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 .
Detailed Action
This action is in response to the claims filed 12/11/2023:
Claims 1 – 20 are pending.
Claims 1, 10, and 19 are independent.
Claim Objections
Claims 2, 11, 19, and 20 are objected to because of the following informalities:
Regarding claims 2, 11, and 19, "the next layer of the predetermined layer" should read "the next layer subsequent to the predetermined layer" or "the next layer after the predetermined layer." "Of the predetermined layer" suggests that the predetermined layer comprises the next layer.
Regarding claims 19 and 20, "instructions that, when executed by at least one processor, cause the processor configured to" should read "instructions that, when executed by at least one processor, cause the processor to".
Regarding claim 20, "periodically determine whether decrement or increment" should read "periodically determine whether to decrement or increment".
Appropriate correction is required.
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 2, 8, 9, 11, 17, and 18 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.
Regarding claims 2 and 11, "so that shortcut debiasing online continual learning is progressed in the target model" is indefinite. The claim limitation fails to clearly define the scope of the claim because it is unclear whether "shortcut debiasing" modifies "online continual learning," identifies a separate operation, or refers to a result achieved by the preceding step. Further, "is progressed" does not identify a clear action, state, metric, or structural/functional relationship by which one of ordinary skill could determine whether the limitation is met. In the interest of further examination, the next layer receiving a feature map with shortcut features removed is interpreted as satisfying the claim.
Regarding claims 8 and 17, "beneficial to prediction performance" is indefinite. A determination on whether or not something is beneficial is a relative subjective judgement which has no clear bounds in the claim. For example, "prediction performance of the target model" could be related to accuracy or processing speed which typically have an inverse relationship. Since the scope of the claim cannot reasonably be determined the claim is seen as being indefinite. In the interest of further examination any decrement or increment of a current drop intensity is interpreted as beneficial to prediction performance of the target model.
Regarding claims 9 and 18, "significantly better direction" is indefinite. One of ordinary skill in the art would not readily be able to determine the scope of "a significantly better direction". In the interest of further examination any update of the current drop intensity is interpreted as being in a significantly better direction.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-6, 10-15, and 19 are rejected under U.S.C. §103 as being unpatentable over the combination of Gao (“Cascade Attentive Dropout for Weakly Supervised Object Detection”, 2020) and Jung (“NEW INSIGHTS FOR THE STABILITY-PLASTICITY DILEMMA IN ONLINE CONTINUAL LEARNING”, 2023).
PNG
media_image1.png
330
1070
media_image1.png
Greyscale
FIG. 3 of Gao (CADM)
PNG
media_image2.png
200
1110
media_image2.png
Greyscale
FIG. 4 of Gao (GCM)
PNG
media_image3.png
348
1078
media_image3.png
Greyscale
FIG. 2 of Gao
Regarding claim 1, Gao teaches An operating method of an apparatus operated by at least one processor, the operating method comprising: ([p. 9] "Our experiments are implemented based on PyTorch deep learning framework and a NVIDIA GTX 1080Ti GPU")
fusing at least some feature maps generated by layers of a target model [performing online continual learning] to generate a fused feature map; ([p. 6] "Perform feature fusion to obtain enhanced feature maps")
identifying features with high attention in the fused feature map as shortcut features ([p. 4] "The Attention mechanism is inspired by the human vision which does not treat all data equally but enhances or weakens them" [p. 5] "Weakly supervised object detectors tend to learn only the most discriminative features" [p. 4] "The proposed CADM, which is designed to eliminate negative effects of discriminative features, is employed on pooling 3 feature map. Different from ADL[21] which erased the maximally activated spatial parts, we purposely discard attentive elements in both channel and space dimension." [p. 6] "Perform feature fusion to obtain enhanced feature maps" Gao's "most discriminative features" / "attentive elements" are interpreted as "features with high attention in the fused feature map as shortcut features". Gao explicitly drops out features of the fused feature map.)
in a ratio based on a drop intensity; and ([p. 5] "We first get the self-attention map Xa ∈RN× 1× H× W via a channel-wise average pooling layer. Since the activation value of more discriminative areas in the attention map is higher, we set a threshold λ2 to erase these areas to force the network to learn less discriminative features for classification but meaningful features for object localization, thereby avoiding location part domination. For the self-attention map Xa, the maximum value of ith row is recorded as gimax. When the element gij in row i and column j of the attention map is greater than the corresponding drop threshold gimax ·λ2, the element is dropped; otherwise, the element is retained" [p. 6] "During the training process, the network stochastically chooses either of the drop mask or importance map according to drop rate, and the selected one is merged into the input feature map Xcd to gain the spatial-dropped feature" Gao’s drop rate is a parameterized ratio that controls the extent or frequency of the feature removal interpreted as a ratio based on a drop intensity. Gao's threshold control which channel/spatial attention elements are dropped, and the drop rate controls whether the network applies the drop mask during training.)
removing the shortcut features from a target feature map output from a predetermined layer of the target model and input to a next layer. ([p. 5] "Channel-Dropout. Given a feature map X3 ∈ RN×D×H×W extracted from CNN, the channel-dropout module takes it as input and outputs a global information embedding via a global average pooling (GAP) layer [...] We first get the self-attention map Xa ∈RN× 1× H× W via a channel-wise average pooling layer. Since the activation value of more discriminative areas in the attention map is higher, we set a threshold λ2 to erase these areas to force the network to learn less discriminative features" [p. 4] "CADM, which is designed to eliminate negative effects of discriminative features, is employed on pooling 3 feature map [...] we purposely discard attentive elements in both channel and space dimension" Gao identifies the predetermined intermediate feature map: the pooling3 / X3 feature map. CADM is immediately after pooling3. The FIG. 2 architecture then routes the CADM processed feature map onward into VGG16 Conv4-5, then GCM, ROI pooling, and MIL refinement. That ties the removal step to a target feature map output from a predetermined layer and input to downstream layers. See also FIG. 3).
However, Gao does not explicitly teach fusing at least some feature maps generated by layers of a target model performing online continual learning to generate a fused feature map;.
Jung, in the same field of endeavor, teaches fusing at least some feature maps generated by layers of a target model performing online continual learning to generate a fused feature map; ([p. 1] "we propose an online continual learning framework named multi-scale feature adaptation network (MuFAN) that utilizes a richer context encoding extracted from different levels of a pre-trained network" [p. 3] "MuFAN obtains an aggregated multi-scale feature map using a top-down module and replaces every BN in m with the proposed SPN").
Gao as well as Jung are directed towards feature fusion. Therefore, Gao as well as Jung are reasonably pertinent analogous art. It would have been obvious before the effective filing date of the claimed invention to combine the teachings of Gao with the teachings of Jung by applying multi-layer feature fusion to online continual learning. Jung provides as additional motivation for combination ([p. 9] “we propose a novel online CL framework named MuFAN, which enhances both stability and plasticity while being less impediment to one another. This is the first CL method that leverages multi-scale feature maps, which are constructed by projecting raw images into meaningful spaces, as the input for the classifier in order to improve plasticity”). This motivation for combination also applies to the remaining claims which depend on this combination.
Regarding claim 2, the combination of Gao and Jung teaches The operating method of claim 1, wherein the next layer of the predetermined layer receives a feature map with the shortcut features removed as input, so that shortcut debiasing online continual learning is progressed in the target model.(Gao [p. 5] "Channel-Dropout. Given a feature map X3 ∈ RN×D×H×W extracted from CNN, the channel-dropout module takes it as input and outputs a global information embedding via a global average pooling (GAP) layer [...] We first get the self-attention map Xa ∈RN× 1× H× W via a channel-wise average pooling layer. Since the activation value of more discriminative areas in the attention map is higher, we set a threshold λ2 to erase these areas to force the network to learn less discriminative features" [p. 4] "CADM, which is designed to eliminate negative effects of discriminative features, is employed on pooling 3 feature map [...] we purposely discard attentive elements in both channel and space dimension" Gao identifies the predetermined intermediate feature map: the pooling3 / X3 feature map. CADM is immediately after pooling3. The FIG. 2 architecture then routes the CADM processed feature map onward into VGG16 Conv4-5, then GCM, ROI pooling, and MIL refinement. That ties the removal step to a target feature map output from a predetermined layer and input to downstream layers. See also FIG. 3).
Regarding claim 3, the combination of Gao and Jung teaches The operating method of claim 1, wherein the removing the shortcut features includes removing the shortcut features from the target feature map by applying a drop mask (Gao [p. 5] "a binary channel-dropout mask mcd ∈RN× D× 1× 1 is generated to indicate whether each channel is dropped or not, as shown in formula 1")
that masks regions of the shortcut features in the target feature map with zero.(Gao [p. 5] "a binary channel-dropout mask mcd ∈RN× D× 1× 1 is generated to indicate whether each channel is dropped or not, as shown in formula 1 […] where mi cd equal to 0 means the i-th channel is dropped. The binary drop mask is then multiplied to input map X3 to get the channel-dropped feature map").
Regarding claim 4, the combination of Gao and Jung teaches The operating method of claim 1, wherein the generating the fused feature map includes fusing a first feature map including structural information with a second feature map including semantic information.(Jung [p. 4] "low-level features extracted by shallow layers encode more pattern-wise and general information, whereas the high-level features extracted by deeper layers contain more contextual information that focuses on prominent features. That is, the aggregated multi-scale feature map from shallow to deeper layers of the pre-trained encoder can compensate for the shortcomings of each layer" Pattern-wise/general interpreted as synonymous with structural. Contextual/prominent interpreted as synonymous with semantic.).
Regarding claim 5, the combination of Gao and Jung teaches The operating method of claim 4, wherein the generating the fused feature map includes adjusting the first feature map and the second feature map to have the same resolution and then fusing the first feature map and the second feature map.(Jung [p. 17] "For the top-down module, to match the resolution and the number of channels of shallow layers, the feature from deeper layers enlarges a resolution by bilinear upsampling and reduces the number of channels by a random 3 × 3 convolution. For the bottom-up module, oppositely, the feature from shallow layers shrinks a resolution by max-pooling and increases the number of channels by a random 3 × 3 convolution [...] the resized feature is added to another feature from a different scale element-wisely, as illustrated in Figure 3").
Regarding claim 6, the combination of Gao and Jung teaches The operating method of claim 1, wherein the identifying the shortcut features includes: performing pooling on the fused feature map along channel dimension to generate an attention map;(Gao [p. 2] "A global context module (GCM), which uses sigmoid to enhance nonlinearity and perform feature fusion through element-wise multiplication and additions" [p. 4] "Generate comprehensive feature maps by VGG16 with CADM and GCM. (2)Generate fixed-size ROI features. (3)Feed the proposal feature vectors to MIL and Refinement submodule to predict categories and locations" See FIG. 2 where output enhanced feature map of CADM is upscaled and fused through GCM and then immediately pooled)
identifying a certain percentage of features with high attention scores in the attention map as the shortcut features based on the drop intensity; and(Gao [p. 5] "we set a threshold λ2 to erase these areas" [p. 6] "When λ2 decreases, more element values will be discarded." [p. 6] "the network stochastically chooses either of the drop mask or importance map according to drop rate")
generating a drop mask that masks regions of the shortcut features in the attention map with zero.(Gao [p. 5] "a binary channel-dropout mask mcd ∈RN× D× 1× 1 is generated to indicate whether each channel is dropped or not, as shown in formula 1 […] where mi cd equal to 0 means the i-th channel is dropped. The binary drop mask is then multiplied to input map X3 to get the channel-dropped feature map").
Regarding claim 10, claim 10 is directed towards an apparatus for performing the method of claim 1. Therefore, the rejection applied to claim 1 also applies to claim 10. Claim 10 recites additional elements An apparatus for online continual learning, the apparatus comprising: a memory; and a processor executing instructions stored in the memory, wherein the processor is configured, by executing the instructions, to (Gao [p. 9] "Our experiments are implemented based on PyTorch deep learning framework and a NVIDIA GTX 1080Ti GPU").
Similarly, regarding claims 11-15, claims 11-15 are directed towards an apparatus for performing the methods of claims 2-5, respectively. Therefore, the rejections applied to claims 2-5 also apply to claims 11-15.
Regarding claim 19, Gao teaches A computer program stored in a non-transitory computer-readable storage medium, the computer program comprising instructions that, when executed by at least one processor, cause the processor configured to:([p. 9] "Our experiments are implemented based on PyTorch deep learning framework and a NVIDIA GTX 1080Ti GPU")
fuse at least some feature maps generated by layers of a target model [performing online continual learning] to generate a fused feature map;([p. 6] "Perform feature fusion to obtain enhanced feature maps")
perform pooling on the fused feature map along a channel to generate an attention map;([p. 2] "A global context module (GCM), which uses sigmoid to enhance nonlinearity and perform feature fusion through element-wise multiplication and additions" [p. 4] "Generate comprehensive feature maps by VGG16 with CADM and GCM. (2)Generate fixed-size ROI features. (3)Feed the proposal feature vectors to MIL and Refinement submodule to predict categories and locations" See FIG. 2 where output enhanced feature map of CADM is upscaled and fused through GCM and then immediately pooled. Gao generates attention/drop masks by channel-wise average pooling and then fuses the resulting feature representation and then follows it up immediately with additional ROI pooling)
identify a certain percentage of features with high attention scores in the attention map as the shortcut features based on the drop intensity;([p. 5] "we set a threshold λ2 to erase these areas" [p. 6] "When λ2 decreases, more element values will be discarded." [p. 6] "the network stochastically chooses either of the drop mask or importance map according to drop rate")
generate a drop mask that masks regions of the shortcut features in the attention map with zero; and([p. 5] "a binary channel-dropout mask mcd ∈RN× D× 1× 1 is generated to indicate whether each channel is dropped or not, as shown in formula 1")
apply the drop mask to a target feature map output from a predetermined layer of the target model, ([p. 5] "a binary channel-dropout mask mcd ∈RN× D× 1× 1 is generated to indicate whether each channel is dropped or not, as shown in formula 1")
and input a new target feature map with shortcut features removed into a next layer of the predetermined layer.([p. 5] "Channel-Dropout. Given a feature map X3 ∈ RN×D×H×W extracted from CNN, the channel-dropout module takes it as input and outputs a global information embedding via a global average pooling (GAP) layer [...] We first get the self-attention map Xa ∈RN× 1× H× W via a channel-wise average pooling layer. Since the activation value of more discriminative areas in the attention map is higher, we set a threshold λ2 to erase these areas to force the network to learn less discriminative features" [p. 4] "CADM, which is designed to eliminate negative effects of discriminative features, is employed on pooling 3 feature map [...] we purposely discard attentive elements in both channel and space dimension" Gao identifies the predetermined intermediate feature map: the pooling3 / X3 feature map. CADM is immediately after pooling3. The FIG. 2 architecture then routes the CADM processed feature map onward into VGG16 Conv4-5, then GCM, ROI pooling, and MIL refinement. That ties the removal step to a target feature map output from a predetermined layer and input to downstream layers. See also FIG. 3).
However, Gao does not explicitly teach fuse at least some feature maps generated by layers of a target model performing online continual learning to generate a fused feature map;.
Jung, in the same field of endeavor, teaches fuse at least some feature maps generated by layers of a target model performing online continual learning to generate a fused feature map;([p. 1] "we propose an online continual learning framework named multi-scale feature adaptation network (MuFAN) that utilizes a richer context encoding extracted from different levels of a pre-trained network" [p. 3] "MuFAN obtains an aggregated multi-scale feature map using a top-down module and replaces every BN in m with the proposed SPN").
Gao as well as Jung are directed towards feature fusion. Therefore, Gao as well as Jung are reasonably pertinent analogous art. It would have been obvious before the effective filing date of the claimed invention to combine the teachings of Gao with the teachings of Jung by applying multi-layer feature fusion to online continual learning. Jung provides as additional motivation for combination ([p. 9] “we propose a novel online CL framework named MuFAN, which enhances both stability and plasticity while being less impediment to one another. This is the first CL method that leverages multi-scale feature maps, which are constructed by projecting raw images into meaningful spaces, as the input for the classifier in order to improve plasticity”). This motivation for combination also applies to the remaining claims which depend on this combination.
Claims 7, 8, 9, 16, 17, 18, and 20 are rejected under U.S.C. §103 as being unpatentable over the combination of Gao and Jung and in further view of Goel (US20160307098A1).
Regarding claim 7, the combination of Gao and Jung teaches The operating method of claim 1.
However, the combination of Gao and Jung doesn't explicitly teach further comprising adaptively shifting the drop intensity.
Goel, in the same field of endeavor, teaches The operating method of claim 1, further comprising adaptively shifting the drop intensity. ([¶0024] "The model (or ensemble of models) may be subsequently trained for additional iterations with the dropout rate (e.g., inferred dropout rate) held fixed, or the dropout range (inferred dropout range) may be dynamically changed").
The combination of Gao and Jung as well as Goel are directed towards neural network dropout. Therefore, the combination of Gao and Jung as well as Goel are reasonably pertinent analogous art. It would have been obvious before the effective filing date of the claimed invention to combine the teachings of the combination of Gao and Jung with the teachings of Goel by making the dropout rate adaptive. Goel provides as additional motivation for combination that the annealed dropout is effective ([Abstract] “to improve a generalization performance of the network”).
Regarding claim 8, the combination of Gao, Jung, and Goel teaches The operating method of claim 7, wherein the adaptively shifting the drop intensity includes periodically determining whether decrement or increment of a current drop intensity is beneficial to prediction performance of the target model, and increasing or decreasing a next drop intensity from the current drop intensity, or maintaining the current drop intensity. (Goel [¶0058] "the adjusting of the dropout rate may include applying a linear, or geometric, fixed decaying schedule to change the probability. This may include inferring (e.g., exactly or approximately), an optimal annealing schedule 311 (or an optimal joint annealing schedule for an ensemble of models) for the learning rate, the dropout probability pd, and any other such so-called hyperparameters of a learning procedure using the annealing schedule generator 310. Determining an annealing schedule 311 (e.g., optimized annealing schedule) according to the present principles may remove the uncertainty of random dropout training" [¶0060] "the generating of the optimal annealing schedule 311 (or optimal joint annealing schedule for an ensemble of models) for all hyperparameters (e.g., dropout rate, learning rate, etc.) by the annealing schedule generator 310 may include considering all or a subset of the set of combinations for an ensemble of models by holding fixed, increasing, or decreasing each hyperparameter by a specified amount, and selecting a subset (e.g., the N best performing models, N>=1) of models that result (e.g., based on one or more iterations of learning), for the application of additional training iterations").
Regarding claim 9, the combination of Gao, Jung, and Goel teaches The operating method of claim 8, wherein the adaptively shifting the drop intensity includes comparing loss reduction according to a reduced drop intensity and loss reduction according to an increased drop intensity while alternately using the reduced drop intensity and the increased drop intensity for a certain number of iterations, and updating the current drop intensity in a significantly better direction of the loss reduction. (Goel [¶0069] "Annealed dropout may be viewed as a ‘noisy’ training procedure, which can greatly increase the realized capacity of the learned model, (e.g., by mitigating against the convergence to poor local optima). By mitigating against the convergence to poor, locally optimal solutions, the ‘noise’ in the training procedure may converge to be ‘deep’ but ‘narrow’ local minima of a loss function, and may therefore be less likely to be present than in conventional systems and methods, and network performance may be improved. Annealing the dropout rate may result in a very noisy procedure initially, but may result in a less noisy procedure after each iteration of annealed dropout training for fine tuning the network optimization").
Regarding claims 16-18, claims 16-18 are directed towards an apparatus for performing the methods of claims 7-9, respectively. Therefore, the rejections applied to claims 7-9 also apply to claims 16-18.
Regarding claim 20, the combination of Gao and Jung teaches The computer program of claim 19.
However, the combination of Gao and Jung doesn't explicitly teach further comprising instructions to cause the processor configured to periodically determine whether decrement or increment of a current drop intensity is beneficial to prediction performance of the target model, and increase or decrease a next drop intensity from the current drop intensity, or maintain the current drop intensity.
Goel, in the same field of endeavor, teaches to cause the processor configured to periodically determine whether decrement or increment of a current drop intensity is beneficial to prediction performance of the target model, and increase or decrease a next drop intensity from the current drop intensity, or maintain the current drop intensity. ([¶0058] "the adjusting of the dropout rate may include applying a linear, or geometric, fixed decaying schedule to change the probability. This may include inferring (e.g., exactly or approximately), an optimal annealing schedule 311 (or an optimal joint annealing schedule for an ensemble of models) for the learning rate, the dropout probability pd, and any other such so-called hyperparameters of a learning procedure using the annealing schedule generator 310. Determining an annealing schedule 311 (e.g., optimized annealing schedule) according to the present principles may remove the uncertainty of random dropout training" [¶0060] "the generating of the optimal annealing schedule 311 (or optimal joint annealing schedule for an ensemble of models) for all hyperparameters (e.g., dropout rate, learning rate, etc.) by the annealing schedule generator 310 may include considering all or a subset of the set of combinations for an ensemble of models by holding fixed, increasing, or decreasing each hyperparameter by a specified amount, and selecting a subset (e.g., the N best performing models, N>=1) of models that result (e.g., based on one or more iterations of learning), for the application of additional training iterations").
The combination of Gao and Jung as well as Goel are directed towards neural network dropout. Therefore, the combination of Gao and Jung as well as Goel are reasonably pertinent analogous art. It would have been obvious before the effective filing date of the claimed invention to combine the teachings of the combination of Gao and Jung with the teachings of Goel by making the dropout rate adaptive. Goel provides as additional motivation for combination that the annealed dropout is effective ([Abstract] “to improve a generalization performance of the network”).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Zhang (“Online Residual-Based Key Frame Sampling with Self-Coach Mechanism and Adaptive Multi-Level Feature Fusion”, 2023) is directed towards continuous online learning with adaptive multi-level feature fusion.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SIDNEY VINCENT BOSTWICK whose telephone number is (571)272-4720. The examiner can normally be reached M-F 7:30am-5:00pm 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, Miranda Huang can be reached on (571)270-7092. 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.
/SIDNEY VINCENT BOSTWICK/Examiner, Art Unit 2124
Prosecution Insights