DETAILED ACTION
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
2. This communication is in response to the Applicant’s submission filed 27 November 2023, where:
Claims 1-20 are pending.
Claims 1-20 are rejected.
Claim Rejections - 35 U.S.C. § 112
3 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.
Claim 7, line 3, recites “the one or more other data mining models.” There is insufficient antecedent basis for this limitation in the claim.
Claim Rejections - 35 U.S.C. § 101
4. 35 U.S.C. § 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
5. Claims 1-20 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to an abstract idea without significantly more.
Claim 1 recites a “computer-implemented method,” which is a process, and thus one of the statutory categories of patentable subject matter. (35 U.S.C. § 101).
However, under Step 2A Prong One, the claim recites the limitation of “[(c)]1 identifying, by the trained data source mining model based on the input data, a portion of the raw data to classify as mining data.” The activity of “[(c)] identifying,” contains limitations that can practically be performed in the human mind, including, for example, observations, evaluations, judgments, and opinions, and accordingly, are a mental process, (MPEP § 2106.04(a)(2) sub III), which is one of the groupings of abstract ideas. (MPEP § 2106.04(a)(2)). Thus, claim 1 recites an abstract idea.
Under Step 2A Prong Two, the claim as a whole is not integrated into a practical application, because the additional elements recited in the claim beyond the identified judicial exception include a “processing device,” which is recited at a high-level of generality, and accordingly, is a generic computer component used to implement the abstract idea, (MPEP § 2106.05(f)), that does not serve to integrate the abstract idea into a practical application. The claim also recites a “data source proxy head of a machine learning (ML) model,” and a “trained data source mining model,” which are also recited at a high-level of generality, and accordingly, are generic computer components used to implement the abstract idea, (MPEP §2106.05(f)), that does not serve to integrate the abstract idea into a practical application. The claim also recites “an autonomous vehicle (AV),” which generally linking the use of a judicial exception (that is, abstract idea) to particular technological environment or field of use," (MPEP § 2106.05(h)), that does not integrate the abstract idea into a practical application.
The claim also recites limitations of “[(a)] receiving, by a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV), a set of features selected from raw data by a backbone network of the ML model,” and “[(d)] providing, by the data source proxy head, identification of the portion of the raw data as a data mining output.” These activities “[(a)] receiving” and “[(d)] providing” are pre-processing and post-processing insignificant extra-solution activities of receiving and transmitting data, (MPEP §2106.05(g)), that does not serve to integrate the abstract idea into a practical application.
The claim also recites “[(b)] utilizing, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head,” which is the use of the generic computer components (processing device, trained data source mining model, data source proxy head) to implement the abstract idea, (MPEP § 2106.05(f)), that does not serve to integrate the abstract idea into a practical application. Therefore, claim 1 is directed to the abstract idea.
Finally, under Step 2B, the additional elements, taken alone or in combination, do not represent significantly more than the abstract idea itself. The additional elements recited in the claim beyond the identified judicial exception include a “processing device,” which is recited at a high-level of generality, and accordingly, is a generic computer component used to implement the abstract idea, (MPEP § 2106.05(f)), that does not amount to significantly more than the abstract idea. The claim also recites a “data source proxy head of a machine learning (ML) model,” and a “trained data source mining model,” which are also recited at a high-level of generality, and accordingly, are generic computer components used to implement the abstract idea, (MPEP §2106.05(f)), that does not amount to significantly more than the abstract idea. The claim also recites “an autonomous vehicle (AV),” which generally linking the use of a judicial exception (that is, abstract idea) to particular technological environment or field of use," (MPEP § 2106.05(h)), that does not amount to significantly more than the abstract idea.
The claim also recites limitations of “[(a)] receiving, by a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV), a set of features selected from raw data by a backbone network of the ML model,” and “[(d)] providing, by the data source proxy head, identification of the portion of the raw data as a data mining output.” These activities “[(a)] receiving” and “[(d)] providing” are well-understood, routine, and conventional activities of receiving or transmitting data over a network, (MPEP § 2106.05(d) sub II.i), that does not amount to significantly more than the abstract idea.
The claim also recites “[(b)] utilizing, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head,” which is the use of the generic computer components (processing device, trained data source mining model, data source proxy head) to implement the abstract idea, (MPEP § 2106.05(f)), that does not amount to significantly more than the abstract idea. Therefore, claim 1 is subject-matter ineligible.
Claim 10 recites an “apparatus,” which is a product, and thus one of the statutory categories of patentable subject matter. (35 U.S.C. § 101).
However, under Step 2A Prong One, the claim recites the limitation of “[(c)] identify, by the trained data source mining model based on the input data, a portion of the raw data to classify as mining data.” The activity of “[(c)] identify,” contains limitations that can practically be performed in the human mind, including, for example, observations, evaluations, judgments, and opinions, and accordingly, are a mental process, (MPEP § 2106.04(a)(2) sub III), which is one of the groupings of abstract ideas. (MPEP § 2106.04(a)(2)). Thus, claim 10 recites an abstract idea.
Under Step 2A Prong Two, the claim as a whole is not integrated into a practical application, because the additional elements recited in the claim beyond the identified judicial exception include a “one or more hardware processors,” which is recited at a high-level of generality, and accordingly, is a generic computer component used to implement the abstract idea, (MPEP § 2106.05(f)), that does not serve to integrate the abstract idea into a practical application. The claim also recites a “data source proxy head of a machine learning (ML) model,” and a “trained data source mining model,” which are also recited at a high-level of generality, and accordingly, are generic computer components used to implement the abstract idea, (MPEP §2106.05(f)), that does not serve to integrate the abstract idea into a practical application. The claim also recites “an autonomous vehicle (AV),” which generally linking the use of a judicial exception (that is, abstract idea) to particular technological environment or field of use," (MPEP § 2106.05(h)), that does not integrate the abstract idea into a practical application.
The claim also recites limitations of “[(a)] receive, by a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV), a set of features selected from raw data by a backbone network of the ML model,” and “[(d)] provide, by the data source proxy head, identification of the portion of the raw data as a data mining output.” These activities of “[(a)] receive” and “[(d)] provide” are pre-processing and post-processing insignificant extra-solution activities of receiving and transmitting data, (MPEP §2106.05(g)), that does not serve to integrate the abstract idea into a practical application.
The claim also recites “[(b)] utilize, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head,” which is the use of the generic computer components (one or more hardware processors, trained data source mining model, data source proxy head) to implement the abstract idea, (MPEP § 2106.05(f)), that does not serve to integrate the abstract idea into a practical application. Therefore, claim 10 is directed to the abstract idea.
Finally, under Step 2B, the additional elements, taken alone or in combination, do not represent significantly more than the abstract idea itself. The additional elements recited in the claim beyond the identified judicial exception include a “one or more hardware processors,” which is recited at a high-level of generality, and accordingly, is a generic computer component used to implement the abstract idea, (MPEP § 2106.05(f)), that does not amount to significantly more than the abstract idea. The claim also recites a “data source proxy head of a machine learning (ML) model,” and a “trained data source mining model,” which are also recited at a high-level of generality, and accordingly, are generic computer components used to implement the abstract idea, (MPEP §2106.05(f)), that does not amount to significantly more than the abstract idea. The claim also recites “an autonomous vehicle (AV),” which generally linking the use of a judicial exception (that is, abstract idea) to particular technological environment or field of use," (MPEP § 2106.05(h)), that does amount to significantly more than the abstract idea.
The claim also recites limitations of “[(a)] receive, by a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV), a set of features selected from raw data by a backbone network of the ML model,” and “[(d)] provide, by the data source proxy head, identification of the portion of the raw data as a data mining output.” These activities “[(a)] receive” and “[(d)] provide” are well-understood, routine, and conventional activities of receiving or transmitting data over a network, (MPEP § 2106.05(d) sub II.i), that does not amount to significantly more than the abstract idea.
The claim also recites “[(b)] utilize, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head,” which is the use of the generic computer components (one or more hardware processors, trained data source mining model, data source proxy head) to implement the abstract idea, (MPEP § 2106.05(f)), that does not amount to significantly more than the abstract idea. Therefore, claim 10 is subject-matter ineligible.
Claim 16 recites a “non-transitory computer-readable medium,” which is a product, and thus one of the statutory categories of patentable subject matter. (35 U.S.C. § 101).
However, under Step 2A Prong One, the claim recites the limitation of “[(c)] identify, by the trained data source mining model based on the input data, a portion of the raw data to classify as mining data.” The activity of “[(c)] identify,” contains limitations that can practically be performed in the human mind, including, for example, observations, evaluations, judgments, and opinions, and accordingly, are a mental process, (MPEP § 2106.04(a)(2) sub III), which is one of the groupings of abstract ideas. (MPEP § 2106.04(a)(2)). Thus, claim 10 recites an abstract idea.
Under Step 2A Prong Two, the claim as a whole is not integrated into a practical application, because the additional elements recited in the claim beyond the identified judicial exception include a “non-transitory computer-readable medium having stored thereon instructions that, when executed by one or more processors, cause the one or more processors to,” which is recited at a high-level of generality, and accordingly, is a generic computer component used to implement the abstract idea, (MPEP § 2106.05(f)), that does not serve to integrate the abstract idea into a practical application. The claim also recites a “data source proxy head of a machine learning (ML) model,” and a “trained data source mining model,” which are also recited at a high-level of generality, and accordingly, are generic computer components used to implement the abstract idea, (MPEP §2106.05(f)), that does not serve to integrate the abstract idea into a practical application. The claim also recites “an autonomous vehicle (AV),” which generally linking the use of a judicial exception (that is, abstract idea) to particular technological environment or field of use," (MPEP § 2106.05(h)), that does not integrate the abstract idea into a practical application.
The claim also recites limitations of “[(a)] receive, by a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV), a set of features selected from raw data by a backbone network of the ML model,” and “[(d)] provide, by the data source proxy head, identification of the portion of the raw data as a data mining output.” These activities of “[(a)] receive” and “[(d)] provide” are pre-processing and post-processing insignificant extra-solution activities of receiving and transmitting data, (MPEP §2106.05(g)), that does not serve to integrate the abstract idea into a practical application.
The claim also recites “[(b)] utilize, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head,” which is the use of the generic computer components (one or more hardware processors, trained data source mining model, data source proxy head) to implement the abstract idea, (MPEP § 2106.05(f)), that does not serve to integrate the abstract idea into a practical application. Therefore, claim 16 is directed to the abstract idea.
Finally, under Step 2B, the additional elements, taken alone or in combination, do not represent significantly more than the abstract idea itself. The additional elements recited in the claim beyond the identified judicial exception include a “one or more hardware processors,” which is recited at a high-level of generality, and accordingly, is a generic computer component used to implement the abstract idea, (MPEP § 2106.05(f)), that does not amount to significantly more than the abstract idea. The claim also recites a “data source proxy head of a machine learning (ML) model,” and a “trained data source mining model,” which are also recited at a high-level of generality, and accordingly, are generic computer components used to implement the abstract idea, (MPEP §2106.05(f)), that does not amount to significantly more than the abstract idea. The claim also recites “an autonomous vehicle (AV),” which generally linking the use of a judicial exception (that is, abstract idea) to particular technological environment or field of use," (MPEP § 2106.05(h)), that does not amount to significantly more than the abstract idea.
The claim also recites limitations of “[(a)] receive, by a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV), a set of features selected from raw data by a backbone network of the ML model,” and “[(d)] provide, by the data source proxy head, identification of the portion of the raw data as a data mining output.” These activities “[(a)] receive” and “[(d)] provide” are well-understood, routine, and conventional activities of receiving or transmitting data over a network, (MPEP § 2106.05(d) sub II.i), that does not amount to significantly more than the abstract idea.
The claim also recites “[(b)] utilize, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head,” which is the use of the generic computer components (one or more hardware processors, trained data source mining model, data source proxy head) to implement the abstract idea, (MPEP § 2106.05(f)), that does not amount to significantly more than the abstract idea. Therefore, claim 16 is subject-matter ineligible.
Claims 2, 3, 4, 8, and 9 depend directly or indirectly from claim 1. Claims 11, 12, and 13 depend directly or indirectly from claim 10. The claims recite more details or specifics to the additional element of “[(a)] receiving,” (claims 2 and 11: “[(a.1)] wherein the ML model comprises a plurality of heads including a primary model head and the data source proxy head”; claims 3 and 12: “[(a.2)] wherein, during training of the ML model, the data source proxy head and the primary model head are optimized during training at a same time”; claims 4 and 13: “[(a.2)] wherein, during training of the ML model, the data source proxy head is optimized separately from the primary model head by freezing weights and parameters of the backbone network”; claim 8: “[(a.1)] wherein the ML model comprises a trajectory generation model deployed on the AV”; and claim 9: “[(a.1)] wherein the data source proxy head implements a classifier model”), and accordingly, are merely more specific to the abstract idea. Therefore, claims 2-4, 8, 9, and 11-13 are subject-matter ineligible.
Claim 5 depends directly or indirectly from claim 1. Claim 14 depends directly or indirectly from claim 10. Claim 19 depends directly or indirectly from claim 16. The claims recite more details or specifics to the additional element of “the trained data source mining model,” “[(c.1)] wherein the trained data source mining model of the data source proxy head is trained separately from the ML model and is to consume the set of features that the ML model consumes without utilizing the backbone network of the ML model,” which is merely more specific to the additional element. Therefore, claims 5, 14, and 19 are subject-matter ineligible.
Claims 6 and 7 depend directly or indirectly from claim 1. Claim 15 depends directly or indirectly from claim 10. Claim 20 depends directly or indirectly from claim 16. The claims recite more details or specifics of the additional element of the “data source proxy head” (claim 6: [(b.1)] wherein the data source proxy head is to mitigate regression of the ML model by maintaining a loss weight of the data source proxy head below a determined weight value”; claims 7, 15, and 20: “[(b.1)] wherein, during training, the data source proxy head is to bootstrap one or more other data mining models by utilizing data sources of the one or more other data mining models and by leveraging manual user labels”), and accordingly, are merely more specific to the additional element. Therefore, claims 6, 7, 15, and 20 are subject-matter ineligible.
Claims 17 and 18 depends directly or indirectly from claim 16. The claim recites more details or specifics to the additional element of the “ML model,” (claim 17: “wherein the ML model comprises a plurality of heads including a primary model head and the data source proxy head,” and “wherein, during training of the ML model, the data source proxy head and the primary model head are optimized during training at a same time”; claim 18: “wherein the ML model comprises a plurality of heads including a primary model head and the data source proxy head,” and “wherein, during training of the ML model, the data source proxy head is optimized separately from the primary model head by freezing weights and parameters of the backbone network”), and accordingly, are merely more specific to the additional element. Therefore, claims 17 and 18 are subject-matter ineligible.
Claim Rejections - 35 U.S.C. § 102
6. 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)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(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.
7. Claims 1, 2, 4, 8, 9, 10, 11, 13, 16, and 18 are rejected under 35 U.S.C. § 102(a)(1) as being anticipated by US Published Application 20230057509 to Emmons et al. [hereinafter Emmons].
Regarding claims 1, 10, and 16, Emmons teaches [a] computer implemented method (Emmons, claim 1, teaches “[a] method implemented by a vehicles processor system [(that is, a computer implemented method )]”) of claim 1, [a]n apparatus (Emmons ¶ 0030 teaches “an example processor system 120 [(that is, an apparatus )]”) of claim 10, and [a] A non-transitory computer-readable medium having stored thereon instructions that, when executed by one or more processors, cause the one or more processors (Emmons, claim 10, teaches a “system comprising one or more processors and non-transitory computer storage media storing instructions that when executed by the one or more processors”) of claim 16, comprising:
[(a)] receiving, by a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV), a set of features selected from raw data by a backbone network of the ML model (Emmons, Fig. 2, teaches a vision-based machine learning model which includes a vulnerable road user (VRU) branch a non-VRU branch [Examiner annotations in dashed-line text boxes]:
PNG
media_image1.png
728
881
media_image1.png
Greyscale
Emmons ¶ 0049 teaches “a vulnerable road user (VRU) network 210 a non-VRU network 230 [(that is, “VRU network 210” and “non-VRU network 230” are a data source proxy head of a machine learning model)]. The example model may be executed by an autonomous vehicle, such as vehicle 100. Thus, actions of the model may be understood to be performed by a processor system (e.g., system 120) included in the vehicle [(that is, a processing device hosting a data source proxy head of a machine learning (ML) model deployed on an autonomous vehicle (AV))]”; Emmons ¶ 0051 teaches “a vision-based machine learning model includes backbone networks 200 which receive respective images as input [(that is, a backbone network of the ML model)]. Thus, the backbone networks 200 process the raw pixels included in the images 202A-202H [(that is, raw data)]. In some embodiments, the backbone networks 200 may be convolutional neural networks [(that is, receiving . . . a set of features selected from raw data by a backbone network of the ML model)]”);
[(b)] utilizing, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head (Emmons, Fig. 4A, teaches a trained data source mining model [Examiner annotations in dashed-line text boxes]:
PNG
media_image2.png
737
912
media_image2.png
Greyscale
Emmons ¶ 0079 teaches “vision information 204 from the backbone networks is received as input to the non-VRU network 230. A transformer network engine 402 receives the vision information 204 as input. In some embodiments, the transformer network engine 402 is trained to project the information 204 into a virtual camera space (e.g., vector space) [(that is, utilizing, by the data source proxy head, the set of features selected from the raw data as input data to a trained data source mining model of the data source proxy head)]”; Emmon ¶0084 teaches “[h]eads 410-414 may then determine output as illustrated in FIG. 2 [above] [(that is, “heads 410-414” are a trained data source mining model)]”);
[(c)] identifying, by the trained data source mining model based on the input data, a portion of the raw data to classify as mining data (Emmons ¶ 0060 & Fig. 2 (above) teaches “output 212, 232, from the VRU network 210 and non-VRU network 230 are illustrated in FIG. 2. The output may represent information associated with objects, such as location (e.g., position with a virtual camera space), depth, and so on. For example, the information may relate to cuboids associated with objects positioned about the autonomous vehicle [(that is, the “output 212, 232” is identifying, by the trained data source mining model based on the input data, a portion of the raw data to classify as mining data)]”);
[Examiner notes that the claim term “mining data” has the plain and ordinary meaning of sifting of large volumes of raw data resulting in hidden patterns, trends, and relationships, which may then be used to make informed decisions. The broadest reasonable interpretation of the claim term “mining data” covers the teachings of Emmons relating to the output 212, 232 from the VRU network 210 and non-VRU network 230, which is not inconsistent with the Applicant’s disclosure. (MPEP § 2111; Specification ¶ 0019 (“a data source proxy head that can run on the edge (edge platform) anywhere that a base ML model is running that includes source mining data sought after for mining purposes”))]; and
[(d)] providing, by the data source proxy head, identification of the portion of the raw data as a data mining output (Emmons, Fig. 2, teaches the identification of the portion of the raw data as a data mining output [Examiner annotations in dashed-line text boxes]:
PNG
media_image3.png
805
939
media_image3.png
Greyscale
Emmons ¶ 0060 teaches “output 212, 232, from the VRU network 210 and non-VRU network 230 [(that is, data source proxy head)] are illustrated in FIG. 2. The output may represent information associated with objects, such as location (e.g., position with a virtual camera space), depth, and so on. For example, the information may relate to cuboids associated with objects positioned about the autonomous vehicle [(that is, providing, by the data source proxy head, identification of the portion of the raw data as a data mining output)]”)
Regarding claims 2 and 11, Emmons teaches all of the limitations of claims 1 and 10, respectively as described above in detail. Emmons teaches -
[(a.1)] wherein the ML model comprises a plurality of heads including a primary model head and the data source proxy head (Emmons, Fig. 4A, teaches the ML model comprising a plurality of heads [Examiner annotation in dashed-line text boxes]:
PNG
media_image4.png
797
946
media_image4.png
Greyscale
Emmons ¶ 0073 teaches “the vision-based machine learning model described herein may include a multitude of trunks or heads. As known by those skilled in the art, these trunks or heads (collectively referred to herein as heads) may extend from a common portion of a neural network and be trained as experts in determining specific information”).
Regarding claims 4 and 13, Emmons teaches all of the limitations of claims 1 and 10, respectively, as described above in detail.
Emmons teaches -
[(a.2)] wherein, during training of the ML model, the data source proxy head is optimized separately from the primary model head by freezing weights and parameters of the backbone network (Emmons ¶ 0074 teaches “[i]n addition to being experts in specific information, the separation into different heads allows for piecemeal training [(that is, “piecemeal training” is optimized separately from the primary model head)] to quickly incorporate new training data. As new training information is obtained, portions of the machine learning model which would most benefit from the training information may be quickly updated. In this example, the training information may represent images or video clips of specific real-world scenarios gathered by vehicles in real-world operation. Thus, a particular head or heads may be trained, and the weights included in these portions of the network may be updated. For example, other portions (e.g., earlier portions of the network [(that is, the backbone network)]) may not have weights updated to reduce a training time and time to updating end-user autonomous vehicles [(that is, “earlier portions of the network” is freezing weights and parameters of the backbone network)]”; Emmons ¶ 0075 teaches that “[d]uring training, the error generated may be used to train for the loss [(that is, “error generated” is the data source proxy head is optimized)] in the pixels which a labeler has associated with the object or signal. Thus, only a head associated with this type of object or signal may be updated [(that is, wherein, during training of the ML model, the data source proxy head is optimized separately from the primary model head by freezing weights and parameters of the backbone network)]”).
Regarding claim 8, Emmons teaches all of the limitations of claim 1, as described above in detail.
Emmons teaches -
[(a.1)] wherein the ML model comprises a trajectory generation model deployed on the AV (Emmons ¶ 0025 teaches “outputs of the described model and the birdseye view network may be used by, for example, a planning and/or navigation model or engine [(that is, a trajectory generation model)] to effectuate autonomous or semi-autonomous driving:; Emmons ¶ 0093 teaches “the information (e.g., the outputs described herein) determined by the machine learning model described herein may be presented in a display of the vehicle. For example, the information may be used to inform autonomous driving (e.g., used by a planning and/or navigation engine) and optionally be presented as a visualization for a driver or passenger to view [(that is, the ML model comprises a trajectory generation model deployed on the AV)]”)
Regarding claim 9, Emmons teaches all of the limitations of claim 1, as described above in detail.
Emmons teaches -
[(a.1)] wherein the data source proxy head implements a classifier model (Emmons ¶ 0045 teaches “the vision-based machine learning model engine 126 may output object/signal information 124. This information 124 may represent information identifying objects depicted in the image information 122. For example, the information 122 may include one or more of positions of the objects (e.g., information associated with cuboids about the objects), velocities of the objects, accelerations of the objects, types or classifications of the objects [(that is, “classifications” is wherein the data source proxy head implements a classifier model)], whether a car object has its door open, and so on”).
Regarding claim 18, Emmons teaches all of the limitations of claim 16, as described above in detail.
Emmons teaches -
wherein the ML model comprises a plurality of heads including a primary model head and the data source proxy head (Emmons, Fig. 4A, teaches the ML model comprising a plurality of heads [Examiner annotation in dashed-line text boxes]:
PNG
media_image4.png
797
946
media_image4.png
Greyscale
Emmons ¶ 0073 teaches “the vision-based machine learning model described herein may include a multitude of trunks or heads. As known by those skilled in the art, these trunks or heads (collectively referred to herein as heads) may extend from a common portion of a neural network and be trained as experts in determining specific information”), and
wherein, during training of the ML model, the data source proxy head is optimized separately from the primary model head by freezing weights and parameters of the backbone network (Emmons ¶ 0074 teaches “[i]n addition to being experts in specific information, the separation into different heads allows for piecemeal training [(that is, “piecemeal training” is optimized separately from the primary model head)] to quickly incorporate new training data. As new training information is obtained, portions of the machine learning model which would most benefit from the training information may be quickly updated. In this example, the training information may represent images or video clips of specific real-world scenarios gathered by vehicles in real-world operation. Thus, a particular head or heads may be trained, and the weights included in these portions of the network may be updated. For example, other portions (e.g., earlier portions of the network [(that is, the backbone network)]) may not have weights updated to reduce a training time and time to updating end-user autonomous vehicles [(that is, “earlier portions of the network” is freezing weights and parameters of the backbone network)]”; Emmons ¶ 0075 teaches that “[d]uring training, the error generated may be used to train for the loss [(that is, “error generated” is the data source proxy head is optimized)] in the pixels which a labeler has associated with the object or signal. Thus, only a head associated with this type of object or signal may be updated [(that is, wherein, during training of the ML model, the data source proxy head is optimized separately from the primary model head by freezing weights and parameters of the backbone network)]”).
Claim Rejections – 35 U.S.C. § 103
8. 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.
9. 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.
10. Claims 3, 5, 12, 14, 17, and 18 are rejected under 35 U.S.C. § 103 as being unpatentable over US Published Application 20230057509 to Emmons et al. [hereinafter Emmons] in view of US Published Application 20220398405 to Hassan et al. [hereinafter Hassan].
Regarding claims 3 and 12, Emmons teaches all of the limitations of claims 1 and 10, respectively, as described above in detail.
Though Emmons teaches piecemeal training of different heads to quickly incorporate new training data, Emmons, however, does not explicitly teach –
[(a.2)] wherein, during training of the ML model, the data source proxy head and the primary model head are optimized during training at a same time.
But Hassan teaches -
[(a.2)] wherein, during training of the ML model, the data source proxy head and the primary model head are optimized during training at a same time (Hassan ¶ 0029 teaches “multiple heads can be used to improve the training of the backbone network and classification head [(that is, “multiple heads” is the data source proxy head and the primary model head)]. Once trained, the unneeded heads can be discarded”; Hassan ¶ 0038 teaches “[b]y minimizing the joint loss, the disclosed embodiments are able to train all prediction heads at the same time. In some embodiments, additional prediction heads can be added to further improve the accuracy of the distraction classification by learning good features capable of solving all tasks performed by the prediction heads [(that is, during training of the ML model, the data source proxy head and the primary model head are optimized during training at a same time)]”;
[Examiner notes that the plain and ordinary meaning of model “training” involves techniques like hyperparameter tuning, model pruning, and using efficient optimizers to enhance performance and reduce computational costs, which is synonymous to “optimized during training.” The broadest reasonable interpretation of the claim term “trained” or “training” includes optimization of a model as taught by the Hassan reference, which is not inconsistent with the Applicant’s disclosure. (MPEP § 2111)]).
Emmons and Hassan are from the same or similar field of endeavor. Emmons teaches as new training information is obtained, portions of the machine learning model which would most benefit from the training information may be quickly updated. Hassan teaches the distraction model can utilize multiple other heads during training, and minimize a joint loss among the heads.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Emmons pertaining to piecemeal training of multiple heads with the training all prediction heads at the same time of Hassan.
The motivation to do so is because “[d]uring training, a joint loss can be minimized while all prediction heads are used to generate predictions 210a-210n, 214a-214n. . . . Since each of the prediction heads 208a-208n, 212a-212n can be tuned to detect specific features of an image, the joint loss ensures that the backbone network 204 and individual prediction heads 208a-208n, 212a-212n are trained to emphasize critical regions of images useful for distraction classification.” (Hassan ¶ 0036).
Regarding claims 5, 14, and 19, Emmons teaches all of the limitations of claims 1, 10, and 16, respectively, as described above in detail.
Though Emmons teaches piecemeal training of different heads to quickly incorporate new training data, Emmons, however, does not explicitly teach –
[(c.1)] wherein the trained data source mining model of the data source proxy head is trained separately from the ML model and is to consume the set of features that the ML model consumes without utilizing the backbone network of the ML model
But Hassan teaches -
[(c.1)] wherein the trained data source mining model of the data source proxy head is trained separately from the ML model and is to consume the set of features that the ML model consumes without utilizing the backbone network of the ML model (Hassan ¶ 0058 teaches “heads 300, 400, and 500 can all be connected to the backbone network and FPN during training. In this arrangement, back-propagation is used to perform a gradient descent operation to minimize a loss function. This back-propagation results in changes to the weights and other model parameters to minimize such a loss function. Since the model, during training, includes multiple heads, there are multiple loss functions for each head and, in some instances, multiple loss functions for a single head, such as the object detection prediction head. Thus . . . a joint loss function is used to perform back-propagation. Since a joint loss function is used, the entire network is optimized based on the outputs of all prediction heads, which forces the network to focus on the features detected by the prediction heads [(that is, training with “back-propagation” is the trained data source mining model of the data source proxy head is trained separately from the ML model and is to consume the set of features that the ML model consumes without utilizing the backbone network of the ML model)]”).
Emmons and Hassan are from the same or similar field of endeavor. Emmons teaches as new training information is obtained, portions of the machine learning model which would most benefit from the training information may be quickly updated. Hassan teaches the distraction model can utilize multiple other heads during training, and minimize a joint loss among the heads.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Emmons pertaining to piecemeal training of multiple heads with the back propagation joint-loss-based training without using a backbone network of Hassan.
The motivation to do so is because “existing systems rely primarily on image classification, which is unable to consider features such as pose estimation and object detection and instead relies exclusively on labeled image data to train a classifier. As a result, such systems are relatively inaccurate, since the model does not fundamentally understand which regions of an image are of interest for the problem of distraction classification.” (Hassan ¶ 0058).
Regarding claim 17, Emmons teaches all of the limitations of claim 16, as described above in detail.
Emmons teaches -
wherein the ML model comprises a plurality of heads including a primary model head and the data source proxy head (Emmons, Fig. 4A, teaches the ML model comprising a plurality of heads [Examiner annotation in dashed-line text boxes]:
PNG
media_image4.png
797
946
media_image4.png
Greyscale
Emmons ¶ 0073 teaches “the vision-based machine learning model described herein may include a multitude of trunks or heads. As known by those skilled in the art, these trunks or heads (collectively referred to herein as heads) may extend from a common portion of a neural network and be trained as experts in determining specific information”), and
Though Emmons teaches piecemeal training of different heads to quickly incorporate new training data, Emmons, however, does not explicitly teach –
[(a.2)] wherein, during training of the ML model, the data source proxy head and the primary model head are optimized during training at a same time.
But Hassan teaches -
[(a.2)] wherein, during training of the ML model, the data source proxy head and the primary model head are optimized during training at a same time (Hassan ¶ 0029 teaches “multiple heads can be used to improve the training of the backbone network and classification head [(that is, “multiple heads” is the data source proxy head and the primary model head)]. Once trained, the unneeded heads can be discarded”; Hassan ¶ 0038 teaches “[b]y minimizing the joint loss, the disclosed embodiments are able to train all prediction heads at the same time. In some embodiments, additional prediction heads can be added to further improve the accuracy of the distraction classification by learning good features capable of solving all tasks performed by the prediction heads [(that is, during training of the ML model, the data source proxy head and the primary model head are optimized during training at a same time)]”).
Emmons and Hassan are from the same or similar field of endeavor. Emmons teaches as new training information is obtained, portions of the machine learning model which would most benefit from the training information may be quickly updated. Hassan teaches the distraction model can utilize multiple other heads during training, and minimize a joint loss among the heads.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Emmons pertaining to piecemeal training of multiple heads with the training all prediction heads at the same time of Hassan.
The motivation to do so is because “[d]uring training, a joint loss can be minimized while all prediction heads are used to generate predictions 210a-210n, 214a-214n. . . . Since each of the prediction heads 208a-208n, 212a-212n can be tuned to detect specific features of an image, the joint loss ensures that the backbone network 204 and individual prediction heads 208a-208n, 212a-212n are trained to emphasize critical regions of images useful for distraction classification.” (Hassan ¶ 0036).
11. Claims 6, 15, and 20 are rejected under 35 U.S.C. § 103 as being unpatentable over US Published Application 20230057509 to Emmons et al. [hereinafter Emmons] in view of US Published Application 20240095945 to Zhu et al. [hereinafter Zhu].
Regarding claims 6, 15, and 20, Emmons teaches all of the limitations of claims 1, 10, and 16, respectively, as described above in detail.
Though Emmons teaches piecemeal training of different heads to quickly incorporate new training data, Emmons, however, does not explicitly teach –
[(b.1)] wherein the data source proxy head is to mitigate regression of the ML model by maintaining a loss weight of the data source proxy head below a determined weight value.
But Zhu teaches -
[(b.1)] wherein the data source proxy head is to mitigate regression of the ML model by maintaining a loss weight of the data source proxy head below a determined weight value (Zhu ¶ 0074 teaches a “first architecture 300a of the model includes two main components. A network backbone 310 and an object detection head 320 [(that is, data proxy head)]”; Zhu ¶ 0086 teaches a “value of the first and second loss weights may be between an upper weight limit value and a lower weight limit value [(that is, “upper weight value” is maintaining a loss weight . . . below a determined weight value)]. The first and second loss weight may be adapted during training 400 based on a value of the first 470 loss and second loss 480 respectively. . . . As a result, when training 400 the model 410, only the regression loss 470 may be considered for minimizing [(that is, to mitigate regression of the ML model)], since the uncertainty regression loss 480 is multiplied by 0 and is thus not considered in the final loss 490. Once the regression loss 470 is sufficiently low (e.g., the regression loss is smaller than or equal to a detection quality threshold), meaning that the accuracy of the object detection is sufficiently high, the value of the second loss weight may be adapted [(that is, wherein the data source proxy head is to mitigate regression of the ML model by maintaining a loss weight of the data source proxy head below a determined weight value)]”).
Emmons and Zhu are from the same or similar field of endeavor. Emmons teaches as new training information is obtained, portions of the machine learning model which would most benefit from the training information may be quickly updated. Zhu teaches an object detection head configured to output a predicted feature of an object within the scene representation.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Emmons pertaining to piecemeal training of multiple heads with minimizing regression loss and associated loss weight of Zhu.
The motivation to do so is that, by “determining based on the estimated uncertainty quality whether the model needs further training or is application ready, it is verified that only a model, which accurately and reliably predicts uncertainty is deployed into technical systems/mechanical agents, which increases the efficiency and safety of the system. The object detection may for example be applied to a vicinity of a mechanical agent (e.g., a vehicle, a car, a drone, a ship or a robot).” (Zhu ¶ 0009).
12. Claim 7 is rejected under 35 U.S.C. § 103 as being unpatentable over US Published Application 20230057509 to Emmons et al. [hereinafter Emmons] in view of Gokmen et al., “Asking for Help: Failure Prediction in Behavioral Cloning through Value Approximation,” arXiv (Feb 2023) [hereinafter Gokmen].
Regarding claim 7, Emmons teaches all of the limitations of claim 1, as described above in detail.
Though Emmons teaches piecemeal training of different heads to quickly incorporate new training data, Emmons, however, does not explicitly teach –
[(b.1)] wherein, during training, the data source proxy head is to bootstrap one or more other data mining models by utilizing data sources of the one or more other data mining models and by leveraging manual user labels.
But Gokmen teaches -
[(b.1)] wherein, during training, the data source proxy head is to bootstrap one or more other data mining models by utilizing data sources of the one or more other data mining models and by leveraging manual user labels (Gokmen, right column of p. 3, “III. Problem Setup,” third paragraph, teaches a ”dataset of rollouts from the first policy then needs to be collected and labelled under full human supervision [(that is, leveraging manual user labels)], and discounted returns computed offline to be used in bootstrapping the value estimate [(that is, “value estimate” is bootstrap . . . by utilizing data sources)]. Then, each following policy πk(a | s) is trained jointly with a value estimate Vk(s) from previous episodes’ labelled-rollouts”; Gokmen, right column of p. 6, “VI. Conclusion,” first paragraph, teaches “[Behavioral Cloning Value Approximation (BCVA)] on a complex mobile-manipulation task of latched door opening with rich environment observations and contacts, and our model was able to identify failure risks with 86% precision and 81% recall. This ability is relatively straightforward to implement for typical neural network-based Behavioral Cloning policies, as a separate prediction head on top of a shared feature-encoder backbone [(that is, a “separate prediction head” is utilizing data sources of the one or more other data mining models)]. Weight sharing allows for both policy learning and state value estimation to jointly improve and cue on shared features (e.g. leveraging failed episodes for representation learning)”; Gokmen, right column of p. 5, “C. Asking for Help,” second paragraph, teaches that for “operational deployment, we tune the thresholds ϵ and ν by running the value estimate on rollouts from the (human labelled) validation set [(that is, “tune” is during training)], computing the episode-level confusion matrix across different values of ϵ and ν, and picking appropriate values such that the model satisfies requirements in terms of both overall precision and recall, as well as being able to correctly flag a small, hand-selected sample of failures of concern [(that is, during training, the data source proxy head is to bootstrap one or more other data mining models by utilizing data sources of the one or more other data mining models and by leveraging manual user labels)]”).
Emmons and Gokmen are from the same or similar field of endeavor. Emmons teaches as new training information is obtained, portions of the machine learning model which would most benefit from the training information may be quickly updated. Gokmen teaches Behavioral Cloning Value Approximation (BCVA) can be learned simply as a regression head on top of an existing Behavioral Cloning model, allowing weight sharing.
Thus, it would have been obvious to a person having ordinary skill in the art as of the effective filing date of the Applicant’s invention to modify Emmons pertaining to piecemeal training of multiple heads with the bootstrap valuation values and/or approximation training of Gokmen.
The motivation to do so is because “[Behavioral Cloning Value Approximation (BCVA)] can be learned simply as a regression head on top of an existing Behavioral Cloning model, allowing for low cost training and inference as well as weight sharing.” (Gokmen, left column of p. 2, “I. Introduction,” third paragraph).
Conclusion
13. The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
(Grigorescu et al., "Cloud2Edge Elastic AI Framework for Prototyping and Deployment of AI Inference Engines in Autonomous Vehicles," (2020)) teaches training samples and labels are aggregated into single input and output layers, respectively. The automatic configuration of the input-output layers can be manually tuned if the samples and labels are composed of multiple types of sensory measurements (e.g. images and LiDAR) and/or multiple prediction heads (e.g. 2D object detection and 3D object reconstruction). Once the input-output shapes of the DNN model have been defined, we proceed to computing the architecture of the inner layers. We split the anatomy of a DNN architecture into three main blocks: i) backbone model, ii) feature extractor model and iii) prediction heads. These are chosen from subsets of state-of-the-art DNN architectures, such as VGG16, MobileNetV2 and Darknet for the backbone, or YoloV3-V5, RetinaNet and SSD (Single Shot Detector) for the case of prediction heads in object detection. We compute the DNN structure depending on the model capacity determined at Data Processor stage. Namely, in order to avoid overfitting if training data is scarce, we use a light-weighted backbone, such as VGG16, coupled to prediction heads having a lower number of convolutional layers. Different DNN models are suggested based on this analysis. The models are subsequently adapted and modified in our SiL and HiL paradigm for obtaining an optimal AI Inference Engine for the task at hand.
(US Published Application 20210012116 to Urtasun et al.) teaches determining a feature embedding and at least one of an instance embedding, class embedding, and/or background embedding for each of the plurality of three-dimensional points. The method can include determining a first subset of points associated with one or more known instances within the environment based on the class embedding and the background embedding associated with each point in the plurality of points.
14. Any inquiry concerning this communication or earlier communications from the Examiner should be directed to KEVIN L. SMITH whose telephone number is (571) 272-5964. Normally, the Examiner is available on Monday-Thursday 0730-1730.
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, KAKALI CHAKI can be reached on 571-272-3719. 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.
/K.L.S./
Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122
1 References to the limitations are provided for the limited purpose of aiding in the subject-matter eligibility evaluation under the Office guidance and not for the purpose of oversimplifying the claims
Prosecution Insights