DETAILED ACTION
Remarks
This office action is issued in response to communication filed on 4/15/2026. Claims 1-2,4-12 and 14-22 are pending in this Office Action.
Objection of claims 2 and 12 has been withdrawn by the examiner.
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 Arguments
Applicant’s arguments filed 4/15/2026 with respect to the 103 rejection have been considered and are not persuasive. The examiner respectfully traverses applicant’s arguments.
Applicant argues: “Kang does not teach any multi-task distillation framework where different teachers perform different classification tasks. As the Examiner acknowledges, Kang teaches selecting multiple teacher models to train a student model. However, Kang's teacher models and student model are all "configured to recognize the same target." (Kang, par. [0045].) Kang's Equation 3 (see par. [0074]) compares the single student output PStudent against outputs of teacher model i and teacher model j, but all models obtain classifications for the same set of classes. The loss function evaluates the same student output against multiple teachers that all perform the same classification task. There is no "plurality of sets of classes" and no per-set-of-classes comparison as recited in the claims.”(Applicant’s argument at page 10-11)
Examiner response: The examiner respectfully disagrees because one cannot show non-obviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). The examiner relies on Fukuda , NOT Kang for teaching “plurality set of classes” . Specifically, Fukuda par [006] teaches plurality of teacher models, each of which is specialized for different one of the plurality of domains. Fukuda par [0013] teaches plurality of teacher models and student model are image processing models and the plurality of domains has difference in color modes of an input image signal.
Applicant argues: “During the Examiner interview on April 7, 2026, the Examiner has indicated that the "plurality of sets of classes" can be met by the different levels of accuracy of different teacher models. Applicant respectfully submits that this interpretation would be unreasonable for several reason..”(Applicant’s argument at page 11)
Examiner response: The examiner respectfully disagrees. Although different levels of accuracy of different teacher models under broadest reasonable interpretation, can be interpreted as different classes. The “plurality set of classes” is not expressly discloses by Kang and the examiner relies on Fukuda for the teaching of this feature as indicates in the above response.
Applicant argues: “Fukuda does not cure the deficiency of Kang with respect to the combined loss function. The Examiner has alleged that Fukuda teaches "a plurality of teacher models, each of which is specialized for different one of the plurality of the domains." (Office Action, p. 4.) However, Fukuda's teachers are specialized for different input domains (e.g., narrowband vs. broadband audio, or RGB vs. B&W images), not different classification tasks with different sets of classes. Fukuda explicitly states that "the number of the members in the class set for the mixed-band model and the members in the class set are typically same as those of the teacher models." (Fukuda, par. [0064].) Fukuda's "domains" refer to different input modalities, not different classification tasks with different sets of output classes. All of Fukuda's teachers classify into the same HMM states or image classes.”(Applicant argument at page 12)
Examiner response: The examiner respectfully disagrees because one cannot show non-obviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). The examiner relies on Kang , NOT Fukuda for teaching “the combined loss function” . Specifically, Kang par [0077] teaches the model training may calculate a loss based on output data of at least one teacher model and output data of student model and train the student model to reduce the loss. In addition, it is noted that the features upon which applicant relies (i.e., “different classification tasks with different sets of output classes”) are not recited in the rejected claim(s). Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993).
Applicant’s remaining arguments with respect to claims are substantially encompassed in the argument above, therefore examiner responds with the same rationale as stated above.
For at least the foregoing reasons, the examiner maintains prior art rejections.
Allowable Subject Matter
Claims 9-10 and 18-19 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
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.
Claims 1-2,4-5, 11-12,14 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Kang et al.( US Patent Application Publication 2017/0083829 A1, hereinafter “Kang”) in view of Fukuda et al.(US Patent Application Publication 2019/0378006 A1, hereinafter “Fukuda”) , in view of Iscen et al.(US Patent Application Publication 2024/0320493 A1, hereinafter “Iscen”)
As to claim 1, Kang teaches a method comprising: obtaining a plurality of target outputs (bold emphasis added. Kang par [0073] teaches selecting 2 teacher models to train student model),
[each target output of the plurality of target outputs comprising a respective classification of a training input among a respective set of classes of a plurality of sets of classes, wherein the classification among the respective set of classes obtained using a respective teacher machine learning model (MLM) of a plurality of teacher MLMs], and wherein the training input comprises a representation of one or more object (Kang par 115] teaches A model refers to a neural network capable of detecting objects from the target data, or classifying or clustering the objects); and
using the training input, a combined loss function and the plurality of target outputs to train a student MLM to classify the one or more objects among each of the plurality of sets of classes, (Kang par [0076] teaches the student model 320 is trained based on output data of the selected teacher models. Kang par [0073] teaches the model training apparatus may trin the student model 320 to reduce the loss calculated through equation 3)
wherein the combined loss function characterizes similarity of (i) classifications generated by the student MLM [among said each of the plurality of sets of classes] and (ii) classifications obtained using corresponding teacher MLMs [among said each of the plurality of sets of classes]; (Kang par [0077] teaches the model training may calculate a loss based on output data of at least one teacher model and output data of student model and train the student model to reduce the loss) and
[causing the trained student model to be configured to control a driving path of a vehicle modified in view of one or more additional objects identified, by the trained student model, in an environment of the vehicle].
Kang fails to expressly teach each target output of the plurality of target outputs comprising a respective classification of a training input [among a respective set of classes of a plurality of sets of classes, wherein the classification among the respective set of classes obtained using a respective teacher machine learning model (MLM) of a plurality of teacher MLMs.
However, in an analogous art directed to machine learning, Fukuda teaches each target output of the plurality of target outputs comprising a respective classification of a training input among a respective set of classes of a plurality of sets of classes, wherein the classification among the respective set of classes obtained using a respective teacher machine learning model (MLM) of a plurality of teacher MLMs.(Fukuda par [006] teaches plurality of teacher models, each of which is specialized for different one of the plurality of domains. Fukuda par [0013] teaches plurality of teacher models and student model are image processing models and the plurality of domains has difference in color modes of an input image signal)
Therefore , it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kang and Fukuda to achieve the claimed invention. One would have been motivated to make such combination to reduce computational resources (Fukuda par [0007])
Kang and Fukuda fail to expressly teach causing the trained student model to be configured to control a driving path of a vehicle modified in view of one or more additional objects identified, by the trained student model, in an environment of the vehicle.
However, Iscen teaches causing the trained student model to be configured to control a driving path of a vehicle modified in view of one or more additional objects identified, by the trained student model, in an environment of the vehicle.(Iscen par [0075] teaches a robotic agent of autonomous vehicle may use the output of a computer vision task as to determine a suitable trajectory to follow or an alteration to make its current trajectory and adapt it motion accordingly)
Therefore , it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kang , Fukuda and Iscen to achieve the claimed invention. One would have been motivated to make such combination to enhance recognition capabilities.(Iscen par [0026])
As to claim 2, Kang , Fukuda and Iscen teach the method of claim 1, wherein the representation of the one or more objects comprises at least one of a camera image of the one or more objects, a lidar image of the one or more objects, or a radar image of the one or more objects.(Iscen par [0011] teaches one or more objects recognized in the image)
As to claim 4, Kang, Fukuda and Iscen teach the method of claim 1 wherein the classification of the training input among the respective set of classes comprises a set of values, wherein each value of the set of values characterizes a likelihood of the one or more objects belonging to a corresponding class of the respective set of classes.(Iscen par [0073] teaches image processing task can be image classification where the output is a set of scores , each score corresponding to a different object class representing the likelihood that the one or more images depict an object belong to the object class)
As to claim 5, Kang , Fukuda and Iscen teach the method of claim 4, wherein the training input comprises a plurality of frames, wherein each of the plurality of frames depicts the one or more objects at a respective one of a plurality of times. (Iscen par [0061] teaches training dataset include image data , video data)
Claims 11-12 and 14 merely recite a system to perform the method of claims 1-2 and 4 respectively. Accordingly, Kang , Fukuda and Iscen teach every limitation of claims 11-12 and 14 as indicates in the above rejection of claims 1-2 and 4 respectively.
Claim 20 merely recites a non-transitory computer readable medium storing instructions when executed by a processor, performs the method of claim 1. Accordingly, Kang , Fukuda and Iscen teach every limitation of claim 20 as indicates in the above rejection of claim 1.
Claims 6 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Kang , Fukuda and Iscen and further in view of Pu Wang et al.(US Patent Application Publication 2023/0305136 A1, hereinafter “Pu”)
As to claim 6, Kang , Fukuda and Iscen teach the method of claim 5 but fail to teach wherein using the training input to train the student MLM comprises: using a first neural network (NN) to obtain a plurality of sets of object embeddings, wherein each of the plurality of sets of object embeddings is associated with a respective time of the plurality of times ; and using a second NN to perform a temporal processing of the plurality of sets of the object embeddings,
wherein performing the temporal processing comprises at least one of: , or performing a sequential memory-based processing of the plurality of sets of the object embeddings.
However, in an analogous art directed to machine learning, Pu teaches using a first neural network (NN) to obtain a plurality of sets of object embeddings, wherein each of the plurality of sets of object embeddings is associated with a respective time of the plurality of times (Pu par [0014] teaches collecting a sequence of radar image frames indicative of radar measurements of a scene at different consecutive instances of time and processing each permuted sequence of radar image frames with a first neural network trained to extract features of a frame at the dominant position in a temporal correlation with features of one or multiple frames in a subordinate position to produce temporally enhanced features for each of the frames in the sequence of radar image frames) and using a second NN to perform a temporal processing of the plurality of sets of the object embeddings (Pu par [0014] teaches processing a list of feature vectors from each feature map of the sequence of feature maps with a second neural network ), wherein performing the temporal processing comprises at least one of: performing a concurrent attention-based processing of the plurality of sets of the object embeddings, or performing a sequential memory-based processing of the plurality of sets of the object embeddings (Pu par [0014] teaches processing a list of feature vectors from each feature map of the sequence of feature maps with a second neural network ).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Pu with the teachings of Kang , Fukuda and Iscen to achieve the claimed invention. One would have been motivated to make such combination to enhance capacity of the automotive radar.(Pu par [0006])
As to claim 15, see the above rejection of claim 6.
Claims 7 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Kang , Fukuda and Iscen, in view of Wang et al.(US Patent Application Publication 2023/0078218 A1, hereinafter “Wang”)
As to claim 7, Kang , Fukuda and Iscen teach the method of claim 1 but fail to teach wherein one or more of the plurality of target outputs further comprise one or more manual annotations for at least one of the one or more objects.
However, Wang teaches wherein one or more of the plurality of target outputs further comprise one or more manual annotations for at least one of the one or more objects. (Wang par [0166] teaches human annotated)
Therefore , it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kang , Fukuda, Iscen and Wang to achieve the claimed invention. One would have been motivated to make such combination to improve accuracy of the classification.
As to claim 16, see the above rejection of claim 7.
Claims 8 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Kang , Fukuda and Iscen and further in view of Telpaz et al.(US Patent Application Publication 2023/0177840 A1, hereinafter “Telpaz”)
As to claim 8, Kang , Fukuda and Iscen teach the method of claim 1 but fail to teach wherein the student MLM comprises a common backbone of neural layers and a plurality of classification heads, each classification head of the plurality of classification heads outputting a classification of the one or more objects among a respective set of the plurality of sets of classes.
However, in an analogous art directed to machine learning, Telpaz teaches the MLM comprises a common backbone of neural layers and a plurality of classification heads, each classification head of the plurality of classification heads outputting a classification of the one or more objects among a respective set of the plurality of sets of classes. (Telpaz par [0049] teaches common backbone and multiple task heads to perform image classification tasks)
Therefore , it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kang , Fukuda , Iscen and Telpaz to achieve the claimed invention. One would have been motivated to make such combination to helps the machine learning training process to focus on features that contain identifiably “strong visual” information instead of simply overfitting the model to the training dataset.(Telpaz par [0049])
As to claim 17, see the above rejection of claim 8.
Claims 21-22 are rejected under 35 U.S.C. 103 as being unpatentable over Kang , Fukuda and Iscen, in view of Haidar et al.(US Patent Application Publication 2022/0335303 A1, hereinafter “Haidar” )
As to claim 21, Kang , Fukuda and Iscen teach the method of claim 1 but fail to teach wherein the combined loss function further characterizes similarity of one or more intermediate embeddings outputted by at least one hidden layer of the respective teacher MLM and one or more intermediate embeddings outputted by the student MLM.
However, Haidar teaches wherein the combined loss function further characterizes similarity of one or more intermediate embeddings outputted by at least one hidden layer of the respective teacher MLM and one or more intermediate embeddings outputted by the student MLM. ( Haidar par [0076]-[0077] teaches at 512, the intermediate representation loss module 222 processes the teacher intermediate representations and the student intermediate representations for the training batch 302 to compute an intermediate representation loss 316)
Therefore , it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention was made to combine the teachings of Kang , Fukuda, Iscen and Haidar to achieve the claimed invention. One would have been motivated to make such combination to improve the performance of the trained student at inference and provide better generalization capability and /or training student efficiently.(Haidar par [0056])
As to claim 22, see the above rejection of claim 21.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Vinyals et al. US Patent Application Publication 2020/0234192 A1, par [0019] discloses “the cumbersome machine learning model 110 is an ensemble machine learning model that is a compilation of multiple individual machine learning models that have been trained separately, with the outputs of the individual machine learning models being combined to generate the output of the cumbersome machine learning model 110. Further, in some cases, the models in the ensemble machine learning model include one or more full models that generate scores for each of the classes and one or more specialist models that generate scores for only a respective subset of the classes.”
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 HIEN DUONG whose telephone number is (571)270-7335. The examiner can normally be reached Monday-Friday 8:00AM-5:00PM.
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, Viker Lamardo can be reached at 571-270-5871. 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.
/HIEN L DUONG/Primary Examiner, Art Unit 2147
