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 .
Election/Restrictions
Applicant’s election without traverse of Species I (claims 1–2, 4–5, 7–8, and 10–11) in the reply filed on August 6th, 2026, is acknowledged. The application has pending claims 1–12 (withdrawn claims 3, 6, 9 and 12 are withdrawn from further consideration).
Drawings
The drawings are objected to because:
FIG. 10: flowchart box S22 uses the label "LEARNING UNIT 7", whereas boxes S21, S24, and S25 (and FIG. 1/7 and paragraph [0026]) consistently label element 7 as "TRAINING UNIT 7".
FIG. 10: flowchart box S25 recites "...AND UPDATES WEIGHS IN SECOND MODEL". The word WEIGHS is a misspelling and should be corrected to WEIGHTS.
Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. The figure or figure number of an amended drawing should not be labeled as “amended.” If a drawing figure is to be canceled, the appropriate figure must be removed from the replacement sheet, and where necessary, the remaining figures must be renumbered and appropriate changes made to the brief description of the several views of the drawings for consistency. Additional replacement sheets may be necessary to show the renumbering of the remaining figures. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Specification
The abstract of the disclosure is objected to because it contains the following grammatical informalities: “-- takes a resolution-converted data as an input --” should be corrected to “-- takes resolution-converted data as an input --”; and “a part of activation output” should be corrected to “a part of an activation output”. A corrected abstract of the disclosure is required and must be presented on a separate sheet, apart from any other text. See MPEP § 608.01(b).
Claim Objections
Claims 1-2, 7-8, and 10-11 are objected to because of the following informalities:
In claims 1, 7, and 10: “a resolution-converted data” should be corrected to “resolution-converted data”; and “a part of activation output” should be corrected to “a part of an activation output”.
In claim 2: the “wherein the one or more processors configured to execute the software instructions to...” should be amended to “wherein the one or more processors are configured to execute the software instructions to...”.
In claims 2, 8, and 11: “the part of activation after size conversion” should be corrected to “the part of the activation after size conversion” or equivalent grammatically correct language; and “input concatenation result” or “inputting concatenation result” should be corrected to “input the concatenation result” or “inputting the concatenation result,” respectively.
Appropriate correction is required.
Claim Rejections - 35 USC § 112(b)
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1–2, 4–5, 7–8, and 10–11 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 1, 7, and 10 recite the limitation “convert a resolution of the data to the predicted resolution” in claims. There is insufficient antecedent basis for this limitation in the claim, as no earlier limitation establishes a "predicted resolution", only "a minimum resolution" is previously recited. It is unclear whether "the predicted resolution" is intended to refer back to the recited "minimum resolution" or introduces an additional, undefined resolution value. Claims 2, 4, 5, 8, and 11 depend from claims 1, 7 and 10 and inherit this deficiency.
Claim 4 recites the limitation “the given layer of the second model” in claim. There is insufficient antecedent basis for this limitation in the claim. Claim 4 depends from claim 1. Claim 1 introduces “a given layer in the first model,” but does not introduce a “given layer in the second model”. It is unclear whether this limitation refers to the previously recited given layer in the first model, a corresponding layer in the second model, or a separately selected layer of the second model. To overcome this rejection, claim 4 should be amended to clarify the initial input layer and provide an explicit antecedent reference (e.g., "weights from an initial layer to the given layer in the first model are the same as initial weights from an initial layer to a corresponding layer in the second model").
Claim 4 further recites that “weights from a first layer to the given layer are the same as weights from a first layer to the given layer of the second model as an initial state”. This language does not clearly identify the model associated with the first-recited set of weights or clearly establish which layers of the respective models are being compared. Additionally, it is unclear whether “as an initial state” modifies the weights of the first model, the weights of the second model, or the state in which the first model is trained. For clarification, applicant may consider amending the claims to specify that “the first model is trained using, as an initial state, weights from the first layer through the given layer of the first model that are the same as weights from the first layer through a corresponding layer of the second model.”
Claim 5 contains the same indefinite language. Although claim 2 recites “a layer in the second model corresponding to the given layer,” claim 5 instead refers to “the given layer of the second model.” It is therefore unclear whether claim 5 is referring to the corresponding layer recited in claim 2 or to another layer of the second model. The ambiguity concerning the two sets of weights and the phrase “as an initial state” also applies to claim 5.
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 (i.e., changing from AIA to pre-AIA ) 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, 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 1–2, 7–8 and 10–11 are rejected under 35 U.S.C. §103 as being unpatentable over Zhu (Zhu et al. (2021). Dynamic Resolution Network. arXiv.Org), as provided by Applicant's disclosure filed on 11/18/2024, in view of Zang (Yang et al. (2020). Resolution Adaptive Networks for Efficient Inference. arXiv.Org).
Regarding claim 1, Zhu teaches an inference apparatus comprising:
a memory that stores software instructions, and
one or more processors configured to execute the software instructions to
( Abstract, § 4.1, § 4.4: Zhu discloses computer code implementing the Dynamic Resolution Network (“DRNet”), implementation of the framework in PyTorch on NVIDIA Tesla V100 GPUs, and execution of the inference operations model on an Intel Xeon Gold 6151 CPU. )
take data of a certain resolution as an input and predict a minimum resolution, among multiple resolution candidates, by which a label of the data can be inferred with a predetermined accuracy, using a first model including multiple layers,
( Abstract, Fig. 1-2, § 3.1 and Eqs. (1)–(2), Appendix A.1: Zhu receives an input image X having an input resolution and employs a resolution predictor as a first model. The resolution predictor includes multiple convolutional and fully connected layers and produces a probability distribution over multiple candidate resolutions r1, r2, …, rm. Zhu explains that the predictor learns and selects the smallest or minimal resolution that is performance-sufficient (i.e., that retains or exceeds the original recognition accuracy for the input image); thereby predicting a minimum resolution among multiple candidates by which an image label can be inferred with a predetermined accuracy. )
convert a resolution of the data to the predicted resolution, and
( Abstract, Fig. 2, § 3.1: Zhu teaches that, after the resolution predictor selects the resolution, the original input image is resized to the selected or predicted resolution. )
take a resolution-converted data as an input and infer a label of the data using a second model including multiple layers
( Fig. 2, § 3.1, § 3.2: Zhu teaches resizing the original input image to the resolution selected by the resolution predictor and supplying only the resized image to a large image classifier. Zhu identifies the large classifier as, for example, a ResNet or EfficientNet and depicts the classifier as including multiple convolutional, batch-normalization, adaptive-global-average-pooling, and fully connected layers. Zhu further explains that the base classifier F receives the resized image and outputs probability predictions y for image classification, thereby inferring a label of the resolution-converted data using a second model including multiple layers. )
However, Zhu does not expressly teach inferring the label using part of an activation output from a given layer in the first model when performing classification with a second multilayer model, where Yang teaches:
take generated data as an input and infer a label of the data using a second model including multiple layers and a part of activation output from a given layer in the first model.
( § 3.2, Fig. 2 and § 3.3.2, Fig. 3 and § 3.3.2: Yang teaches that an Initial Layer generates multiple base feature maps at different spatial-resolution scales for corresponding multilayer subnetworks. When Sub-network 1 does not produce a sufficiently confident classification, Sub-network 2 receives and processes its corresponding generated larger-scale base feature map (x02,2) to further classify the input sample. Sub-network 2 also receives intermediate feature outputs from Sub-network 1. In particular, Yang discloses that the output (xi1,1) from each of multiple layers of Sub-network 1 is propagated to Sub-network 2 for reuse and is fused with the generated base-feature data in the Fusion Blocks of Sub-network 2. Sub-network 2 thereafter processes the fused features through its multiple layers and classifiers to infer the classification label. Accordingly, Yang teaches using a second model including multiple layers to infer a label from generated feature-map data and part of an activation output from a given layer of the first model. )
It would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to modify Zhu’s DRNet to reuse and fuse, in the downstream classifier, intermediate features already produced by the resolution predictor, as taught by Yang, to avoid duplicative feature extraction and improve computational efficiency. Because both systems process the same input using compatible CNN feature maps, the modification would have been a predictable application of a known feature-reuse technique, with a reasonable expectation of reducing computation while maintaining classification accuracy.
Regarding claim 2, Zhu [as modified by Yang] teaches the inference apparatus according to claim 1, wherein the one or more processors configured to execute the software instructions to
convert a size of the part of activation according to the predicted resolution, and
( Yang > §3.2, §3.3.2, Figs. 2 and 3(b): Yang propagates an intermediate activation output from a layer of a preceding sub-network to a subsequent sub-network and processes the reused activation using an Up-Conv layer comprising a Regular-Conv layer and bilinear upsampling, thereby converting the activation to the same spatial size as the activation in the receiving sub-network. In Zhu [as modified by Yang], the receiving feature-map size corresponds to Zhu’s predicted input resolution; therefore, the reused activation is size-converted according to the predicted resolution. )
when performing inference, concatenate, in a channel direction, an activation output from a layer in the second model corresponding to the given layer and the part of activation after size conversion, and input concatenation result to a next layer in the second model.
( Yang > §1.1, §3.3.2, Fig. 2 and 3(b); Appendix A: Yang’s Fusion Block concatenates a corresponding-layer activation from the receiving sub-network with the size-converted activation from the preceding sub-network through a dense connection and supplies the concatenated feature tensor to the following convolutional layer. Because the spatial dimensions are first made equal and the resulting feature channels are contributed by both sub-networks, the disclosed concatenation is along the channel direction. )
Regarding claims 7–8 and 10–11, the rationale provided in the rejection of claims 1–2 is incorporated herein. In addition, the apparatus of claims 1–2 corresponds to the method of claims 7–8, as well as the non-transitory computer readable recording medium of claims 10–11, and performs the steps disclosed herein. Therefore, the claims are all rejected.
Claims 4–5 are rejected under 35 U.S.C. §103 as being unpatentable over Zhu [as modified by Zang], in view of Gebre (Gebre et al, US 2019/0156205 A1, 2019).
Regarding claim 4, Zhu [as modified by Yang] teaches the inference apparatus according to claim 1, wherein
However, Zhu [as modified by Yang] does not expressly teach where Gebre teaches:
the first model is a model trained with a state such that weights from a first layer to the given layer are the same as weights from a first layer to the given layer of the second model as an initial state.
( Gebre, [0059–0062], [Fig. 3-4], [0078–0079]: Gebre teaches initializing a neural-network model by copying weights from the input and hidden layers of another, previously trained neural-network model to the corresponding layers of the model being initialized. Gebre further teaches copying all weights of the input and hidden layers so that the corresponding weights initially have the same values, and thereafter training or fine-tuning the initialized model. Gebre’s source “first model 402” corresponds to the claimed second model, and Gebre’s receiving “second model 404” corresponds to the claimed first model; thereby copying all corresponding input- and hidden-layer weights includes copying the weights continuously from the first layer through the claimed given layer, thereby providing the recited same-weight initial state. )
It would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, to initialize Zhu’s resolution-prediction model with corresponding early-layer weights of the Zhu [as modified by Yang]'s classifier, as taught by Gebre, because both neural networks process the same image data and their early layers learn common low-level visual features. Gebre teaches that such weight transfer shortens training time and reduces computational burden. The modification therefore would have been a predictable use of a known transfer-learning technique, with a reasonable expectation of success.
Regarding claim 5, which depends from claim 2 rather than claim 1, the rationale provided in the rejection of claim 4 is still incorporated herein. In addition, the same citations above apply identically and performs the steps disclosed herein. Therefore, the claim is rejected.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KEN KUDO whose telephone number is (571)272-4498. The examiner can normally be reached M-F 8am - 5pm.
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, Vincent Rudolph can be reached at 571-272-8243. 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.
KEN KUDO
Examiner
Art Unit 2671
/KEN KUDO/Examiner, Art Unit 2671
/VINCENT RUDOLPH/Supervisory Patent Examiner, Art Unit 2671
Prosecution Insights