.Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
Claims 1-12 are presented for examination.
Claim 13 was cancelled.
Claims 1, 6 and 8 were amended.
This is a Non-Final Action.
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 05/01/2026 has been entered.
Response to Arguments
Applicant's arguments filed 03/06/2026 have been fully considered but they are not persuasive.
Applicant’s arguments with respect to claim(s) 1-12 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Specifically, claim 13 as a whole was not incorporated into claims 1 and 6; Claim limitations in relation to central node, transmission, first processor and second processor architecture was missing. Furthermore, in view of updated search examiner found a prior art that would read on the current amendments to the claim.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102 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, 3-4, 7-8, 10-11 are rejected under 35 U.S.C. 103 as being unpatentable over Shen et al. (US 9,996,768) in view of Choi et al. (US 2018/0268292) further in view of Raychaudhuri et al. (US 2022/0292819)
1. Shen teaches, A learning apparatus comprising a processor configured to:
divide target data into pieces of partial data (Col 1: lines 21-24- teaches patches are generated from image .. randomly , Shen);
input the pieces of partial data into a first network model to output a first prediction result (Col 1: lines 24-32 - teaches Activations are calculated … for each of the plurality of patches using a neural network (per-patch logits = first prediction result), Shen);
calculate a first confidence indicating a degree of contribution to the first prediction result, for each of the pieces of partial data (Col 2: lines 46-61- teaches outputs … are aggregated … to generate an image descriptor … indicative of a confidence …” (weight per patch), the patch weights fulfil the first-confidence element, Shen); and
update a parameter of the first network model, based on the first prediction result, the first confidence (Claim 7, Fig 8 and 9 - teaches patch weights propagate gradients into CNN (see Procedure 800/900), Shen).
Shen does not explicitly teach,
input the target data into a second network model to output a second prediction result;
calculate a second confidence indicating a degree of contribution to the second prediction result, fora region corresponding to each of the pieces of partial data in the target data; and
update a parameter of the first network model, the second prediction result and the second confidence to generate a trained first network model, wherein the trained first network model includes a feature extractor configured to extract a feature from partial data of target data and a predictor configured to perform processing corresponding to a task on the feature, and wherein the processor is further configured to receive a plurality of features extracted from pieces of partial data by the feature extractor deployed in a plurality of processing nodes, perform ensemble processing on the plurality of features, and input the features subjected to the ensemble processing into the predictor to generate an inference result.
However, Choi teaches,
input the target data into a second network model to output a second prediction result (Paragraphs 5, 36-40 - teaches teacher network: input a plurality of images into the Faster R-CNN (global model) – supplies the second model & second prediction, Choi);
calculate a second confidence indicating a degree of contribution to the second prediction result, fora region corresponding to each of the pieces of partial data in the target data (Paragraph 28 & Claim 7 - teaches confidence-weight binary activation loss … when a confidence of the teacher model is greater.. (per-box/region score); the teacher’s per-region score is the second confidence, Choi); and
update a parameter of the first network model, the second prediction result and the second confidence (Abstract and Claim 1 - teaches a loss layer combines student logits & confidences and teacher logits & confidences during back-prop (confidence-weighted train intermediate layers), Choi);
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which said subject matter pertains to allow Shen’s invention to be combined with Choi’s invention because they are both in the same field endeavor of neural network, wherein Shen supplies the student-side mechanics: patch subdivision, per-patch inference, and a patch level confidence signal that already back-propagates; and Choi teaches teacher-side mechanics: whole-image second network, per-region confidence and training loss that uses both student & teacher predictions plus their confidences to update the shared parameters; therefore teachings in Shen and Choi would naturally combine a patch-based student with a confidence-weighted teacher.
Raychaudhuri teaches,
to generate a trained first network model, wherein the trained first network model includes a feature extractor configured to extract a feature from partial data of target data (Paragraph 31 – teaches first they use CNN network to extract feature maps from the input and generate region proposals (RPs) for every object; Paragraph 36 – teaches edge servers 14a-14c perform one or more computer vision processes on the RP boxes 22 (e.g., using one or more neural network or other computer vision processes) to generate object metadata 24, Raychaudhuri) and a predictor configured to perform processing corresponding to a task on the feature (Paragraph 31 - teaches Second, they use a CNN network to evaluate each RP and output the fine-grained result such as the classified object type or the key body points of a person, Raychaudhuri), and wherein the processor is further configured to receive a plurality of features extracted from pieces of partial data by the feature extractor deployed in a plurality of processing nodes (Paragraph 36 – teaches edge servers 14a-14c perform one or more computer vision processes on the RP boxes 22 (e.g., using one or more neural network or other computer vision processes) to generate object metadata 24, Raychaudhuri), perform ensemble processing on the plurality of features (Paragraph 37 - teaches these partial results are finally integrated at the mobile side to render the final result, Raychaudhuri), and input the features subjected to the ensemble processing into the predictor to generate an inference result (Paragraph 36 – teaches the mobile device 12 then combines the object metadata 24 with the input frame 16 to produce a final result 26 which can be displayed to the user; Claim 1 – teaches processing at the mobile device the object metadata and the input frame to generate a final image, Raychaudhuri).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which said subject matter pertains to modify the patch-based neural network system of Shen, as trained using Choi’s confidence weighted teacher/student framework, to use Raychaudhuri’s distributed partial inference architecture because Raychaudhuri teaches the partitioning high-resolution vision data and offloading the partitions to multiple edge servers reduces end-to-end latency while maintaining accuracy. Thus, a POSITA would have been motivated to deploy the trained feature-extraction processing across multiple nodes and integrate the resulting partial features/inference outputs to improve speed and efficiency for high resolution image inference.
3. The combination of Shen, Choi and Raychaudhuri teach, The apparatus according to claim 1, wherein the processor calculates an objective function based on the first prediction result, the second prediction result, the first confidence, and the second confidence, and updates the parameter such that a value of the objective function is optimized ( Fig 14, Paragraphs 56-57- teaches overall distillation loss (equations 3-5) further explaining the Binary-Activation loss: a positive gradient is applied when a confidence of the teach model is greater than a confidence of the student model, both confidences explicitly modulate the loss term; Further teaching, the sign of the gradient drives back-prop to previous layers, thereby updating the shared parameters to minimize the confidence-weighted loss, Choi).
4. The combination of Shen, Choi and Raychaudhuri teach, The apparatus according to claim 1, wherein the processor is further configured to:
generate intermediate data regarding feature extraction of the pieces of partial data from the first network model ( Col 5: lines 40-60, Fig 4 & 5 teaches - for every image patch the CNN’s backbone produces an activation feature vector (intermediate data), Shen);
weight the intermediate data, based on the first confidence (Col 6: lines 32-48 - teaches a learned patch weight wi (the first confidence) is multiplied with each patch feature vector before aggregation , Shen);
perform ensemble processing to the weighted intermediate data (Col 6: lines 49-65 teaches – aggregator 430 sums/ averages the weighted patch vectors to build a single image descriptor. This is ensemble fusion of multiple patch features , Shen); and
output the first prediction result, based on the intermediate data after the ensemble processing (Col 6: lines 49-65 teaches – the aggregated descriptor is passed to classifier head 440 to output the final prediction for the image , Shen).
7. The combination of Shen, Choi and Raychaudhuri teach, The apparatus according to claim 1, wherein each of the pieces of partial data corresponds to at least one of a region partially overlapping in the target data, a region not overlapping in the target data, a region randomly selected from the target data, and a region regarding a prediction target included in the target data (Col 5: lines 40-60 and fig 4 – teaches the images is divided into a plurality of randomly generated patches… the patches may overlap, Procedure 500, grid-based non overlapping variant; Paragraphs 38-42 and Fig 2 – teaches teacher Faster R-CNN outputs object-bounding boxes; each box region is cropped and used to supervise the student, Choi).
Claims 8-12 are similar to claims 1-6 hence rejected similarly
All the limitations of claim 1 are taught above
Claims 2, 6 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Shen et al. (US 9,996,768) in view of Choi et al. (US 2018/0268292) and Raychaudhuri et al. (US 2022/0292819) further in view of Chakraborty (US 2020/0167772)
2. The combination of Shen, Choi and Raychaudhuri teach, wherein the first network model operates on pieces of partial data (Abstract, Fig 4, Col 5: lines 48-56 teaches – Generating patches … from an image and feeds each patch to a CNN column, Shen).
The combination of Shen and Choi do not explicitly teach,
… first and the second network model are partially or entirely identical in model structure and share part or an entirety of the parameter.
However, Chakraborty teaches,
… first and the second network model are partially or entirely identical in model structure and share part or an entirety of the parameter (Fig 3, Paragraph 83-84 - teaches network architecture includes two identical sub-networks processing paired inputs; the convolution networks are twin networks with shared weights including shared weights 314/514 Chakraborty) .
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which said subject matter pertains to allow Shen’s invention to be combined with Chakraborty because both prior arts work to reduce GPU memory by processing small patches, using Chakraborthy/Shared-weight twin path lets the same parameters serve both a patch view and a full-image view, shrinking model size even more. Adding a second, weight-tied branch to Shen’s architecture requires only duplicating the graph and declaring the variables shared – no new hardware needed.
Claim 6 is similar to the combination of claim 1 and 2 hence rejected similarly.
Claim 9 is similar to claim 2 hence rejected similarly.
All the limitation of claim 1 is taught above.
Claim 5 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Shen et al. (US 9,996,768) in view of Choi et al. (US 2018/0268292) and Raychaudhuri et al. (US 2022/0292819) further in view of Tung et al. (US 2020/0302295)
5. The combination of Shen, Choi and Raychaudhuri do not teach, wherein the first confidence is calculated based on saliency or attention of intermediate data of the first network model, and the second confidence is calculated based on Saliency or attention of intermediate data of the second network model.
However Tung teaches,
wherein the first confidence is calculated based on saliency or attention of intermediate data of the first network model, and the second confidence is calculated based on Saliency or attention of intermediate data of the second network model (Paragraphs 24-25, 56-57 – teaches computing a student activation map A, intermediate feature tensor and then derives a pair-wise similarity matrix Gs=f(As), which expressly describes an attention-transfer signal that guides training. The similarity/attention values are used in the loss therefore act as the student confidence, Tung).
It would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which said subject matter pertains to allow Shen’s invention to be combined with Tung because both patents attack problems of deploying CNNs on resource-constrained devices. Tung explicitly cites attention transfer as a way to capture richer supervisory signal – exactly the motivation Shen’s dual-confidence design needs.
Claim 12 is similar to claim 5 hence rejected similarly.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AMRESH SINGH whose telephone number is (571)270-3560. The examiner can normally be reached on Monday-Friday 8am-5pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Mariela D. Reyes can be reached on (571) 272-9767. 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.
/Amresh Singh/
Primary Examiner, Art Unit 2159
Prosecution Insights