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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 04/17/2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
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 1-7 are rejected under 35 U.S.C. 103 as being unpatentable over Dwivedi et al. ("Duality Diagram Similarity: a generic framework for initialization selection in task transfer learning") in view of Cui et al. ("Large Scale Fine-Grained Categorization and Domain-Specific Transfer Learning").
Regarding Claim 1, Dwivedi teaches a training device(Fig. 3 shows the architecture for training a DNN) comprising: processing circuitry(Page 1, Abstract discloses DNN which would have to be stored and ran on a GPU/CPU inherently.) configured to acquire a plurality of training images(Page 8, Paragraph 1, “we select images from NYUv2, Taskonomy, and Pascal VOC dataset and evaluate the proposed methods on both, Taskonomy and Pascal VOC benchmark. We further compute the variance performing bootstrap by randomly sampling 200 images from the same dataset 100 times to compute similarity”, discloses acquiring images from a dataset used to train a DNN to compute similarity.); calculate a similarity between the shared feature space feature and the existing feature as an index by using the shared feature space feature for each of the plurality of training images and using a distance in the shared feature space based on distribution of the shared feature space features plotted in the shared feature space and the existing feature (Page 5, Paragraph 3, “The duality diagram of matrix X ∈ Rn×d1 is a triplet (X,Q,D) consisting of a matrix Q ∈ Rd1×d1 that quantifies dependencies between the individual feature dimensions, and a matrix D ∈ Rn×n that assigns weights on the observations, i.e., images in our case. Hence, a DNN representation for a set of n examples can be expressed by its duality diagram. By comparing duality diagrams of two DNNs we can obtain a similarity score.” , discloses calculating a similarity between feature spaces so a similarity score can be calculated.); select, as a base model, one of the trained models suitable for a purpose of training, on a basis of the index(Fig.1. “Duality Diagram Similarity (DDS): We apply DDS to compare features of a set of initialization options (encoders) with features of a new task to get model initialization rankings to select the encoder initialization for learning a new task.”, discloses selecting the best model for learning a new task.);perform retraining for the base model(Section 4 Our Approach and Section 4.1. disclose using transfer learning to learn tasks from the base model when training.); evaluate inference performance of the retrained base model and output the retrained base model(Section 6, Results, “we first report the comparison results of different similarity measures. After selecting the best similarity measure we apply it for identifying the representation type at different depth of the pre-trained encoder. Finally, we validate if the branching selection suggested using similarity measures gives the best transfer performance, by training models with different branching locations on NYUv2 and Pascal VOC datasets.”, discloses selecting the best similarity measure for the encoder to ensure the best transfer performance for the model so representations of source tasks are reused.)
Dwivedi does not explicitly teach calculate a shared feature space feature for each of the plurality of training images; acquire a pre-stored trained model and an existing feature corresponding to the pre-stored trained model;
Cui teaches calculate a shared feature space feature for each of the plurality of training images (Section 4.1, Paragraph 1, “Suppose we have a source domain S and a target domain T . We define the distance between two images s ∈Sand t ∈T as the Euclidean distance between their feature rep resentations: d(s, t)=∥g(s) − g(t)∥ (1) where g(·) denotes a feature extractor for an image. To bet ter capture the image similarity, the feature extractor g(·) needs to be capable of extracting high-level information from images in a generic, unbiased manner.”, discloses performing feature extraction from the source and target image to capture image similarity.) acquire a pre-stored trained model and an existing feature corresponding to the pre-stored trained model (Page 4114, Section 5.3, Paragraph 1,” We evaluate the proposed transfer learning method by pre-training the network on source domain from scratch, and then fine-tune on target domains for fine-grained visual categorization. Other than training separately on Im ageNet and iNat, we also train networks on a combined ImageNet + iNat dataset that contains 1,946,640 training images from 6,089 categories (i.e., 1,000 from ImageNet and 5,089 from iNat)”, discloses acquiring a pre-trained model from memory and fine tuning the model using existing features from the target domain.)
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the claimed invention as taught by Dwivedi with Cui in order to calculate a shared feature space and use a pre-stored model to fine tune the model. One skilled in the art would have been motivated to modify Dwivedi in this manner in order to offer an effective solution for domain-specific fine-grained visual categorization (FGVC) tasks (e.g., recognizing bird species or car make & model). (Cui, Abstract)
Regarding Claim 2, claim 2 is considered an apparatus claim substantially corresponding to claim 1. Please see the discussion of claim 1 above for a discussion of similar limitations. Furthermore, Dwivedi teaches a training system comprising a training device, an operation input device, a storage device(Page 1, Abstract discloses DNN which would have to be stored and ran on a GPU/CPU inherently.), and a display output device connected to each other(See Fig. S4 the images are output to a display device for presentation.),
Regarding Claim 3, claim 3 is considered a storage medium claim substantially corresponding to claim 1. Please see the discussion of claim 1 above for a discussion of similar limitations. Furthermore, Dwivedi teaches a non-transitory computer readable medium with an executable program stored thereon, wherein the program instructs a computer to perform (Page 1, Abstract discloses DNN which would have to be stored and ran on a GPU/CPU inherently.)
Regarding Claim 4, claim 4 is considered a method claim substantially corresponding to claim 1. Please see the discussion of claim 1 above for a discussion of similar limitations. Furthermore, Dwivedi teaches an information processing method for a training device(See Fig. 1),
Regarding Claim 5, Dwivedi teaches a training device(Fig. 3 shows the architecture for training a DNN) comprising: processing circuitry(Page 1, Abstract discloses DNN which would have to be stored and ran on a GPU/CPU inherently.) configured to acquire a plurality of training images(Page 8, Paragraph 1, “we select images from NYUv2, Taskonomy, and Pascal VOC dataset and evaluate the proposed methods on both, Taskonomy and Pascal VOC benchmark. We further compute the variance performing bootstrap by randomly sampling 200 images from the same dataset 100 times to compute similarity”, discloses acquiring images from a dataset used to train a DNN to compute similarity.); calculate a similarity between the shared feature space feature and the existing feature as an index by using the shared feature space feature for each of the plurality of training images and using a distance in the shared feature space based on distribution of the shared feature space features plotted in the shared feature space and the existing feature (Page 5, Paragraph 3, “The duality diagram of matrix X ∈ Rn×d1 is a triplet (X,Q,D) consisting of a matrix Q ∈ Rd1×d1 that quantifies dependencies between the individual feature dimensions, and a matrix D ∈ Rn×n that assigns weights on the observations, i.e., images in our case. Hence, a DNN representation for a set of n examples can be expressed by its duality diagram. By comparing duality diagrams of two DNNs we can obtain a similarity score.” , discloses calculating a similarity between feature spaces so a similarity score can be calculated.); select, as a base model, one of the trained models suitable for a purpose of training, on a basis of the index(Fig.1. “Duality Diagram Similarity (DDS): We apply DDS to compare features of a set of initialization options (encoders) with features of a new task to get model initialization rankings to select the encoder initialization for learning a new task.”, discloses selecting the best model for learning a new task.);perform retraining for the base model(Section 4 Our Approach and Section 4.1. disclose using transfer learning to learn tasks from the base model when training.);
Dwivedi does not explicitly teach calculate a shared feature space feature for each of the plurality of training images; acquire a pre-stored trained model and an existing feature corresponding to the pre-stored trained model;
Cui teaches calculate a shared feature space feature for each of the plurality of training images (Section 4.1, Paragraph 1, “Suppose we have a source domain S and a target domain T . We define the distance between two images s ∈Sand t ∈T as the Euclidean distance between their feature rep resentations: d(s, t)=∥g(s) − g(t)∥ (1) where g(·) denotes a feature extractor for an image. To bet ter capture the image similarity, the feature extractor g(·) needs to be capable of extracting high-level information from images in a generic, unbiased manner.”, discloses performing feature extraction from the source and target image to capture image similarity.) acquire a pre-stored trained model and an existing feature corresponding to the pre-stored trained model (Page 4114, Section 5.3, Paragraph 1,” We evaluate the proposed transfer learning method by pre-training the network on source domain from scratch, and then fine-tune on target domains for fine-grained visual categorization. Other than training separately on Im ageNet and iNat, we also train networks on a combined ImageNet + iNat dataset that contains 1,946,640 training images from 6,089 categories (i.e., 1,000 from ImageNet and 5,089 from iNat)”, discloses acquiring a pre-trained model from memory and fine tuning the model using existing features from the target domain.)
It would have been obvious to one of ordinary skill in the art before the effective filing date to modify the claimed invention as taught by Dwivedi with Cui in order to calculate a shared feature space and use a pre-stored model to fine tune the model. One skilled in the art would have been motivated to modify Dwivedi in this manner in order to offer an effective solution for domain-specific fine-grained visual categorization (FGVC) tasks (e.g., recognizing bird species or car make & model). (Cui, Abstract)
Regarding Claim 6, claim 6 is considered a storage medium claim substantially corresponding to claim 1. Please see the discussion of claim 5 above for a discussion of similar limitations. Furthermore, Dwivedi teaches a non-transitory computer readable medium with an executable program stored thereon, wherein the program instructs a computer to perform (Page 1, Abstract discloses DNN which would have to be stored and ran on a GPU/CPU inherently.)
Regarding Claim 7, claim 7 is considered a method claim substantially corresponding to claim 1. Please see the discussion of claim 5 above for a discussion of similar limitations. Furthermore, Dwivedi teaches an information processing method for a training device(See Fig. 1),
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HAN D HOANG whose telephone number is (571)272-4344. The examiner can normally be reached Monday-Friday 8-5.
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, JOHN M VILLECCO can be reached at 571-272-7319. 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.
/HAN HOANG/Primary Examiner, Art Unit 2661
Prosecution Insights