach The device of claim 8, as shown above.
Kapoor also teaches wherein the one or more processors are further configured to one or more of: provide the retrained model to another device for implementation; or store the retrained model. ([0023]: “The operating system (154) and device clustering module (304) in the example of FIG. 1 are shown in RAM (168), but many components of such software typically are stored in non-volatile memory also, such as, for example, on a disk drive (170).” [0065]: “The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified…” The device clustering module (i.e. trained model) can be stored on a disk drive and loaded onto another computer for implementation.)
Regarding Claim 11, Khanfor, Singla, Sugiyama, and Kapoor teach The device of claim 8, as shown above.
Khanfor also teaches wherein the one or more processors are further configured to:
analyze the device representations of the different types of IoT devices to generate analysis results; and modify the retrained model based on the analysis results. (Pg. 3, section II.B: “The two message passing layers are followed by a fully-connected layer which makes use of the softmax function to produce a probability distribution over the class labels…. We then label few nodes of our data and we train the model in a semi-supervised way in order to learn better representations. In fact, the model is trained as a classifier… We train the model over 100 epochs.” In an epoch of training, the device representations are passed through the softmax function to generate predictions which are compared to labeled data (i.e. analyzed), and the model is updated to improve performance based on the results. Since the model is trained over 100 epochs, in later epochs, the model will already have been trained and retrained when it is modified by the current epoch.)
Regarding Claim 12, Khanfor, Singla, Sugiyama, and Kapoor teach The device of claim 8, as shown above.
Khanfor also teaches wherein the one or more processors are further configured to:
receive additional device data identifying activation and usage of other IoT devices; (Pg. 3, section III: “To examine the proposed framework, we use a data set of real-world IoT devices from a smart city in Santander, Spain, provided by Marche et al. [13]. The data set includes different types of private and public devices. We select 1000, 1500, and 2000 private devices out of 16216 devices to analyze the possibility of the framework for scalability and applicability in different sizes of the IoT system.” This testing data is additional device data for other IoT devices.)
identify another device activation sequence based on the additional device data; and (Pg. 3, section III: “Following that, the links between the devices are established based on the various social relations, namely CLOR, SFOR, and SOR, described in Section II-A1.” Establishing links between devices based on social relations (creating the graph) amounts to identifying a device activation sequence.)
process the other device activation sequence, with the retrained model, to generate additional device representations of the other IoT devices. (Pg. 3-4, section III: “In Fig. 3, we illustrate the clustering results of applying algorithms K-means, DBSCAN, and Louvain for the evaluation metrics, modularity, and coverage… We also notice that both K-means and DBSCAN clustering of the GNN embeddings outperform Louvain community detection…” The GNN embeddings are the testing data device representations generated by processing the testing data graph (i.e. the device activation sequence) with the model.
Regarding Claim 13, Khanfor, Singla, Sugiyama, and Kapoor teach The device of claim 12, as shown above.
Khanfor also teaches wherein the one or more processors are further configured to:
determine additional groups of similar IoT devices based on the additional device representations; (Pg. 3-4, section III: “In Fig. 3, we illustrate the clustering results of applying algorithms K-means, DBSCAN, and Louvain for the evaluation metrics, modularity, and coverage… With DBSCAN, we notice that some nodes are detected as outliers. Therefore, we assign to each of those nodes a new cluster… We also notice that both K-means and DBSCAN clustering of the GNN embeddings outperform Louvain community detection…” The testing data is clustered (grouped) by GNN embeddings (device representations), including the creation of additional groups when necessary.)
Kapoor teaches wherein the one or more processors are further configured to:
generate additional configuration data for each of the additional groups; and cause the other IoT devices to be configured based on the additional configuration data. ([0015]: “Clustering similar Internet connected devices within a particular cluster group can enable a larger system to better manage a collection of heterogeneous devices by, for example, imposing usage rules and policies on the set of similar devices that are included in a cluster group, providing access control and security restrictions to the set of similar devices that are included in a cluster group, performing device configuration operations on the set of similar devices that are included in a cluster group, and so on.”)
Regarding Claim 14, Khanfor, Singla, Sugiyama, and Kapoor teach The device of claim 8, as shown above.
Khanfor also teaches wherein the model includes an input layer to receive the device activation sequence, a hidden layer to process the device activation sequence and generate the output data, and an output layer to store the output data. (Pg. 3, section II.B: “Our GNN model consists of two message passing hidden layers, where the first hidden layer has h1 = 64 units and the second hidden layer has h2 = 32 units… Zt is the learned embeddings of the graph in the tth layer. As an initialization, Z0 = X where X ∈ Rn×d is a matrix whose jth row contains the feature vector Xj of the IoT node j.” Z0 represents the input layer of the model, which receives the graph (i.e. the device activation sequence) “…we tend to only make use of the nodes hidden representations Z2 ∈ Rn×h2 that are produced from the second message passing layer.” Z2 represents the output of the second hidden message passing layer, which is used as the output of the model to be fed into the clustering algorithm, and thus must be stored.)
Claim 15 is a product claim containing substantially the same elements as method claim 1. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 1, as shown above.
Kapoor also teaches A non-transitory computer-readable medium storing a set of instructions, ([0058]: “Furthermore, aspects of the present invention may take the form of a computer program product embodied in one or more computer readable medium(s) having computer readable program code embodied thereon.”)
Khanfor also teaches wherein the model is a neural network model; (Pg. 3, section II.B: “…we tend to use the GNN [Graph Neural Network] model as an embedder where we extract the feature representations of IoT devices in a forward pass using the propagation rule.” A graph neural network (GNN) is a type of neural network model.)
Claim 17 is a product claim containing substantially the same elements as method claim 4. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 4, as shown above.
Claim 18 is a product claim containing substantially the same elements as method claim 7. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 7, as shown above.
Claim 19 is a product claim containing substantially the same elements as method claim 2. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 2, as shown above.
Claim 20 is a product claim containing substantially the same elements as system claim 11. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 11, as shown above.
Claim 21 is a product claim containing substantially the same elements as system claim 14. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 14, as shown above.
Claim 22 is a product claim containing substantially the same elements as system claim 12. Khanfor, Singla, Sugiyama, and Kapoor teach the elements of claim 12, as shown above.
Conclusion
Claims 1-2, 4-15, and 17-22 are rejected.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BENJAMIN M ROHD whose telephone number is (571)272-6445. The examiner can normally be reached Mon-Thurs 8:00-6:00 EST.
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, Viker Lamardo can be reached at (571) 270-5871. 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.
/B.M.R./Examiner, Art Unit 2147 /VIKER A LAMARDO/Supervisory Patent Examiner, Art Unit 2147
Prosecution Insights