DETAILED ACTION
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 .
Response to Amendment
Applicant’s amendments filed on 05/11/2026 to the specification have overcome the specification objection as preciously set forth in the Non-Final Rejection Office Action mailed on 01/12/2026.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 02/18/2026 has/have been considered by the examiner.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claim(s) 1-9 is/are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Sultana et al (US 20240203098 A1), hereinafter Sultana.
-Regarding claim 1, Sultana discloses an information processing apparatus comprising a processor configured to (Abstract; FIGS. 1-19
PNG
media_image1.png
313
584
media_image1.png
Greyscale
; [0119]): acquire training data that is used for training of a first feature extractor (subblocks,
f
1
…
f
i
) and a second feature extractor (subblocks,
f
i
+
1
…
f
n
) (FIGS. 4-5B; input image 402, 502; [0061]; [0067], “soft supervision … training”); determine a model size of the second feature extractor (FIGS. 4-5B; 2nd feature extractor is the subblocks in trained model 410 or 510 without the subblocks,
f
1
…
f
i
(i.e., first feature extractor) selected by selector 520. Once selected subblocks decided, the size of 2nd feature extractor is determined); extract a first feature by inputting the training data to the first feature extractor (FIG. 5A, output of
f
i
); extract a second feature by inputting the first feature to the second feature extractor (FIG. 5A, output of block
f
n
; output of
f
i
is the input of 2nd feature extractor); and generate a trained model of the first feature extractor ([0043], “a self-distilled ViT … a simple and light-weight …”; [0060], “arranges ViT into transformer modules to obtain a light-weight plug-and-play DG approach for ViTs, referred to as self-distillation for VIT (SDVIT) …”; Abstract, “optimized based on a difference between the predicted class token and the sub-model prediction”; the final selected ,
f
1
…
f
i
or optimized vision transformer neural network) by training the first feature extractor in such a manner as to make the first feature closer to the second feature such that an inference is executed by the first feature extractor without using the second feature extractor (FIGS. 5A-5B; [0072]; equations (1)-(4); [0016], “predicting a class of the target medical image using the optimized vision transformer neural network”).
-Regarding claim 2. Sultana discloses the apparatus of claim 1. Sultana further discloses wherein the first feature extractor and the second feature extractor have an equal number of dimensions of features that are extracted (FIG. 4; FIG. 5A, tokens 506, 508; [0064], “A ViT is arranged as … any transformer block produces equi-dimensional features … each token has d dimensions”).
-Regarding claim 3. Sultana discloses the apparatus of claim 1. Sultana further discloses wherein pre-trained parameters in the first feature extractor are set as an initial value ([0055], “the present self-distillation approach seamlessly modularizes the architecture of ViTs and avoids introducing any new parameters”; [0062]; [0072]; FIGS. 4-5A; [0075]; Note: blocks 410 or 510 are pretrained” ).
-Regarding claim 4. Sultana discloses the apparatus of claim 1. Sultana further discloses comprising a storage configured to store a plurality of the second feature extractors, wherein the processor is configured to select the second feature extractor that is used for the training, from among the plurality of the second feature extractors (FIG. 5A; [0066], “extract a lot of knowledge from the training data is to learn many different models in parallel. The models should be as different as possible to minimize the correlations between their errors. The models can be made different by using different initializations or different architectures or different subsets of the training data”; [0075], “In one embodiment, the DeiT backbone is arranged as the intermediate blocks 510. In one embodiment, the CvT backbone is arranged as the intermediate blocks 510. In one embodiment, the T2T-ViT backbone is arranged as the intermediate blocks 510”).
-Regarding claim 5, Sultana discloses the apparatus of claim 1. Sultana further discloses wherein the processor is configured to determine the model size of the second feature extractor, based on at least one of a memory size, a calculation cost, and an inference accuracy ([0013], “capacity is smaller”; [0018], “optimized”; [0068], “a self-distilled vision transformer”; [0072], “minimizing the overall loss”; equations (2)-(3); FIGS. 5A-6D).
-Regarding claim 6. Sultana discloses the apparatus of claim 1. Sultana further discloses wherein each of the first feature extractor and the second feature extractor is a model using a Transformer configuration, or a model using an MLP-Mixer (FIGS. 4-5A).
-Regarding claim 7. Sultana discloses the apparatus of claim 1. Sultana further discloses extracting a feature of an intermediate layer of the first feature extractor as the first feature, and extracting a feature of an intermediate layer of the second feature extractor as the second feature (FIG. 5A).
-Regarding claim 8, Sultana discloses an information processing method comprising (Abstract; FIGS. 1-19; [0119]): acquire training data that is used for training of a first feature extractor (subblocks,
f
1
…
f
i
) and a second feature extractor (subblocks,
f
i
+
1
…
f
n
) (FIGS. 4-5A; input image 402, 502; [0061]; [0067], “soft supervision … training”); determine a model size of the second feature extractor (FIGS. 4-5A; 2nd feature extractor is the subblocks in trained model 410 or 510 without the subblocks,
f
1
…
f
i
(i.e., first feature extractor) selected by selector 520. Once selected subblocks decided, the size of 2nd feature extractor is determined); extract a first feature by inputting the training data to the first feature extractor (FIG. 5A, output of
f
i
); extract a second feature by inputting the first feature to the second feature extractor (FIG. 5A, output of block
f
n
; output of
f
i
is the input of 2nd feature extractor); and generate a trained model of the first feature extractor ([0043], “a self-distilled ViT … a simple and light-weight …”; [0060], “arranges ViT into transformer modules to obtain a light-weight plug-and-play DG approach for ViTs, referred to as self-distillation for VIT (SDVIT) …”; Abstract, “optimized based on a difference between the predicted class token and the sub-model prediction”; the final selected ,
f
1
…
f
i
or optimized vision transformer neural network) by training the first feature extractor in such a manner as to make the first feature closer to the second feature such that an inference is executed by the first feature extractor without using the second feature extractor (FIGS. 5A-5B; [0072]; equations (1)-(4); [0016], “predicting a class of the target medical image using the optimized vision transformer neural network”).
-Regarding claim 9, Sultana discloses a non-transitory computer readable medium including computer executable instructions, wherein the instructions, when executed by a processor (FIG. 19; [0113]), cause the processor to perform a method comprising (Abstract; FIGS. 1-19; [0119]): acquire training data that is used for training of a first feature extractor (subblocks,
f
1
…
f
i
) and a second feature extractor (subblocks,
f
i
+
1
…
f
n
) (FIGS. 4-5A; input image 402, 502; [0061]; [0067], “soft supervision … training”); determine a model size of the second feature extractor (FIGS. 4-5A; 2nd feature extractor is the subblocks in trained model 410 or 510 without the subblocks,
f
1
…
f
i
(i.e., first feature extractor) selected by selector 520. Once selected subblocks decided, the size of 2nd feature extractor is determined); extract a first feature by inputting the training data to the first feature extractor (FIG. 5A, output of
f
i
); extract a second feature by inputting the first feature to the second feature extractor (FIG. 5A, output of block
f
n
; output of
f
i
is the input of 2nd feature extractor); and generate a trained model of the first feature extractor ([0043], “a self-distilled ViT … a simple and light-weight …”; [0060], “arranges ViT into transformer modules to obtain a light-weight plug-and-play DG approach for ViTs, referred to as self-distillation for VIT (SDVIT) …”; Abstract, “optimized based on a difference between the predicted class token and the sub-model prediction”; the final selected ,
f
1
…
f
i
, or optimized vision transformer neural network) by training the first feature extractor in such a manner as to make the first feature closer to the second feature such that an inference is executed by the first feature extractor without using the second feature extractor (FIGS. 5A-5B; [0072]; equations (1)-(4); [0016], “predicting a class of the target medical image using the optimized vision transformer neural network”).
-Regarding claim 10, Sultana discloses the apparatus of claim 1. Sultana discloses wherein the second feature extractor is connected to downstream of the first feature extractor (FIG. 5A).
Response to Arguments
Applicant's arguments filed on 05/11/2026 have been fully considered but they are not persuasive. Applicant argues “Sultana fails to disclose or even suggest "generate a trained model of the first feature extractor by training the first feature extractor in such a manner as to make the first feature closer to the second feature such that an inference is executed by the first feature extractor without using the second feature extractor," as recited in amended independent claim 1” (Remarks, page 9, 1st paragraph), the examiner respectfully disagrees (See this office action, pages 3-4).
Regarding claim 1 and in response to applicant’s argument “Sultana discloses a vision transformer (ViT) self-distillation framework, which does not remove any portion of the model at inference time. To the contrary, inference continues to rely on the entire ViT, including all transformer blocks and the final classifier” (Remarks: page 9. 2nd paragraph), it is known that knowledge distillation is the procedure of transferring "knowledge" from a large model (the teacher) to a more compact one (the student), often being used in the context of model compression. When both models have the same architecture, this procedure is called self-distillation. During inference, the learned features will be represented using smaller model, i.e., the student model (See Pham et al, Revisiting Self-Distillation, arXiv:2206.08491v1 17 Jun 2022, Abstract; Sec. 1). In other word, the trained student model will be used at inference time for prediction or classification. As indicated by the applicant, Sultana discloses a vision transformer (ViT) self-distillation framework. Sultana discloses arranging ViT into transformer modules to obtain a light-weight plug-and-play DG approach for ViTs (Sultana: [0060]) and discloses self-distillation to transfer “knowledge” from a baseline vision transformer (ViT) (Sultana: FIG. 4, 404; FIG. 5, 510) to a smaller model (
f
1
…
f
i
)
through selector 520 (Sultana: FIG. 5, 520) assuming that
f
i
is the final selected sub-model after learning or optimization based difference between the predicted class token (through baseline vision transformer (ViT)) and the sub-model prediction (though the smaller model (
f
1
…
f
i
)
)
.
In this case, the optimized smaller model (
f
1
…
f
i
) will be used at inference time.
In response to applicant’s argument “Sultana does not disclose the claimed two-extractor architecture”, “the "sub-models" described in Sultana are intermediate layers of the same ViT model, rather than a second feature extractor separate from a first feature extractor” (Remarks: page 10, 1st paragraph), and “Sultana also fails to disclose determining a model size of a second feature extractor” (Remarks: page 10, 2nd paragraph), Sultana discloses acquire training data that is used for training of a first feature extractor (subblocks,
f
1
…
f
i
) and a second feature extractor (subblocks,
f
i
+
1
…
f
n
) (FIGS. 4-5B; input image 402, 502; [0061]; [0067], “soft supervision … training”); determine a model size of the second feature extractor (FIGS. 4-5B; 2nd feature extractor is the subblocks in trained model 410 or 510 without the subblocks,
f
1
…
f
i
(i.e., first feature extractor) selected by selector 520. Once selected subblocks decided, the size of 2nd feature extractor is determined) (See also this office action, page 3). Sultana discloses the vision transformer neural network having a sequence of feature extraction transformer blocks and a classifier component wherein each transformer block is a sub-model. The sub-models or feature extraction transformer blocks
f
i
+
1
…
f
n
is the second feature extractor. As stated above, once final
f
i
is determined, the number of block in the 2nd feature extractor is determined. Thus, the size of the 2nd feature extractor can be determined based on the number of block in the 2nd feature extractor and the determined size of each block in the 2nd feature extractor.
In response to applicant’s argument that Sultana fails to disclose training the first feature extractor in such a manner as to make the first feature closer to the second feature (Remarks, page 11), See this office action pages 3-4. Sultana discloses computing the difference between the final and randomly sampled intermediate classification tokens by comparing the KL divergence between their logit distributions using equations (2)-(4) wherein
Ƒ
(
x
)
is feature of output 2nd feature extractor or baseline ViT model and
Ƒ
j
(
x
)
is feature output of 1st feature extractor (Sultana: [0072]; FIG. 5A; equations (2)-(4)). KL divergence measures the difference the two feature outputs based on the distributions of the two features.
In response to applicant’s argument that Sultana fails to disclose "the second feature extractor is connected to downstream of the first feature extractor," as recited in claim 10, please see Sultana’s FIG. 5A, the output of the first feature extractor
f
1
…
f
i
is the input of the second feature extractor
f
i
+
1
…
f
n
.
Conclusion
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 nonprovisional extension fee (37 CFR 1.17(a)) 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 XIAO LIU whose telephone number is (571)272-4539. The examiner can normally be reached Monday-Thursday and Alternate Fridays 8:30-4:30.
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, Jennifer Mehmood can be reached at (571) 272-2976. 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.
/XIAO LIU/Primary Examiner, Art Unit 2664
Prosecution Insights