DETAILED ACTION
This action is in response to the application filed on 11/08/2023. Claims 1-20 are pending in the application and have been examined.
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 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.
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 1-4, 10-13, 19 are rejected under 35 U.S.C. 103 as being unpatentable by Wang et al. (“When, Where, and How Does It Fail? A Spatial-Temporal Visual Analytics Approach for Interpretable Object Detection in Autonomous Driving” [2022], hereinafter “Wang”) in view of Bharadhwaj (“Layer-wise Relevance Propagation for Explainable Recommendations” [2018]).
Regarding Claim 1,
Wang discloses A method of visualizing neurons in an Artificial Intelligence (AI) model for autonomous driving, comprising: obtaining, from a number of neurons of the AI model for a task, one or more neurons (Wang [Section 2.3]; “Various visual analytics systems have been developed to support the understanding and improvement of machine learning models, especially those based on neural networks, which are often seen as a black box. Zeiler and Fergus [52] proposed a method to show the feature map of a model to help users better understand the role of each layer in a CNN and the importance of features in the original data through visualization. This method has been widely used in CNN interpretability research, such as in Simonyan et al. [46], which used saliency maps”
Wang [Section 5.1.1];
PNG
media_image1.png
236
341
media_image1.png
Greyscale
Wang [Figure 1.a];
PNG
media_image2.png
222
114
media_image2.png
Greyscale
wherein the global discriminative density map comprises pooled feature maps; wherein feature maps being visualizations of spatially arranged neuron activation values thus reads on a method of visualizing neurons in an artificial intelligence model; wherein the generation of such a density map comprising feature maps thus reads on obtaining some number of neurons of the AI model for a feature processing task
Wang [Page 5034 Paragraph 2];
PNG
media_image3.png
289
346
media_image3.png
Greyscale
wherein such visualization systems are being developed for the context of autonomous driving)
determining, for each of the one or more neurons, a respective Region of Interest (ROI) of an input related to the task, wherein the respective ROI is encoded by the one or more neurons for the task; (Wang [Figure 1.a];
PNG
media_image4.png
630
325
media_image4.png
Greyscale
Wang [Page 5039 Paragraph 1];
PNG
media_image5.png
257
372
media_image5.png
Greyscale
PNG
media_image6.png
168
360
media_image6.png
Greyscale
wherein the determined density maps obtained through average pooling of a plurality of feature maps highlighting regions of interest associated with high semantic information thus reads on, for each of the plurality of neurons, determining a ROI of the input wherein the ROI is encoded by the neurons (since the density maps comprising feature maps thus reads on the ROI being encoded by neurons))
and producing a human-interpretable representation of the determined respective ROI of the input for at least a portion of the one or more neurons (Wang [Section IV Subsection B2];
PNG
media_image7.png
426
319
media_image7.png
Greyscale
wherein the comparison of the density map with the temporal information and bounding boxes providing developers answers as to how the model detects objects in addition to the object projection visualizing the processed features of each detection thus reads on a produced human-interpretable representation of the determined respective ROIs of the inputs for the neurons)
Wang fails to explicitly disclose but Bharadhwaj discloses by applying a first operation including Layer-wise Relevance Propagation (LRP) (Bharadhwaj [Section 2.3];
PNG
media_image8.png
226
335
media_image8.png
Greyscale
PNG
media_image9.png
372
336
media_image9.png
Greyscale
wherein select input variables comprised of select neurons are interpreted as regions of interest (evaluated variables as informative) associated with the one or more neurons; wherein the layer-wise relevance propagation is performed to generate an outputted heatmap evaluation of important variables and their associated neurons in the neural network)
It would have been obvious to modify Wang’s method of visualizing neurons in an AI model through human-interpretable global discriminative density maps by instead computing human-interpretable heatmaps signifying importance through Bharadhwaj’s layer-wise relevance propagation. One would have been motivated to do so because one could “incorporate layer-wise relevance propagation in this system so that features of the recommended images which were most central to the recommendation task are identified” (Bharadhwaj [Section 1 Paragraph 3]) as opposed to Wang’s density maps which only identify general neuron regions of importance instead of neurons of specific features.
Regarding Claim 2,
Wang/Bharadhwaj teaches the method of Claim 1 (and thus the rejection of Claim 1 is incorporated). Wang/Bharadhwaj already discloses wherein the input is collected by a sensor, a recorded human driving database, and/or a cloud storage for the task (Wang [Section 4.4];
PNG
media_image10.png
295
321
media_image10.png
Greyscale
wherein the input data including tracklets, GPS, point clouds, and images thus read on the input collected by at least a recorded human driving database (images captured by the vehicle’s camera))
Regarding Claim 3,
Wang/Bharadhwaj teaches the method of Claim 1 (and thus the rejection of Claim 1 is incorporated). Wang/Bharadhwaj already discloses wherein the input is a processed image frame, and the respective Region of Interest (ROI) for each of the one or more neurons includes a set of pixels of the processed image frame that corresponds to the respective ROI (Wang [Section 4.4];
PNG
media_image10.png
295
321
media_image10.png
Greyscale
wherein the input data including tracklets, GPS, point clouds, and images thus read on the input collected by at least a recorded human driving database (images captured by the vehicle’s camera), thus the input is a processed image frame
Wang [Figure 1.a];
PNG
media_image4.png
630
325
media_image4.png
Greyscale
Wang [Section IV Subsection B2];
PNG
media_image7.png
426
319
media_image7.png
Greyscale
wherein the density maps’ regions rich in semantic information include a set of lighter colored pixels in greyscale space, thus the respective regions of interest associated with the plurality of neurons encoded in the feature maps making up the density maps include a set of pixels correspondent to the respective regions of interest)
Regarding Claim 4,
Wang/Bharadhwaj teaches the method of Claim 1 (and thus the rejection of Claim 1 is incorporated). Wang/Bharadhwaj already discloses wherein the input is a sequence of processed image frames, and the respective Region of Interest (ROI) for each of the one or more neurons includes a union of sets of pixels of the sequence of processed image frames that correspond to a respective sub-region of interest (sub-ROI) of each processed image frame in the sequence of processed image frames, respectively (Wang [Section 4.4];
PNG
media_image10.png
295
321
media_image10.png
Greyscale
wherein the input data including tracklets, GPS, point clouds, and images thus read on the input collected by at least a recorded human driving database (images captured by the vehicle’s camera), therefore the input a processed image frame
Wang [Figure 4];
PNG
media_image11.png
193
318
media_image11.png
Greyscale
wherein the input is a sequence of processed images
Wang [Page 5039 Paragraph 1];
PNG
media_image12.png
228
328
media_image12.png
Greyscale
PNG
media_image12.png
228
328
media_image12.png
Greyscale
wherein the identified bounding boxes extracted by the output layer neurons of the network reads on a sub-region of interest present in processed image frames
Wang [Page 5040 Paragraph 2];
PNG
media_image13.png
84
325
media_image13.png
Greyscale
wherein the comparison of the density map with the bounding boxes of detection results reads on the union of sets of pixels associated with the density map and the bounding boxes (discovering common regions of interest identified by both the bounding boxes as well as the density map; wherein the bounding boxes are identifiable as a sub-ROI of the larger density maps due to the bounding boxes being derived merely from processing the conv layers and output layer alone, but the density map is derived by further extracting final feature maps of the conv layers and therefore reads as a more advanced ROI compared to the sub-ROI bounding boxes))
Claims 10-13 recite a non-transitory computer-readable medium configured to execute the exact method of Claims 1-4 respectively. Thus, Claims 10-13 are rejected for reasons set forth in the rejection of Claims 1-4 respectively.
Claim 19 recites a computer-implemented system comprising processors and memory devices storing instructions to cause the processors to execute operations comprising the exact method of Claim 1 respectively. Thus, Claim 19 is rejected for reasons set forth in the rejection of Claim 1 respectively.
Claims 5-7, 14-16, 20 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (“When, Where, and How Does It Fail? A Spatial-Temporal Visual Analytics Approach for Interpretable Object Detection in Autonomous Driving” [2022], hereinafter “Wang”) in view of Bharadhwaj (“Layer-wise Relevance Propagation for Explainable Recommendations” [2018]) in view of Smith et al. (US 20240005063 A1, hereinafter “Smith”).
Regarding Claim 5,
Wang/Bharadhwaj teaches the method of Claim 1 (and thus the rejection of Claim 1 is incorporated). Wang/Bharadhwaj fails to explicitly disclose but Smith discloses wherein the producing of the human-interpretable representation includes applying a second operation including Visual Back-Propagation (VBP) to identify a specific ROI from the determined respective ROI that contributed most to a prediction made by the Artificial Intelligence (AI) model to complete the task in order to improve computational efficiency and enhance model accuracy (Smith [0035]; “A generated model is used to propagate output data of the finite element model back through an inverted neural network to relevant inputs. The propagations may implement a back propagation or a layer-wise relevance propagation.”
Smith [0043]; “After the gradient graph 150 has been generated and presented to a person, the person may submit a query that identifies an area of interest 174 in the gradient graph 150. One or more propagations 146 backwards through the neural network 130 are performed for the gradient values within the area of interest 174. The propagations 146 start from the output nodes and work back to the input nodes. In various embodiments, the propagations 146 may implement layer-wise relevance propagations 146a. Data for a relevance graph 180 is generated in response to the layer-wise relevance propagations 146a. The relevance graph 180, optionally displayed, correlates the input nodes that influence the gradient values 152 in the area of interest 174 within the gradient graph 150. The relevance graph 180 includes multiple relevance values 182a-182e of the input parameters. The relevance values 182a-182e range from most relevant values 182a to least relevant values 182e. While five degrees of the relevance values 182a-182e are illustrated, fewer or more degrees of relevance values may be implemented to meet a design criteria of a particular application.” wherein the production of human-interpretable explanations of neurons of the network comprises the application of both visual back propagations to identify relevant values as well as standard layer-wise relevance propagation; wherein the backward propagations identifying ranges of relevant values within the identified area of interest thus reads on visual back-propagation to identify a specific ROI (ranges of relevant values) that contributed most to a prediction made by the Artificial Intelligence model (identifying values within the range as “relevant” or “not relevant”)
Smith [0063]; “In the step 256, the processor 120 reduces the gradient values 152 within the area of interest 174 by automatically adjusting one or more of the input parameters 116 in the subset 118.” wherein the reduction of gradient values within the area of interest thus reads on improved computational efficiency (since reducing the magnitude of gradients to be processed leads to reduced amount of data necessitating storage and processing, thus improving computational efficiency)
Smith [0040]; “Thereafter, the input parameters are automatically or manually adjusted to improve a fidelity of the model 94 to the structure 92 being represented.” wherein the input parameters adjusted dependent on the identified relevance of values within the region of interest in order to improve the fidelity of the model to the structure being represented is interpreted as enhanced model accuracy of the model in response to the visual back propagation)
It would have been obvious to modify Wang/Bharadhwaj’s method of applying layer-wise relevance propagation to obtain feature maps by applying Smith’s visual back-propagation in addition to the layer-wise propagation. One would have been motivated to do so in order “to improve a fidelity of the model 94 to the structure 92 being represented” (Smith [0040]).
Regarding Claim 6,
Wang/Bharadhwaj/Smith teaches the method of Claim 5 (and thus the rejection of Claim 5 is incorporated). Wang/Bharadhwaj/Smith already discloses wherein the Visual Back-Propagation (VBP) is applied sequentially after the Layer-wise Relevance Propagation (LRP), and the LRP is applied to identify the one or more neurons from the number of neurons that contributed most to a prediction made by the Artificial Intelligence (AI) model to complete the task in order to improve computational efficiency and enhance model accuracy (Smith [0035]; “A generated model is used to propagate output data of the finite element model back through an inverted neural network to relevant inputs. The propagations may implement a back propagation or a layer-wise relevance propagation.”
Smith [0043]; “After the gradient graph 150 has been generated and presented to a person, the person may submit a query that identifies an area of interest 174 in the gradient graph 150. One or more propagations 146 backwards through the neural network 130 are performed for the gradient values within the area of interest 174. The propagations 146 start from the output nodes and work back to the input nodes. In various embodiments, the propagations 146 may implement layer-wise relevance propagations 146a. Data for a relevance graph 180 is generated in response to the layer-wise relevance propagations 146a. The relevance graph 180, optionally displayed, correlates the input nodes that influence the gradient values 152 in the area of interest 174 within the gradient graph 150. The relevance graph 180 includes multiple relevance values 182a-182e of the input parameters. The relevance values 182a-182e range from most relevant values 182a to least relevant values 182e. While five degrees of the relevance values 182a-182e are illustrated, fewer or more degrees of relevance values may be implemented to meet a design criteria of a particular application.” wherein the production of human-interpretable explanations of neurons of the network comprises the application of both visual back propagations is thus interpreted as both types of propagation being applied sequentially (since sequentially merely reads on one being performable before the other, and the interchangeability between back and layer-wise relevance propagation thus reads on a sequential application)))
Regarding Claim 7,
Wang/Bharadhwaj/Smith teaches the method of Claim 6 (and thus the rejection of Claim 6 is incorporated). Wang/Bharadhwaj/Smith already discloses wherein the Artificial Intelligence (AI) model includes a mixing block and a model backbone (Wang [Page 5039 Paragraph 1];
PNG
media_image14.png
232
324
media_image14.png
Greyscale
wherein the existence of the extracted feature maps output by the processing convolutional and output layers of the machine learning model thus imply the existence of some mixing block and model backbone (convolutional layers reads on parameter sharing between weights, thus a sort of “mixing block”; a model backbone is implied since a model backbone extracts features from input data through processing raw inputs across multiple layers to produce feature maps, and the existence of feature maps thus implies a model backbone))
Claims 14-16 recite a non-transitory computer-readable medium configured to execute the exact method of Claims 5-7 respectively. Thus, Claims 14-16 are rejected for reasons set forth in the rejection of Claims 5-7 respectively.
Claim 20 recites a computer-implemented system comprising processors and memory devices storing instructions to cause the processors to execute operations comprising the exact method of Claim 5 respectively. Thus, Claim 20 is rejected for reasons set forth in the rejection of Claim 5 respectively.
Claims 8, 17 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (“When, Where, and How Does It Fail? A Spatial-Temporal Visual Analytics Approach for Interpretable Object Detection in Autonomous Driving” [2022], hereinafter “Wang”) in view of Bharadhwaj (“Layer-wise Relevance Propagation for Explainable Recommendations” [2018]) in view of Smith et al. (US 20240005063 A1, hereinafter “Smith”) in view of Chen et al. (CN112766279A, hereinafter “Chen”)
Regarding Claim 8,
Wang/Bharadhwaj/Smith teaches the method of Claim 7 (and thus the rejection of Claim 7 is incorporated). Wang/Bharadhwaj/Smith already discloses wherein the applying of the first operation and the second operation includes: applying the Layer-wise Relevance Propagation (LRP) through the mixing block to obtain … feature maps of the one or more neurons (Smith [0035]; “A generated model is used to propagate output data of the finite element model back through an inverted neural network to relevant inputs. The propagations may implement a back propagation or a layer-wise relevance propagation.”
Smith [0043]; “After the gradient graph 150 has been generated and presented to a person, the person may submit a query that identifies an area of interest 174 in the gradient graph 150. One or more propagations 146 backwards through the neural network 130 are performed for the gradient values within the area of interest 174. The propagations 146 start from the output nodes and work back to the input nodes. In various embodiments, the propagations 146 may implement layer-wise relevance propagations 146a. Data for a relevance graph 180 is generated in response to the layer-wise relevance propagations 146a. The relevance graph 180, optionally displayed, correlates the input nodes that influence the gradient values 152 in the area of interest 174 within the gradient graph 150. The relevance graph 180 includes multiple relevance values 182a-182e of the input parameters. The relevance values 182a-182e range from most relevant values 182a to least relevant values 182e. While five degrees of the relevance values 182a-182e are illustrated, fewer or more degrees of relevance values may be implemented to meet a design criteria of a particular application.” wherein the production of human-interpretable explanations of neurons of the network comprises the application of both visual back propagations is thus interpreted as both types of propagation being applied sequentially (since sequentially merely reads on one being performable before the other, and the interchangeability between back and layer-wise relevance propagation thus reads on a sequential application))).
and applying the Visual Back-Propagation (VBP) to back-propagate the weighted feature maps of the one or more neurons through the model backbone (Smith [0035]; “A generated model is used to propagate output data of the finite element model back through an inverted neural network to relevant inputs. The propagations may implement a back propagation or a layer-wise relevance propagation.”
Smith [0043]; “After the gradient graph 150 has been generated and presented to a person, the person may submit a query that identifies an area of interest 174 in the gradient graph 150. One or more propagations 146 backwards through the neural network 130 are performed for the gradient values within the area of interest 174. The propagations 146 start from the output nodes and work back to the input nodes. In various embodiments, the propagations 146 may implement layer-wise relevance propagations 146a. Data for a relevance graph 180 is generated in response to the layer-wise relevance propagations 146a. The relevance graph 180, optionally displayed, correlates the input nodes that influence the gradient values 152 in the area of interest 174 within the gradient graph 150. The relevance graph 180 includes multiple relevance values 182a-182e of the input parameters. The relevance values 182a-182e range from most relevant values 182a to least relevant values 182e. While five degrees of the relevance values 182a-182e are illustrated, fewer or more degrees of relevance values may be implemented to meet a design criteria of a particular application.” wherein visual back propagation is performed to back-propagate some feature maps of the one or more neurons)
Wang/Bharadhwaj/Smith does not explicitly disclose but Chen discloses wherein the applying of the first operation and the second operation includes: … to obtain weight masks for feature maps of the one or more neurons; weighting the feature maps of the one or more neurons using the weight masks to obtain weighted feature maps of the one or more neurons (Chen [Abstract]; “The invention discloses an image feature extraction method based on a joint attention mechanism. The steps are as follows: 1: Input the image with features to be extracted into a convolutional neural network to obtain a feature map F; 2: Use a spatial attention module to obtain a spatial weight value Mask matrix W1; 3: The spatial weight mask matrix W1 is multiplied by the feature map F to obtain the feature map F1; 4: The channel attention module is used to obtain the channel weight mask matrix W2 of the feature map F; 5: The channel weight The mask matrix W2 is multiplied by the feature map F to obtain the feature map F2; 6: The feature map F1 and the feature map F2 are connected according to the channel to obtain the feature map F3; 7: The feature map F3 is convolved through c convolution kernels Get the feature map F4. The image feature extraction method based on the spatial-spectral joint attention mechanism of the present invention is used to solve the technical problem of insufficient feature extraction caused by only considering single-layer features in the prior art, and can be widely used in the field of computer vision technology.” wherein the spatial attention module to obtain spatial weight value mask matrix thus reads on obtaining of the weight masks for feature maps through some generic mixing block (spatial attention module); wherein weight masks are applied to the plurality of feature maps to obtain weighted feature maps of the one or more neurons)
It would have been obvious to modify Wang/Bharadhwaj/Smith’s method of applying layer-wise relevance propagation to obtain feature maps by applying Chen’s matrix masks to obtain weighted feature maps of the one or more neurons. One would have been motivated to do so because when one “learns a weight mask related to input, can help the network to highlight interested target information and inhibit the background” (Chen [Page 5 Lines 36-37]).
Wang/Bharadhwaj/Smith discloses wherein the applying of the first operation and the second operation includes: applying the Layer-wise Relevance Propagation (LRP) through the mixing block to obtain … feature maps of the one or more neurons; and applying the Visual Back-Propagation (VBP) to back-propagate the … feature maps of the one or more neurons through the model backbone. Chen discloses weighted feature maps obtained through weight masks. By performing Chen’s weighted masks upon Wang/Bharadhwaj/Smith’s feature maps determined through Layer-wise Relevance Propagation to obtain weighted feature maps, the combination thus discloses applying the Layer-wise Relevance Propagation (LRP) through the mixing block to obtain weight masks for feature maps of the one or more neurons; weighting the feature maps of the one or more neurons using the weight masks to obtain weighted feature maps of the one or more neurons; and applying the Visual Back-Propagation (VBP) to back-propagate the weighted feature maps of the one or more neurons through the model backbone.
Claim 17 recites a non-transitory computer-readable medium configured to execute the exact method of Claim 8 respectively. Thus, Claim 17 are rejected for reasons set forth in the rejection of Claim 8 respectively.
Claims 9, 18 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (“When, Where, and How Does It Fail? A Spatial-Temporal Visual Analytics Approach for Interpretable Object Detection in Autonomous Driving” [2022], hereinafter “Wang”) in view of Bharadhwaj (“Layer-wise Relevance Propagation for Explainable Recommendations” [2018]) in view of Lim et al. (WO 2023086025 A2, hereinafter “Lim”).
Regarding Claim 9,
Wang/Bharadhwaj teaches the method of Claim 1 (and thus the rejection of Claim 1 is incorporated). Wang/Bharadhwaj fails to explicitly disclose but Lim discloses wherein the input is a spectrogram of a speech segment (Lim [0070]; “Deep learning approaches proliferate research on automatic speech emotion recognition. Leveraging the intrinsic time-series structure of speech data, recurrent neural network (RNN) models with attention mechanism have been developed to capture transient acoustic features to understand contextual information. Employing popular techniques from the computer vision domain, audio data can be treated as 1-dimensional arrays or converted to a spectrogram as a 2-dimensional image. Convolutional neural networks (CNNs) can then extract salient features from these audiograms or spectrograms. Current approaches improve performance by combining CNN and RNN, or modelling with multiple modalities. The Relatable Explanation Network (RexNet) model, in accordance with embodiments of the disclosure, starts with a base CNN model to leverage many more XAI techniques available to CNNs than RNNs. The approach is modular and can be generalized to state-of-the-art seeding, evolutionary growth, and reseeding (SER) models” wherein the input is audio speech data converted to a spectrogram as a 2-dimensional image for determining model explanations)
It would have been obvious to modify the generic input images used for processing by the neuron visualization method of Wang/Bharadhwaj to instead be replaced with the speech-segment spectrograms proposed by Lim. One would have been motivated to do so because “few techniques have been developed for audio prediction tasks” (Lim [0071]) thus there is a need for more explanation techniques in the audio-prediction field.
Claim 18 recites a non-transitory computer-readable medium configured to execute the exact method of Claim 9 respectively. Thus, Claim 18 is rejected for reasons set forth in the rejection of Claim 9 respectively.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
“Visualizing and Understanding Convolutional Networks” (Zeiler et al. [2013]) which discloses human-interpretable representations of regions of interest in the neural network model.
“On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation” (Bach et al. [2015]) which discloses neuron explanations derived by layer-wise relevance propagation to determine pixel significance.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JONATHAN J KIM whose telephone number is (571)272-0523. The examiner can normally be reached 8-6.
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, Matt El can be reached on (571) 270-3264. 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.
/JONATHAN J KIM/Examiner, Art Unit 2141
/MATTHEW ELL/Supervisory Patent Examiner, Art Unit 2141
Prosecution Insights