DETAILED ACTION
This action is responsive to the Amendments and Remarks received 05/18/2026 in which no claims are cancelled, 1, 2, 9, 10, 17, and 18 are amended, and no claims are added as new claims.
Response to Arguments
On pages 8–9 of the Remarks, Applicant presents arguments regarding the subject matter eligibility of the claims under 35 U.S.C. 101. Examiner considered Applicant’s Remarks and notes that MPEP 2106.04(a)(2) specifically discusses insurance as a fundamental economic principle or practice that may not be patent eligible subject matter because the claimed subject matter is drawn to methods of organizing human activity. When the claims recite the fundamental principle as opposed to merely limitations based on or involving the principle, the claims may be drawn to an abstract idea. It is further explained that certain activity between a person and a computer may fall within the category of patent ineligible subject matter and that the determination should be based on whether the activity itself, in this case determining an insurance loss, falls within one of the enumerated sub-groupings of methods of organizing human activity. As MPEP 2106(a)(2)(II)(B) explains, “processing insurance claims for a covered loss or policy event under an insurance policy” is an “example[] of subject matter where the commercial or legal interaction is an agreement in the form of contracts.” That Section goes on to explain that a computer that can process an underwriting request is the type of subject matter courts have found lacking subject matter eligibility. In this case, a computer program, i.e. a vision transformer, is simply taking the place of a human decision maker in determining how much damage/loss has occurred to an insured property. Further still, MPEP 2106(a)(2)(III) explains that mental processes are patent ineligible citing Versata Dev. Group v. SAP Am., Inc., 793 F.3d 1306, 1335, 115 USPQ2d 1681, 1702 (Fed. Cir. 2015) (“[c]ourts have examined claims that required the use of a computer and still found that the underlying, patent-ineligible invention could be performed via pen and paper or in a person’s mind.”). As MPEP 2106(a)(2)(III)(C) explains, if the claimed invention covers a mental process and is merely using a computer as a tool for performing the concept, the claim recites a patent ineligible mental process. For these reasons, Examiner is not persuaded by Applicant’s Remarks regarding the patent eligibility of the claimed subject matter. However, because the claims are otherwise obvious under 35 U.S.C. 103, a 35 U.S.C 101 rejection appears duplicative and 35 U.S.C. 101 is not a threshold issue requiring resolution where otherwise the claims are unpatentable. This issue under 35 U.S.C. 101 is held in abeyance.
On pages 10–11 of the Remarks, Applicant contends Dosovitskiy is deficient because it is silent with respect to the input images being, specifically, aerial images and images of a property. In this art, the content of the pictures being subjected to image processing is not relevant and is not patent eligible subject matter, and therefore such a feature cannot serve to distinguish from the prior art. Dosovitskiy’s vision transformer can work on any images, including aerial images and images of a property and the placement and orientation of the camera is technologically irrelevant to the image processing task. Accordingly, Examiner is unpersuaded of error.
On pages 11–12 of the Remarks, in explaining what Hurliman teaches, Applicant essentially describes in detail the feature Applicant argued was missing from the teachings of Dosovitskiy. In other words, Applicant’s Remarks demonstrate the propriety of citing the teachings of Hurliman to teach or suggest the features of the claimed invention related to aerial images used for determining property damage. Applicant contends Hurliman is deficient for failing to teach a vision transformer and associated output embeddings, features that the rejection clearly relied on the teachings of Dosovitskiy to teach or suggest. Therefore, Applicant’s arguments attack the references individually rather than address what the combination would teach or suggest to one of ordinary skill in the art. Applicant contends Hurliman does not teach a loss model to predict a loss score. First, this averred feature represents the mental process or fundamental economic activity that is patent ineligible. If the claim’s novelty hinges on this feature, then the claimed subject matter is squarely drawn to a mental process or fundamental economic process. Second, Hurlimann’s description of insurance loss and damage estimation would teach or suggest the averred feature. Finally, the arguments against the combination of Dosovitskiy and Hurlimann are moot in view of the new grounds of rejection necessitated by amendment. Specifically, the rejection under 35 U.S.C. 103 now additionally relies on the teachings of Portail, which further teaches aerial photographs used for insurance purposes and links those purposes to AI technology specifically involving vision transformers. Therefore, for all the foregoing reasons, Examiner is unpersuaded of patentability.
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 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1–3, 6, 8–11, 14, and 16–20 are rejected under 35 U.S.C. 103 as being unpatentable over Dosovitskiy et al., “An Image Is Worth 16x16 Words: Transformers for Image Recognition at Scale,” In International Conference on Learning Representations (ICLR) 2021 (herein “Dosovitskiy”), Hurliman (US 2025/0384497 A1), and Portail (US 11,676,298 B1).
Examiner notes this application seems to be claiming an abstract idea, i.e. a mental step performed by a computer. It appears to be nothing more than replacing a mental step of an insurance agent looking up Google satellite images of my house to determine what material a home (structure, e.g. roof) is made out of (and perhaps square footage of the property and whether I have a pool) to generate the inputs necessary to generate what the Applicant terms, a loss score, which is the same as an insurance agent looking your home up on Google Maps and generating policy limits for a new home owners policy using the aerial and/or street view images from Google. Perhaps shadows from trees make it hard to see the full outline of the roof shape in the aerial view and perhaps a trained AI model could be better at deciphering roof shape/material than a human when the satellite image is ambiguous in some way. It seems important for Applicant to communicate to the Office what technological advancement is represented by the disclosure (Specification) that is not simply applying existing AI modeling techniques, specifically the vision transformer described by Dosovitskiy, to Google’s aerial images to generate some output, like roof information or pool information.
Regarding claim 1, the combination of Dosovitskiy, Hurliman, and Portail teaches or suggests a computing system comprising: one or more processors; and one or more storage devices that comprise instruction code that is executable by the one or more processors to cause the computing system to (Huliman, ¶ 0005: teaches a computer system for generating a building inspection report or building profile): receive an aerial image that depicts a property (Jenks, ¶ 0056 and Fig. 10: teaches receiving an aerial image of a property; Hurliman, ¶¶ 0029 and 0152: teaches using aerial imagery for generating a building profile for insurance claims or underwriting); generate, by a vision transformer, one or more input embeddings associated with the aerial image (Examiner finds original claim 2 elaborates on what is necessary for this step; Dosovitskiy, Fig. 1, Description and Section 3.1: teaches the input image is split into fixed-size (e.g. 16x16) patches, generating a sequence of input embeddings comprising the 1-D patches, added position embeddings, and added classification token); transform, by the vision transformer, the one or more input embeddings to one or more output embeddings that specify features of the aerial image of the property including at least a feature comprising damage to the feature as a specified feature of the property (Dosovitskiy, Section 3.1: teaches outputting patch embeddings; Applicant’s paragraph [0075] explains the output embeddings are patch embeddings; Examiner notes that a vision transformer outputs an image classification by passing the final state of a special [CLS] token (or a global average of patch embeddings) through a Feed-Forward Network (MLP head); Examiner further notes that unlike CNNs that produce 2D feature maps, ViTs produce a 1D sequence of patch embeddings; For feature extraction, the output of a ViT is a high-dimensional feature vector sequence; Hurliman, ¶¶ 0029 and 0141: teaches a building profile created by a trained AI model can be used to address insurance claims and underwriting (i.e. loss prediction/determination); Hurliman, ¶¶ 0127 and 0160: teaches the AI model can generate repair data such as type of repair and cost; In combination, it would be obvious to use Dosovitskiy’s vision transformer to assess damage to a property for purposes of insurance underwriting and claim adjusting (i.e. processing insurance claims) as further taught by Portail, col. 9, ll. 8–46 and col. 24, ln. 53–col. 25, ln. 30, which explains the use of computer vision technology to facilitate insurance business processes were known to include vision transformers), wherein the vision transformer is trained using a self-supervised learning technique to generate the one or more output embeddings that specify the features of the aerial image of the property including the feature comprising damage (Dosovitskiy, Appx. B.1.2: teaches or suggests using self-supervised learning during training; Examiner notes more information on the teacher-student models used for self-supervised training can be found in Caron, cited under the Conclusion Section of this Office Action); and communicate at least one of the one or more output embeddings to a loss model, wherein the loss model is trained to predict a loss score associated with the property depicted in the aerial image to assess a loss associated with damage to the feature; and output an indication of the loss score associated with the property and the feature comprising damage (Hurliman, ¶¶ 0029 and 0141: teaches a building profile created by a trained AI model can be used to address insurance claims and underwriting (i.e. loss prediction/determination); Hurliman, ¶¶ 0127 and 0160: teaches the AI model can generate repair data such as type of repair and cost; see references under the Conclusion Section of this Office Action explaining that an insurance loss score is a risk assessment, loss rating, property valuation, underwriting value for an insurance premium, etc.; These definitions are subsumed by Hurliman’s teachings to one of ordinary skill in the art).
One of ordinary skill in the art, before the effective filing date of the claimed invention, would have been motivated to combine the elements taught by Dosovitskiy, with those of Hurliman, because both references are drawn to the same field of endeavor and because, as evidenced by the teachings of Portail regarding the use of vision transformers for performing insurance business processes, combining Hurliman’s use of a trained neural network utilizing either supervised or unsupervised learning (¶¶ 0081–0082) for generating an AI insurance claims or underwriting system with Dosovitskiy’s explanation of how to implement Hurliman’s transformer-based neural network, is a mere combination of prior art elements, according to known methods to yield a predictable result. This rationale applies to all combinations of Dosovitskiy and Hurliman used in this Office Action unless otherwise noted.
One of ordinary skill in the art, before the effective filing date of the claimed invention, would have been motivated to combine the elements taught by Dosovitskiy and Hurliman, with those of Portail, because all three references are drawn to the same field of endeavor or drawn to a similar problem, because Portail itself applies the use of vision transformers to achieve or support insurance damage assessment, and because, as evidenced by the teachings of Portail regarding the use of vision transformers for performing insurance business processes, combining Portails’s use of a vision transformer for insurance business processes with Dosovitskiy’s explanation of how to implement Portail’s vision transformer, is a mere combination of prior art elements, according to known methods to yield a predictable result. This rationale applies to all combinations of Dosovitskiy, Hurliman, and Portail used in this Office Action unless otherwise noted.
Regarding claim 2, the combination of Dosovitskiy, Hurliman, and Portail teaches or suggests the computing system according to claim 1, wherein the instruction code that causes the computing system to generate the one or more input embeddings is executable to cause the computing system to: divide the aerial image into a plurality of non-overlapping patches; generate the one or more input embeddings as one or more input embeddings that comprises the plurality of non-overlapping patches, positional embeddings, and a classification token (Dosovitskiy, Fig. 1, Description and Section 3.1: teaches the input image is split into fixed-size (e.g. 16x16) patches, generating a sequence of input embeddings comprising the 1-D patches, added position embeddings, and added classification token); output, by the vision transformer, the one or more output embeddings comprising at least one or more patch embeddings and a classification token embedding along with contextual information about the plurality of non-overlapping patches as influenced by patch relationships therebetween (Dosovitskiy, Section 3.1: teaches outputting patch embeddings; Applicant’s paragraph [0075] explains the output embeddings are patch embeddings; Examiner notes that a vision transformer outputs an image classification by passing the final state of a special [CLS] token (or a global average of patch embeddings) through a Feed-Forward Network (MLP head); Examiner further notes that unlike CNNs that produce 2D feature maps, ViTs produce a 1D sequence of patch embeddings; For feature extraction, the output of a ViT is a high-dimensional feature vector sequence; Dosovitskiy, Fig. 1, Description and Section 3.1: teaches the input image is split into fixed-size (e.g. 16x16) patches, generating a sequence of input embeddings comprising the 1-D patches, added position embeddings, and added classification token; According to Applicant’s published para. [0047], the patch relationships and dependencies are achieved using a self-attention layer; Dosovitskiy, e.g. Sections 3.1 and 4.5: explains that prior art vision transformers consist of self-attention layer(s)); via the loss model and the one or more output embeddings, assess the loss associated with damage to the feature that occupies at least two or more patches of the plurality of non-overlapping patches of the arial image using the classification token embeddings (); and via the loss model and the one or more output embeddings, assess the loss associated with damage to the feature that occupies a single patch of the plurality of non-overlapping patches of the ariel image using the one or more patch embeddings ().
Regarding claim 3, the combination of Dosovitskiy, Hurliman, and Portail teaches or suggests the computing system according to claim 1, wherein the vision transformer is trained using a first dataset that comprises unlabeled aerial images depicting properties (Dosovitskiy, Section 4.6: suggests self-supervised (i.e. non-labeled data) pre-training is a viable alternative to supervised (i.e. labeled data) pre-training).
Regarding claim 6, the combination of Dosovitskiy, Hurliman, and Portail teaches or suggests the computing system according to claim 1, wherein the loss model is trained using a second dataset that comprises labeled aerial images depicting properties (Hurliman, e.g. ¶ 0029: teaches a building profile created by a trained AI model can be used to address insurance claims and underwriting (i.e. loss prediction/determination); Hurliman, ¶ 0081: teaches AI models can be trained using supervised learning, which uses labeled data).
Regarding claim 8, the combination of Dosovitskiy, Hurliman, and Portail teaches or suggests the computing system according to claim 1, wherein the aerial image depicts an entirety of the property within a frame of the aerial image (Examiner notes the content of imagery is non-patentable as non-functional descriptive material; Hurliman, ¶ 0098: teaches aerial drone data, which can obviously be at a height to take a picture of a property wherein the whole property can fit within a single image frame).
Claim 9 lists the same elements as claim 1, but in CRM form rather than system form. Therefore, the rationale for the rejection of claim 1 applies to the instant claim.
Claim 10 lists the same elements as claim 2, but in CRM form rather than system form. Therefore, the rationale for the rejection of claim 2 applies to the instant claim.
Claim 11 lists the same elements as claim 3, but in CRM form rather than system form. Therefore, the rationale for the rejection of claim 3 applies to the instant claim.
Claim 14 lists the same elements as claim 6, but in CRM form rather than system form. Therefore, the rationale for the rejection of claim 6 applies to the instant claim.
Claim 16 lists the same elements as claim 8, but in CRM form rather than system form. Therefore, the rationale for the rejection of claim 8 applies to the instant claim.
Claim 17 lists the same elements as claim 1, but in method form rather than system form. Therefore, the rationale for the rejection of claim 1 applies to the instant claim.
Claim 18 lists the same elements as claim 2, but in method form rather than system form. Therefore, the rationale for the rejection of claim 2 applies to the instant claim.
Claim 19 lists the same elements as claim 3, but in method form rather than system form. Therefore, the rationale for the rejection of claim 3 applies to the instant claim.
Claim 20 lists the same elements as claim 6, but in method form rather than system form. Therefore, the rationale for the rejection of claim 6 applies to the instant claim.
Claims 4 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Dosovitskiy, Hurliman, Portail, and Cobo (US 2022/0261713 A1).
Regarding claim 4, the combination of Dosovitskiy, Hurliman, Portail, and Cobo teaches or suggests the computing system according to claim 3, wherein a first subset of the first dataset comprises one or more images that depict overhead views of properties and a second subset of the first dataset comprises one or more images that depict oblique views of the properties (Cobo, ¶ 0028: teaches the digital data image set for building and roof data can comprise both aerial nadir and/or oblique images, which can be used to develop a digital surface model of the building (DSM)).
One of ordinary skill in the art, before the effective filing date of the claimed invention, would have been motivated to combine the elements taught by Dosovitskiy, Hurliman, and Portail, with those of Cobo, because Hurliman, Portail, and Cobo are insurance references drawn to the same field of endeavor such that one wishing to implement a vision transformer for characterizing building features such as roofs and vegetation for insurance purposes and would be interested in aerial imagery to characterize structures would be led to their relevant teachings and because combining Hurliman’s and Portail’s use of aerial nadir and off-nadir images with Cobo’s nadir images represents nothing more than a mere combination of prior art elements, according to known methods to yield a predictable result. This rationale applies to all combinations of Dosovitskiy, Hurliman, Portail, and Cobo used in this Office Action unless otherwise noted.
Claim 12 lists the same elements as claim 4, but in CRM form rather than system form. Therefore, the rationale for the rejection of claim 4 applies to the instant claim.
Claims 5 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Dosovitskiy, Hurliman, Portail, and Caron et al., “Emerging Properties in Self-Supervised Vision Transformers,” Proceedings of the IEEE/CVF International conference on computer vision, pp. 9650–9660, 2021 (herein “Caron”).
Regarding claim 5, the combination of Dosovitskiy, Hurliman, Portail, and Caron teaches or suggests the computing system according to claim 1, wherein the vision transformer corresponds to a first vision transformer, wherein training the first vision transformer comprises: generating a second vision transformer; and using the second vision transformer to train the first vision transformer (Examiner interprets this claim in view of Applicant’s ¶ 0058, wherein it is explained this training is done using Caron’s teacher-student training model; Caron, Sections 2 and 3: teach self-supervised learning and teach combining the teacher-student self-supervised model with Dosovitskiy’s vision transformer in Section 3.2).
One of ordinary skill in the art, before the effective filing date of the claimed invention, would have been motivated to combine the elements taught by Dosovitskiy Hurliman, and Portail, with those of Caron, because Caron itself combines its teachings with Dosovitskiy’s such that one looking for motivation to combine the teachings of Caron and Dosovitskiy would simply need to look at Caron’s combination for guidance. Therefore, the combination represents nothing more than a mere combination of prior art elements, according to known methods, to yield a predictable result. This rationale applies to all combinations of Dosovitskiy, Hurliman, Portail, and Caron used in this Office Action unless otherwise noted.
Claim 13 lists the same elements as claim 5, but in CRM form rather than system form. Therefore, the rationale for the rejection of claim 5 applies to the instant claim.
Claims 7 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Dosovitskiy, Hurliman, Portail, and Lo (US 2025/0111668 A1).
Regarding claim 7, the combination of Dosovitskiy, Hurliman, Portail, and Lo teaches or suggests the computing system according to claim 6, wherein labels associated with the labeled aerial images of the second dataset comprise an indication of a loss score associated with respective properties depicted in the aerial images (Lo, ¶ 0096: teaches a supervised (i.e. trained with labeled data) machine learning model trained to output a fire risk score based on labeled data such as aerial images).
One of ordinary skill in the art, before the effective filing date of the claimed invention, would have been motivated to combine the elements taught by Dosovitskiy, Hurliman, and Portail, with those of Lo, because Hurliman, Portail, and Lo are drawn to the same field of endeavor such that one wishing to generate insurance risk scores using machine learning would be led to their relevant teachings and because combining Hurliman’s cost score generated by an AI model with Lo’s teachings that aerial images can be fed into a supervised machine learning model that has labelled training data to generate a fire risk score based on previous fire damage represents nothing more than a mere combination of prior art elements, according to known methods, to yield a predictable result. This rationale applies to all combinations of Dosovitskiy, Hurliman, Portail, and Lo used in this Office Action unless otherwise noted.
Claim 15 lists the same elements as claim 7, but in CRM form rather than system form. Therefore, the rationale for the rejection of claim 7 applies to the instant claim.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Hurliman (US 2025/0384497 A1) teaches building inspection for insurance purposes such as claims, underwriting, etc. (¶ 0029) and using various AI models including CNNs and transformers with various unsupervised to supervised learning models (¶¶ 0081 and 0082) wherein data can include aerial imagery (¶ 0152).
Ulin (US 2024/0127348 A1) teaches determining roof characteristics for insurance purposes (e.g. ¶ 0006) using AI models (e.g. ¶¶ 0023 and 0024) for data sources such as aerial imagery (e.g. ¶ 0021). See also ¶ 0077. The publication also describes the output as an expected loss score (¶ 0096).
Kabealo (US 2026/0037819 A1) teaches that self-supervised means trained on unlabeled data (¶¶ 0062 and 0063) and further teaches Vision Transformers (ViT) with embeddings (¶ 0008).
Jagannathan (US 2015/0302529 A1) teaches roof condition assessment for evaluating insurance underwriting and pricing (¶¶ 0002 and 0003) wherein the roof condition risk score can be generated by image processing techniques of a supervised or unsupervised learning model (¶ 0050).
Hoshen (US 2023/0281959 A1) teaches Vision Transformers use input grids of 14x14 or 16x16 (¶ 0207).
Deetlefs (US 2026/0030547 A1) teaches machine learning models can be self-supervised transformer networks (¶ 0017), teaches dividing an image into 256 patches, teaches classification (CLS) tokens, and ViT (vision transformers) (¶ 0120).
Salehi (US 2026/0017921 A1) teaches self-supervised learning of a ViT wherein image data is input into the ViT to generate patches of fixed size PxP which are embedded in a space via linear projection layer and further teaches using class tokens (CLS) adjoined to patch embeddings output by the ViT (¶¶ 0033 and 0053–0054). The publication further teaches a vision transformer fed input patches and linearly embedding the patches wherein position embeddings are added to the linear embedded patches to output a sequence of vectors and those vectors can added to a learnable classification token (CLS token) provided as input to the ViT (¶ 0056).
Torok (US 2026/0004786 A1) teaches a vision transformer (ViT) receiving a sequence of vectors generated from fixed-sized patches of an image for predicting a classification wherein a position embedding is added to the vectors and a classification token is added to the vector sequence to output a classification of the input data (¶ 0106).
Alidoost et al., “A CNN-Based Approach for Automatic Building Detection and Recognition of Roof Types Using a Single Aerial Image,” PFG (2018).
Linus Scheibenreif, et al., “Self-supervised vision transformers for land-cover segmentation and classification,” In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 1422–1431, 2022.
Mathilde Caron, et al., “Emerging properties in self-supervised vision transformers,” In Proceedings of the IEEE/CVF international conference on computer vision, pp. 9650–9660, 2021. This publication teaches self-supervised vision transformers using a student teacher model.
Linus Scheibenreif, Michael Mommert, and Damian Borth. Contrastive self-supervised data fusion for satellite imagery. In International Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2022.
Bao et al., “Channel Vision Transfomers: An Image Is Worth 1x16x16 Words,” In International Conference on Learning Representations (ICLR) 19 Apr 2024. This publication marries together Dosovitskiy’s ViT, in which Dosovitskiy suggests self-supervised pre-training (Section 2), with Caron’s self-supervised learning using teacher-student models (Appx. E).
Jenks (US 2022/0180016 A1) teaches dividing a DSM image into a set of patches, training a learning model, and determining a set of features for a roof of a property for insurance purposes (e.g. ¶¶ 0006, 0026, 0027), wherein the property’s features are fed into a system that generates insurance loss scores such as replacement cost, insurance premium rates, cost of materials, etc. (e.g. ¶ 0124), and wherein aerial or satellite images including off-nadir images can be used to generate overhead and orthorectified oblique views (DSM views) of a building (e.g. ¶ 0058).
Williams (US 2025/0006052 A1) teaches “loss score” in the field of insurance is just a risk assessment or predicted cost of a casualty and is used to determine an insurance premium (i.e. loss) (e.g. ¶ 0075).
Brown (US 2023/0011777 A1) teaches an “insurance loss score” determined by an AI model can include property valuation and real estate condition (e.g. ¶ 0107).
Reznek (US 2022/0318916 A1) teaches labeled training data can include historical images labeled with repair costs (¶ 0072).
Hu et al., “Teacher-Student Architecture for Knowledge Distillation: A Survey,” 8 Aug. 2023.
Vacca et al., "The Use of Nadir and Oblique UAV Images for Building Knowledge," ISPRS Int. J. Geo-Inf. 2017.
Malalur (US 2025/0022274 A1) teaches using vision transformers (¶ 0076) for insurance damage claims (e.g. ¶ 0021).
Wegg (US 2024/0312040 A1) teaches insurance claim data, loss data, damage/risk/valuation models, etc. (e.g. ¶¶ 0058 and 0095) effectuated by vision transformers (¶ 0087).
Dearth (US 2023/0306742 A1) teaches insurance work (e.g. ¶ 0003) performed by computer vision such as a vision transformer (ViT) (¶ 0029).
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Michael J Hess whose telephone number is (571)270-7933. The examiner can normally be reached Mon - Fri 9:00am-5:30pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, William Vaughn can be reached on (571)272-3922. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8933.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/MICHAEL J HESS/Examiner, Art Unit 2481
Prosecution Insights