Prosecution Insights
Last updated: November 20, 2026
Application No. 18/917,252

OPTIMIZATION OF TRANSFORMER ENCODERS

Non-Final OA §101§103
Filed
Oct 16, 2024
Priority
Mar 28, 2024 — IN 202441025344 +1 more
Examiner
BLANCHETTE, JOSHUA B
Art Unit
3684
Tech Center
3600 — Transportation & Electronic Commerce
Assignee
Texas Instruments Incorporated
OA Round
1 (Non-Final)
48%
Grant Probability
Moderate
1-2
OA Rounds
1y 6m
Est. Remaining
80%
With Interview

Examiner Intelligence

Grants 48% of resolved cases
48%
Career Allowance Rate
116 granted / 239 resolved
-3.5% vs TC avg
Strong +31% interview lift
Without
With
+31.0%
Interview Lift
resolved cases with interview
Typical timeline
3y 8m
Avg Prosecution
41 currently pending
Career history
291
Total Applications
across all art units

Statute-Specific Performance

§101
35.7%
-4.3% vs TC avg
§103
38.7%
-1.3% vs TC avg
§102
10.1%
-29.9% vs TC avg
§112
11.4%
-28.6% vs TC avg
Black line = Tech Center average estimate • Based on career data from 239 resolved cases

Office Action

§101 §103
DETAILED ACTION Notices to Applicant This communication is a non-final rejection. Claims 1-20, as filed 10/16/2024, are pending and have been considered below. Foreign priority is generally acknowledged to IN 202441025344 and IN 202441025711, both filed 03/28/2024. The priority documents received for these references on 04/04/2025 are both provisional specifications that lack description of generating key data, query data, value data, the claimed first and second matrix multiplication operations, and the SoftMax operation. Accordingly, claims 1-20 have an effective filing date of 10/16/2024. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . 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. Claim Objections Claims 15 and 16 are objected to because of the following informalities. Claim 15 is missing the word “wherein” before “the processing circuitry”. Claim 16 is missing a semicolon after “generate embedding data based on sensor data”. Appropriate correction is required. Claim Rejections - 35 USC § 101 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Step 1 The claims fall within the statutory categories of a process, a machine, and/or an article of manufacture. Claim 1 recites: 1. A method comprising: generating embedding data (abstract idea – mathematical concept) based on sensor data (additional element – insignificant extra-solution activity, namely, mere data gathering); generating key data, query data, and value data based on the embedding data (abstract idea – mathematical concept); performing a first matrix multiplication operation using the key data and the query data to produce a first result by at least generating a value in the first result using a row of the key data and a column of the query data (abstract idea – mathematical concept, namely, a matrix multiplication); performing a SoftMax operation on the first result to produce a second result (abstract idea – mathematical concept, namely, a SoftMax function); and performing a second matrix multiplication operation using the second result and the value data (abstract idea – mathematical concept, namely, a matrix multiplication). Claim 1 is exemplary, and the same analysis applies to claims 9 and 16. Step 2A Prong One Under the broadest reasonable interpretation, the italicized portions recite mathematical concepts such as generating embedding, key, query, and value data, multiplying matrices, and applying a SoftMax function which are mathematical relationships and calculations (see MPEP 2106.04(a)(2)(I)). Dependent claims further narrow or define the abstract idea embodied in the claims. For example, claims 2, 5, 10, 12, 17, and 18 recite which matrices are the left and right inputs of the matrix multiplications. Claims 3, 4, 11, 12, 17, and 18 recite that the query data and the second result are used in a transposed form. Claims 7, 14, and 20, recite the direction along which the SoftMax is computed. Claims 6, 13, and 19 recite a further result based on the calculations. Claims 8 and 15 recite that the calculations are those of an encoder within a vision transformer network. Step 2A Prong Two The judicial exception is not integrated into a practical application because the additional elements: amount to mere instructions to apply the exception. For example, the processing circuitry of claims 9 and 16, the non-transitory computer-readable medium of claim 9, and the transformer network, encoder, and multi-headed attention block of claim 16 amount to invoking computers as a tool to perform the abstract idea (see MPEP 2106.05(f)). The transformer network, encoder, and multi-headed attention blocks are recited functionally at a high level as the setting in which the calculations are performed. add insignificant extra-solution activity to the abstract idea. For example, the sensor data on which the embedding data is based amounts to mere data gathering (see MPEP 2106.05(g)). generally link the abstract idea to a particular technological environment or field of use such as processing sensor data with a vision transformer network (see MPEP 2106.05(h)). The dependent claims add only subject matter consistent with the additional elements in the independent claims. For example, claims 3, 4, 11, 12, 17, and 18 recite reading the query data from memory and writing the second result to a memory in a transposed order, which amounts to insignificant extra-solution activity, namely, storing and retrieving data in memory (see MPEP 2106.05(g)). Viewed as an ordered combination, the additional elements add nothing beyond the elements taken individually. Nothing indicates that the combination improves the functioning of a computer or any other technology. Collectively, the additional elements provide a generic computer implementation of the abstract idea and do not impose a meaningful limit on it. Step 2B The claims do not include additional elements sufficient to amount to significantly more than the judicial exception. As discussed under Step 2A Prong Two, the additional elements amount to no more than mere instructions to apply the exception, insignificant extra-solution activity, and a general link to a particular technological environment or field of use. The extra-solution activity is also well-understood, routine, and conventional. For example, the transpose-reading of the query data from memory and the transpose-writing of the second result to a memory in claims 3, 4, 11, 12, 17, and 18 amount to storing and retrieving information in memory (Versata Dev. Group, MPEP 2106.05(d)(II)(iv)). Gathering data from sensors which was well-understood, routine, and conventional before the effective filing date as demonstrated by applicant’s specification (as published) in [0037] (“sensors 101 may be representative of cameras, radar devices, or another sensor of the like configured to collect sensor data for executing transformer network 105”) and Chen (US20230334876A1) in, e.g., [0021]. Viewed as an ordered combination, the additional elements add nothing beyond the elements taken individually. As discussed under Step 2A Prong Two, the dependent claims add only invocations of computers as tools to perform the abstract idea and subject matter consistent with the additional elements in the independent claims. Viewed as an ordered combination, the additional elements add nothing beyond the elements taken individually. Nothing indicates that the combination improves the functioning of a computer or any other technology. 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 set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied 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, 2, 6, 8, 9, 10, 13, 15, 16, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Venkataramanan (US20240160896A1) in view of Chen (US20230334876A1). Regarding claim 1, Venkataramanan discloses: A method (“an example method 500 for generating and providing attention propagation output,” [0081]) comprising: --generating embedding data (“Each patch can then be projected into an embedding,” [0034]); --generating key data, query data, and value data based on the embedding data (“the input data may be linearly projected (e.g., multiplied using learned parameters) into three matrices: a query matrix Q,” [0021]; “The tokenized image can then be used as input to the first transformer 110A,” [0034]); --performing a first matrix multiplication operation using the key data and the query data to produce a first result by at least generating a value in the first result using a row of the key data and a column of the query data (“the transformer may compute the dot product of the key matrix and the transposed query matrix,” [0084]); --performing a SoftMax operation on the first result to produce a second result (“apply a softmax operation (or other operation) to the result to generate the attention matrix,” [0084]); and --performing a second matrix multiplication operation using the second result and the value data (“the transformer may compute the dot product of the attention matrix and value matrix,” [0085]). Venkataramanan does not expressly disclose but Chen teaches: based on sensor data (“gridding a red-green-blue (RGB) image captured by a camera sensor mounted on an ego vehicle into a plurality of image patches… generating different image patch embeddings to provide correlations between the plurality of image patches and the RGB image” [0004]). A POSITA before the effective filing date would have been motivated to expand Venkataramanan’s vision transformer to include the RGB image captured by a camera sensor of Chen because this would allow the transformer to detect lane boundaries in camera images to “enable planning and control of an ego vehicle” (Chen [0029]). Additionally, each element is taught by Venkataramanan or Chen. The camera sensor input of Chen does not affect the normal functioning of the elements of the claim which are taught by Venkataramanan. Because the elements do not affect the normal functioning of each other, the results of their combination would have been predictable. Therefore, before the effective filing date of the claimed invention, it would have been obvious to combine the teachings of Chen with the teachings of Venkataramanan since the result is merely a combination of old elements, and, since the elements do not affect the normal functioning of each other, the results of the combination would have been predictable. Regarding claim 2, Venkataramanan further discloses: wherein performing the first matrix multiplication operation comprises: supplying the key data as a left matrix input to the first matrix multiplication operation; supplying the query data as a right matrix input to the first matrix multiplication operation; and matrix multiplying the left matrix input with the right matrix input (“the transformer may compute the dot product of the key matrix and the transposed query matrix,” [0084]). Regarding claim 6, Venkataramanan further discloses: outputting a third result based on an output of the second matrix multiplication operation (“the transformer may process the features using a feedforward sub-block (e.g., an MLP) of the transformer to generate an output,” [0086]). Regarding claim 8, Venkataramanan further discloses: wherein the first matrix multiplication operation, the SoftMax operation, and the second matrix multiplication operation are performed within a context of an encoder within a vision transformer network (“Transformers, and in particular vision transformers, have become increasingly common,” [0020]; “the method 500 may be performed by the transformer 110A of FIGS. 1, 2A, and/or 3A, by encoder transformers 410A-N of FIG. 4 , and/or by decoder transformers 450A-N of FIG. 4,” [0081]). Claim 9 is substantially similar to claim 1 and is rejected with the same reasoning. The Examiner further notes that Venkataramanan discloses: “non-transitory, computer-readable media comprising instructions that, when executed by one or more processors of a processing system, cause the processing system to perform the aforementioned methods” ([0006]). Claims 10, 13, 15, and 19 are substantially similar to claims 2, 6, 8, and 6 (respectively) and are rejected with the same reasoning. Claim 16 is substantially similar to claim 1 and is rejected with the same reasoning. The Examiner further notes that Venkataramanan discloses: “one or more processors of a processing system” ([0006]) and “the method 500 may be performed by the transformer 110A of FIGS. 1, 2A, and/or 3A, by encoder transformers 410A-N of FIG. 4 , and/or by decoder transformers 450A-N of FIG. 4,” ([0081]). Venkataramanan further discloses: wherein the encoder includes one or more multi-headed attention blocks (“self-attention sub-blocks 115 may implement multi-head self-attention (MSA),” [0033]). Claims 3, 11, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Venkataramanan in view of Chen and Tran (US20190187986A1). Regarding claim 3, Venkataramanan further discloses: supplying the query data in a transposed form to the right matrix input (“the transformer may compute the dot product of the key matrix and the transposed query matrix, and apply a softmax operation (or other operation) to the result to generate the attention matrix,” [0084]). Venkataramanan does not expressly disclose but Tran teaches: wherein supplying the query data as the right matrix input comprises transpose-reading the query data from memory (“The streaming engine supports implicit matrix transposition with transposed streams. Transposed streams avoid the cost of explicitly transforming the data in memory,” [0144]). A POSITA before the effective filing date would have been motivated to expand the vision transformer using sensor data of Venkataramanan and Chen to include the transposed data stream of Tran because this would keep a real-time camera application from failing, since such an application “fails if the transformed data stream is not available for output when scheduled” (Tran [0002]). Additionally, each element is taught by Venkataramanan, Chen, or Tran. The camera sensor input of Chen and the transposed streams of Tran do not affect the normal functioning of the elements of the claim which are taught by Venkataramanan. Because the elements do not affect the normal functioning of each other, the results of their combination would have been predictable. Therefore, before the effective filing date of the claimed invention, it would have been obvious to combine the teachings of Chen and Tran with the teachings of Venkataramanan since the result is merely a combination of old elements, and, since the elements do not affect the normal functioning of each other, the results of the combination would have been predictable. Claim 11 is substantially similar to claim 3 and is rejected with the same reasoning. Claim 17 is substantially similar to claims 2 and 3 and is rejected with the same reasoning. Claims 4, 5, 12, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Venkataramanan in view of Chen, Weissenborn (US20210383199A1), and Redfern (US20180253402A1). Regarding claim 4, Venkataramanan does not expressly disclose but Weissenborn teaches: wherein performing the second matrix multiplication operation using the second result and the value data comprises performing the second matrix multiplication operation using the transpose-written second result and the value data (“determining a dot product of a transpose of the attention matrix A and the value input matrix XVALUE,” [0067]). A POSITA before the effective filing date would have been motivated to expand the vision transformer with sensor data of Venkataramanan and Chen to include the transpose of the attention matrix of Weissenborn because computing the output from the transposed attention matrix and the values would improve the accuracy of the resulting representations, “thereby allowing the values of the slot vectors to be refined to improve their representational accuracy over one or more iterations of the slot attention module” (Weissenborn [0024]). Venkataramanan does not expressly disclose but Redfern teaches: transpose-writing the second result to a memory, resulting in a transpose-written second result (“Examples of output formatting 116 include seam removal, stride removal, zero padding, and matrix transpose,” [0033]; “The data vectors are stored into external memory (not shown) via the level one data cache 112 and the level two unified cache 106,” [0033]). A POSITA before the effective filing date would have been motivated to expand the vision transformer with sensor data and attentional matrix transposition of Venkataramanan, Chen, and Weissenborn to include the output formatting matrix transpose of Redfern because this would support applications such as “object detection and recognition, and vision” (Redfern [0019]). Additionally, each element is taught by Venkataramanan, Chen, Weissenborn, or Redfern. The camera sensor input of Chen, the transposed attention matrix of Weissenborn, and the output formatting matrix transpose of Redfern do not affect the normal functioning of the elements of the claim which are taught by Venkataramanan. Because the elements do not affect the normal functioning of each other, the results of their combination would have been predictable. Therefore, before the effective filing date of the claimed invention, it would have been obvious to combine the teachings of Venkataramanan with the teachings of Chen, Weissenborn, and Redfern since the result is merely a combination of old elements, and, since the elements do not affect the normal functioning of each other, the results of the combination would have been predictable. Regarding claim 5, Venkataramanan does not expressly disclose but Weissenborn further teaches: wherein performing the second matrix multiplication operation comprises: supplying the transpose-written second result as a left matrix input to the second matrix multiplication operation; supplying the value data as a right matrix input to the second matrix multiplication operation; and matrix multiplying the left matrix input with the right matrix input (“determining a dot product of a transpose of the attention matrix A and the value input matrix XVALUE,” [0067]). The motivation to combine Weissenborn is the same as in claim 4. Claim 12 is substantially similar to claims 4 and 5 and is rejected with the same reasoning. Claim 18 is substantially similar to claims 4 and 5 and is rejected with the same reasoning. Claims 7, 14, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Venkataramanan in view of Chen and Koohpayegani (Koohpayegani et al., "SimA: Simple Softmax-free Attention for Vision Transformers", arXiv:2206.08898v1). Regarding claim 7, Venkataramanan does not expressly disclose but Koohpayegani teaches: wherein performing the SoftMax operation comprises performing a height-wise SoftMax operation on the first result (“In XCA, we calculate the attention matrix with A = K^TQ where A is a D × D matrix. Next, we apply Softmax on each columns, so that columns sum to one,” page 2). A POSITA before the effective filing date would have been motivated to expand the vision transformer with sensor data of Venkataramanan and Chen to include the column-wise SoftMax of Koohpayegani because this would normalize the attention “so that each token is a weighted average of the values of all tokens” (Koohpayegani page 2). Additionally, each element is taught by Venkataramanan, Chen, or Koohpayegani. The camera sensor input of Chen and the column-wise SoftMax of Koohpayegani do not affect the normal functioning of the elements of the claim which are taught by Venkataramanan. Because the elements do not affect the normal functioning of each other, the results of their combination would have been predictable. Therefore, before the effective filing date of the claimed invention, it would have been obvious to combine the teachings of Venkataramanan with the teachings of Chen and Koohpayegani since the result is merely a combination of old elements, and, since the elements do not affect the normal functioning of each other, the results of the combination would have been predictable. Conclusion The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure. El-Nouby (El-Nouby et al., "XCiT: Cross-Covariance Image Transformers", arXiv:2106.09681v2) discloses computing an attention matrix with the key matrix as the left input and the transposed query matrix as the right input: “attn = (k @ q.transpose(-2, -1)) # Computing the block diagonal cross-covariance matrix” (page III). Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOSHUA BLANCHETTE whose telephone number is (571)272-2299. The examiner can normally be reached on Monday - Thursday 7:30AM - 6:00PM, 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, Shahid Merchant, can be reached on (571) 270-1360. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. 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. /JOSHUA B BLANCHETTE/ Primary Examiner, Art Unit 3684
Read full office action

Prosecution Timeline

Oct 16, 2024
Application Filed
Oct 28, 2026
Non-Final Rejection mailed — §101, §103 (current)

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12802824
AUTOMATED SYSTEM AND METHODS TO EVALUATE PLANNED TREATMENT PERFORMANCE AND REAL TREATMENT PERFORMANCE USING THREE-DIMENSIONAL VIRTUAL DENTAL MODELS
2y 7m to grant Granted Nov 17, 2026
Patent 12784885
DATA COMMAND CENTER VISUAL DISPLAY SYSTEM
6y 8m to grant Granted Nov 03, 2026
Patent 12784883
THERAPY MANAGEMENT SYSTEM
1y 11m to grant Granted Nov 03, 2026
Patent 12755852
Systems and Methods for An Artificial Intelligence System
1y 4m to grant Granted Oct 06, 2026
Patent 12731674
METHOD AND CONTROL UNIT FOR CONTROLLING A MEDICAL IMAGING INSTALLATION
2y 11m to grant Granted Sep 08, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

1-2
Expected OA Rounds
48%
Grant Probability
80%
With Interview (+31.0%)
3y 8m (~1y 6m remaining)
Median Time to Grant
Low
PTA Risk
Based on 239 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month