2The present application, filed on or after 16 March 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
This office action is in response to Applicant’s submission filed on 26 January 2024. THIS ACTION IS NON-FINAL.
Status of Claims
Claims 1-42 are pending.
Claims 16-30 are rejected under 35 U.S.C. 112(b) as indefinite.
Claims 1-11, 14-24, 27-42 are rejected under 35 U.S.C. 103 as unpatentable.
Claims 12-13 are objected to for depending on rejected claims.
There is no art rejection for claims 12-13, 25-26.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
A claim is indefinite if, when read in light of the specification, it fails to inform, with reasonable certainty, those skilled in the art about the scope of the invention. Nautilus, Inc. v. Biosig Instruments, Inc., 110 USPQ.2d 1688, U.S. Supreme Court (2014).
Claims 16-30 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor, or for pre-AIA the applicant regards as the invention.
Regarding claim 16, the claim depends from claim 15, which depends from claim 14, which depends from claim 3. Claim 3 already recites that "the preexisting trained model is trained to perform a first task" and refers to "a second task." Claim 16 again recites "wherein the preexisting trained model is trained to perform a first task" and "to perform a second task." It is unclear whether the first and second tasks of claim 16 are the same tasks recited in claim 3, or additional, different tasks. If they are different tasks, it is further unclear how a single preexisting trained model is "trained to perform" two different first tasks. The metes and bounds of the claim therefore cannot be determined, the claim is therefore indefinite.
Claims 17–30 depend from claim 16, do not cure the deficiency, and are rejected for the same reason.
Regarding claim 17,
(a)the claim recites the following:
"a first category of objects represented in one or more images"
"a second category of objects represented in the one or more images"
Claim 14, from which claim 17 indirectly depends, already introduces "one or more images," "a first category of objects," and "a second category of objects." It is unclear whether claim 17 refers to the same images and categories as claim 14 or to different ones.
(b) the claim also recites "the second categories of objects," which lacks antecedent basis in the claims, the claim is therefore indefinite. For the purpose of applying prior art, this limitation is construed to be “the second category of objects”.
Claims 18–26 depend from claim 17, do not cure the deficiency, and are rejected for the same reasons.
Regarding claim 27, the claim recites the following:
"applying an image segmentation process to one or more images"
"a first category of objects"
"a second category of objects"
Each of these limitations is already introduced in claim 14, from which claim 27 indirectly depends through claims 16 and 15. It is unclear whether claim 27 refers to the same image segmentation process, images and categories as claim 14, or to a second, distinct set, the claim is therefore indefinite. For purposes of examination, the Examiner interprets the limitations of claims 16, 17 and 27 as referring to the same tasks, images and categories previously recited.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1–3, 31–35 and 38–40 are rejected under 35 U.S.C. 103 as being unpatentable over Guo et al., "SpotTune: Transfer Learning through Adaptive Fine-tuning," arXiv:1811.08737v1 (Nov. 2018), published at CVPR 2019 [hereafter Guo], in view of Lee et al., "Surgical Fine-Tuning Improves Adaptation to Distribution Shifts," arXiv:2210.11466v1 (Oct. 20, 2022) [hereafter Lee].
With regards to claim 1, Guo teaches:
"A method for generating an extended trained model, the method comprising:" (Guo, FIG.1-2, 3 Proposed Approach,
PNG
media_image1.png
492
402
media_image1.png
Greyscale
"obtaining a preexisting trained model, the preexisting trained model including a plurality of preexisting weights, wherein each of the plurality of preexisting weights is associated with a preexisting value;" (Guo, FIG.1-2,
PNG
media_image2.png
391
773
media_image2.png
Greyscale
3.Proposed Approach, Eq. (1), and Sec. 4.1 ‘Pre-trained model,’ [The method starts from the l-th residual block F_l of a pre-trained ResNet. The experiments use a PyTorch ResNet-50 pre-trained on ImageNet, which has 16 residual blocks]. The convolutional parameters of each pre-trained block F_l are preexisting weights. Their values, learned during ImageNet pre-training, are the preexisting values.)
"identifying a subset of the plurality of preexisting weights;" (Guo, FIG.1-2, 3 Proposed Approach, and 1.Introduction, 3.Proposed Approach, ‘decides, per training example, which layers of the pre-trained model should be fine-tuned (adapted to the target task) and which layers should have their parameters frozen …’, [The global-k variant constrains every image to fine-tune the same k blocks, which may be placed anywhere in the ResNet. Rather than having k chosen by hand, the variant learns the k blocks that give the best target accuracy]. The learned set of k blocks is an identified subset of the preexisting weights.)
"generating a plurality of extended weights based on a training process using duplicates of the subset of the plurality of preexisting weights …." (Guo, FIG.1-2, 3.Proposed Approach,
PNG
media_image3.png
210
386
media_image3.png
Greyscale
'we freeze the original block F_l and create a new trainable block', [The new block F̂_l is initialized from F_l's parameters and is a replicated block that is optimized toward the target dataset, while the frozen block F_l stays unchanged]. The new block is initialized with the parameters of F_l, so it is a duplicate of the preexisting weights. Optimizing it on the target data is a training process that produces the extended weights.).
Guo does not explicitly detail “generating the extended trained model, wherein the extended trained model includes the plurality of preexisting weights and the plurality of extended weights.”.
However, Lee teaches "generating the extended trained model, wherein the extended trained model includes the plurality of preexisting weights and the plurality of extended weights (Lee, Fig. 1 caption, 'tune only one block of parameters and freeze the remaining parameters';
PNG
media_image4.png
336
646
media_image4.png
Greyscale
and 2.Surgical Fine-Runing: Freezing Parameters During Adaption, [Surgical fine-tuning is defined as fine-tuning only a small subset of the layers (i.e., to generate extended weights) of the pre-trained network and keep the remaining layers frozen (i.e., keep pre-exiting weights), to generate an refined model.]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo and Lee before him or her, to modify the adaptive fine-tuning method of Guo to include the fine tuning shown in Lee.
The motivation for doing so would have been to improve transfer learning (Lee, Abstract).
With regards to claim 2, Guo in view of Lee teaches claim 1:
"The method of claim 2, wherein identifying the subset of the plurality of preexisting weights includes selecting the subset of the plurality of preexisting weights based on a comparison of a performance criterion associated with each of the plurality of preexisting weights to a threshold." (Guo, FIG.1-2, 3.Proposed Approach , Eq. (3), and Eqs. (5)–(6), [For each residual block, a policy trained with the target classification loss to maximize target accuracy outputs two scores, one for freeze and one for fine-tune. The fine-tune decision is the argmax of the two. In the global-k variant, the per-block fraction v_l is driven to exactly 0 or 1 to fix which blocks are fine-tuned].)
With regards to claim 3, Guo in view of Lee teaches:
"The method of claim 2, wherein the preexisting trained model is trained to perform a first task," (Guo, FIG.1-2, 4.Experiment, "Pre-trained model," [ResNet-50 pre-trained on ImageNet classification]. ImageNet classification is the first task.)
"wherein the performance criterion associated with each of the plurality of preexisting weights is determined based on causing the preexisting trained model to perform a second task," (Guo, FIG.1-2, Table 1-2, 4.Experiments,
PNG
media_image5.png
174
379
media_image5.png
Greyscale
4 Experiment "Datasets and metrics," [The target tasks are CUBS, Stanford Cars, Flowers, Sketches and WikiArt classification, which differ from ImageNet].)
With regards to claim 31, Guo in view of Lee teaches:
"The method of claim 1, wherein the preexisting value of each of the plurality of preexisting weights includes a numerical value." (Guo, FIG.1-2, Table 2-3, 3.Introduction, ‘few layers of the deep net-work can be fine-tuned while freezing the parameters of the remaining initial layers to their pre-trained values …,’ and 4.Experiments, "Pre-trained model," [The pre-trained blocks are convolutional layers whose learned parameters initialize the trainable blocks. The PyTorch model reaches 75.15% accuracy on ImageNet]. Learned convolutional parameters are necessarily numerical values.)
With regards to claim 32, Guo in view of Lee teaches:
"The method of claim 1, wherein the preexisting trained model includes a neural network." (Guo, FIG.1-2, 3.Proposed Approach, [The method is described using a Residual Network (ResNet)]. A ResNet is a deep neural network.)
With regards to claim 33, Guo teaches:
"A system for generating an extended trained model, the system comprising: at least one processor programmed to:" (Guo, 4.Experiment, "Implementations details," [The implementation is written in PyTorch, and all models are trained on two NVIDIA V100 GPUs]. The GPUs executing the PyTorch implementation are at least one processor programmed to perform the method.)
"obtain a preexisting trained model, the preexisting trained model including a plurality of preexisting weights, wherein each of the plurality of preexisting weights is associated with a preexisting value;" (Guo, FIG.1-2, 3.Proposed Approach, Eq. (1); Sec. 4.1, See the corresponding limitation of claim 1 above.)
"identify a subset of the plurality of preexisting weights;" (Guo, FIG.1-2, 3.Proposed Approach; 1. Introduction. See claim 1 above.)
"generate a plurality of extended weights based on a training process using duplicates of the subset of the plurality of preexisting weights;" (Guo, FIG.1-2, 3.Proposed Approach. See claim 1 above.)
"and generate the extended trained model, wherein the extended trained model includes the plurality of preexisting weights and the plurality of extended weights." (Guo, FIG.1-2, 3.Proposed Approach, Eq. (2); Fig. 2, p. 4. See claim 1 above.)
Guo does not explicitly detail duplicating only an identified subset while retaining all preexisting weights. However, Lee teaches identifying and tuning only a subset of pre-trained layers while freezing the rest (Lee, Fig. 1; 1. Introduction). The rationale for combining Guo and Lee is the same as set forth for claim 1 and is incorporated here.
With regards to claim 38, Guo teaches:
"A non-transitory computer-readable medium storing instructions executable by at least one processor to perform a method for generating an extended trained model, the method comprising:" (Guo, 4.Experiments "Implementations details," [The method is implemented in PyTorch and trained on GPUs, and the authors state that the source code will be made public]. A PyTorch implementation executed on GPUs is necessarily stored as program instructions in non-transitory memory for execution by those processors.)
The remaining limitations of claim 38 are identical to those of claim 1 and are mapped as set forth for claim 1 (Guo, FIG.1-2, 3.Proposed Approach, Eqs. (1)–(2); Fig. 2, p. 4). The same combination with Lee (Fig. 1, 1.Introduction), and the same motivation, apply.
Claims 34-35, 39-40 are substantially similar to claims 2-3. The arguments as given above for claims 2-3 are applied, mutatis mutandis, to claims 34-35, 39-40, therefore the rejection of claims 2-3 are applied accordingly.
The combined teaching described above will be referred as Guo + Lee hereafter.
Claims 4-5, 7-11, 14–18, 20-24, 27–29, 36-37 and 41-42 are rejected under 35 U.S.C. 103 as being unpatentable over Guo et al., "SpotTune: Transfer Learning through Adaptive Fine-tuning," arXiv:1811.08737v1 (Nov. 2018), published at CVPR 2019 [hereafter Guo], in view of Lee et al., "Surgical Fine-Tuning Improves Adaptation to Distribution Shifts," arXiv:2210.11466v1 (Oct. 20, 2022) [hereafter Lee], and further in view of Yu et al., "BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning," arXiv:1805.04687v2 (Apr. 8, 2020), CVPR 2020 [hereafter Yu].
With regards to claim 4, Guo + Lee teaches
"The method of claim 3, wherein the first task includes identifying a first category of objects represented in one or more images," (Guo, fog.2-4, 4.Experiments "Pre-trained model," [The pre-trained ResNet-50 performs ImageNet classification]. Identifying ImageNet object classes in images is identifying a first category of objects represented in images.)
Guo + Lee does not explicitly detail "wherein the second task includes identifying a second category of objects represented in the one or more images, and wherein the first category of objects and the second category of objects are different categories".
However, Yu teaches "wherein the second task includes identifying a second category of objects represented in the one or more images, and wherein the first category of objects and the second category of objects are different categories" (Yu, FIG.1-6,
PNG
media_image6.png
402
795
media_image6.png
Greyscale
3.BDD100k, [For each video, the frame at the 10th second is the one annotated for all image tasks, Bounding boxes in 10 categories are annotated on each reference frame]; 3.BDD100K and Fig. 5, [Lane markings on those images are labeled in 8 main categories]; 5.Multitask Learning, [The lane marking and drivable area tasks use the same 70K training images]). Yu thus performs several recognition tasks, each identifying a different category of objects, on the same driving images. Vehicle detection is one task and lane-marking detection is another, applied to the same reference frames.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed processing of image
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claim 5, Guo + Lee and Yu teaches:
"The method of claim 4”.
Guo + Lee does not explicitly detail " wherein the one or more images are representative of an environment of at least one host vehicle ".
However, Yu teaches " wherein the one or more images are representative of an environment of at least one host vehicle " (Yu, FIG.1,5-6, , 3.BDD100K, [The driving videos were crowd-sourced from tens of thousands of drivers and come with GPS/IMU recordings of the driving trajectories]. The frames are captured from the vehicle being driven, so they represent that host vehicle's environment.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed processing of image
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claim 7, Guo in view of Lee and Yu teaches:
“The method of claim 4.”
Guo + Lee does not explicitly detail “wherein the first category of objects includes vehicles and the second category of objects includes traffic signals."
However, Yu teaches ”wherein the first category of objects includes vehicles and the second category of objects includes traffic signals." (Yu, FIG.1, 5-6, 3.BDD100K and Table 8, [The road-scene label set includes Car, Truck, Bus and Motorcycle classes as well as a traffic Light class]. Vehicles and traffic lights are each identified in the same driving images.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed processing of image
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claim 8, Guo in view of Lee and Yu teaches:
"The method of claim 4.”
Guo + Lee does not explicitly detail “wherein the first category of objects includes vehicles and the second category of objects includes pedestrians."
However, Yu teaches “wherein the first category of objects includes vehicles and the second category of objects includes pedestrians." (Yu, FIG.1, 5-6, Table 8, 5. Multitask Learning,
PNG
media_image7.png
203
787
media_image7.png
Greyscale
.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed processing of image
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claim 9, Guo + Lee and Yu teaches:
“The method of claim 4.”
Guo + Lee does not explicitly detail “wherein the first category of objects includes vehicles and the second category of objects includes lane markings."
However, Yu teaches ”wherein the first category of objects includes vehicles and the second category of objects includes lane markings." (Yu, FIG.5-6,
PNG
media_image8.png
187
800
media_image8.png
Greyscale
3.BD100K [Objects including vehicles are annotated with bounding boxes, and lane markings are annotated on the same reference frames].)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed processing of image
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claim 10, Guo + Lee and Yu teaches:
“The method of claim 4.”
Guo + Lee does not explicitly detail “wherein the first category of objects includes vehicles and the second category of objects includes signs."
However, Yu teaches "wherein the first category of objects includes vehicles and the second category of objects includes signs." (Yu, FIG.9,
PNG
media_image9.png
310
787
media_image9.png
Greyscale
Table 8, [The label set includes a traffic Sign class, and a Cityscapes-trained model fails to recognize US traffic signs].)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed processing of image
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claim 11, Guo + Lee and Yu teaches:
“The method of claim 4.”
Guo + Lee does not explicitly detail “wherein the first category of objects includes a first type of vehicle and the second category of objects includes a second type of vehicle, and wherein the first type of vehicle and the second type of vehicle are different types."
However, Yu teaches “wherein the first category of objects includes a first type of vehicle and the second category of objects includes a second type of vehicle, and wherein the first type of vehicle and the second type of vehicle are different types." (Yu, FIG.1, Table 8, 5. Multitask Learning, [Car, truck, bus, train, motorcycle and bicycle are annotated as separate categories]. For example, car and truck are different types of vehicle.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed processing of image
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claim 14, Guo + Lee teaches
“The method of claim 3.”,
Guo + Lee does not explicitly detail “wherein the first task includes applying an image segmentation process to one or more images and identifying a first category of objects represented in the one or more images, wherein the second task includes identifying a second category of objects represented in the one or more images, and wherein the first category of objects and the second category of objects are different categories.”
However, Yu teaches “wherein the first task includes applying an image segmentation process to one or more images and identifying a first category of objects represented in the one or more images, wherein the second task includes identifying a second category of objects represented in the one or more images, and wherein the first category of objects and the second category of objects are different categories" (Yu, FIG.1-6, 3.BD100K, [Each pixel receives a label from 40 road-scene object classes]; 5. Multitask Learning, [A base semantic segmentation model is fine-tuned by training it jointly with the detection and the lane marking/drivable area tasks]; 3.BD100K [All image tasks annotate the same 10th-second frame]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed processing of image
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claim 15, Guo + Lee in view of Yu teaches:
"The method of claim 14, wherein the training process using the duplicates of the subset of the plurality of preexisting weights includes modifying the preexisting value of each of the duplicates of the subset of the plurality of preexisting weights." (Guo, FIG.1-2, 3.BD100K, 'the replicated block F̂_l, which is initialized from F_l, can be optimized towards the target dataset'.) Optimizing the replicated block changes its values away from the copied pretrained values.
With regards to claim 16, Guo + Lee in view of Yu teaches:
"The method of claim 15, wherein the preexisting trained model is trained to perform a first task," (Guo, FIG.1-2, 3.BD100K, [The frozen block F_l was trained on the source task].)
"wherein the training process using the duplicates of the subset of the plurality of preexisting weights includes using the duplicates of the subset of the plurality of preexisting weights to perform a second task, and wherein the first task and the second task are different tasks." (Guo, 3. BD100K, [The replicated block is optimized on the target dataset and is jointly trained with the target classification loss]. The target task is different from the source task.)
Claim 17 is substantially similar to claim 4. The arguments as given above for claim 4 are applied, mutatis mutandis, to claim 17, therefore the rejection of claim 4 are applied accordingly.
Claims 18, 20–24 are substantially similar to claims 5,7-11. The arguments as given above for claims 5,7-11 are applied, mutatis mutandis, to claims 18,20-24, therefore the rejection of claims 5,7-11 are applied accordingly.
Claim 27 is substantially similar to claim 14. The arguments as given above for claim 14 are applied, mutatis mutandis, to claim 27, therefore the rejection of claim 14 are applied accordingly.
With regards to claim 28, Guo + Lee in view of Yu teaches:
"The method of claim 16, the method further comprising performing the first task and the second task using the extended trained model." (Guo, FIG.1-2, 4. Experiments and Table 3, [The model built from the ImageNet-pretrained ResNet-26 is evaluated on ImageNet and on the other nine Visual Decathlon domains]).
With regards to claim 29, Guo + Lee in view of Yu teaches:
“The method of claim 28.”,
Guo + Lee does not explicitly detail “wherein at least one processor included in a navigation system of a host vehicle is programmed to perform the first task and the second task using the extended trained model.”
However, Yu teaches "wherein at least one processor included in a navigation system of a host vehicle is programmed to perform the first task and the second task using the extended trained model." (Yu, 3.BDD100K, [Lane marking detection is critical for vision-based vehicle localization and trajectory planning]; 6.Clonclusion, [The tasks each serve a different aspect of a complete driving system].)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee and Yu before him or her, to modify the subset-duplication fine-tuning of Guo + Lee to include detailed vehicle control.
The motivation for doing so would have been to support multitask learning (Yu, Abstract).
With regards to claims 36 and 41, Guo in view of Lee teaches claims 35 and 40, respectively, as set forth above. Claims 36 and 41 recite the same limitations as claim 4 and are rejected for the reasons given.
With regards to claims 37 and 42, Guo in view of Lee teaches claims 35 and 40, respectively. Claims 37 and 42 recite the same limitations as claim 14 and are rejected for the reasons given.
The combined teaching described above will be referred as Guo + Lee + Yu hereafter.
Claims 6, 19 and 30 are rejected under 35 U.S.C. 103 as being unpatentable over Guo et al., "SpotTune: Transfer Learning through Adaptive Fine-tuning," arXiv:1811.08737v1 (Nov. 2018), published at CVPR 2019 [hereafter Guo], in view of Lee et al., "Surgical Fine-Tuning Improves Adaptation to Distribution Shifts," arXiv:2210.11466v1 (Oct. 20, 2022) [hereafter Lee], Yu et al., "BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning," arXiv:1805.04687v2 (Apr. 8, 2020), CVPR 2020 [hereafter Yu] and Afrouzi et al., US_PATENT NO.12736962B1 [hereafter Afrouzi].
With regards to claim 6, Guo + Lee + Yu teaches
"The method of claim 4”.
Guo + Lee + Yu does not explicitly detail “wherein the at least one host vehicle includes an autonomous or a semi-autonomous vehicle".
However, Afrouzi teaches “wherein the at least one host vehicle includes an autonomous or a semi-autonomous vehicle.” (Afrouzi, FIG.107, 136-138,
PNG
media_image10.png
446
732
media_image10.png
Greyscale
)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee + Yu and Afrouzi before him or her, to modify the subset-duplication fine-tuning of Guo + Lee + Yu to autonomous vehicle of Afrouzi
The motivation for doing so would have been for digital representation of environment (Afrouzi, Abstract).
With regards to claim 19, Guo + Lee + Yu teaches
"The method of claim 18”.
Guo + Lee + Yu does not explicitly detail “wherein the at least one host vehicle includes an autonomous or a semi-autonomous vehicle".
However, Afrouzi teaches “wherein the at least one host vehicle includes an autonomous or a semi-autonomous vehicle.” (Afrouzi, FIG.107, 136-138,
PNG
media_image10.png
446
732
media_image10.png
Greyscale
)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee + Yu and Afrouzi before him or her, to modify the subset-duplication fine-tuning of Guo + Lee + Yu to autonomous vehicle of Afrouzi
The motivation for doing so would have been for digital representation of environment (Afrouzi, Abstract).
With regards to claim 30, Guo + Lee + Yu teaches
"The method of claim 29”.
Guo + Lee + Yu does not explicitly detail “wherein the at least one host vehicle includes an autonomous or a semi-autonomous vehicle".
However, Afrouzi teaches “wherein the at least one host vehicle includes an autonomous or a semi-autonomous vehicle.” (Afrouzi, FIG.107, 136-138,
PNG
media_image10.png
446
732
media_image10.png
Greyscale
)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, having the teachings of Guo + Lee + Yu and Afrouzi before him or her, to modify the subset-duplication fine-tuning of Guo + Lee + Yu to autonomous vehicle of Afrouzi
The motivation for doing so would have been for digital representation of environment (Afrouzi, Abstract).
Additional Relevant Art
The prior art made of record is considered pertinent to applicant’s disclosure and is recorded on Form PTO-892. Applicant is required under 37 C.F.R. § 1.111 (c) to consider these references fully when responding to this action, with particular attention paid to:
a. Rabinowitz et al., US-PGPUB NO. 20170337464A1 [hereafter Rabinowitz] shows a progressive neural network for learning a sequence of machine-learning tasks.
b. Desjardins et al., US-PGPUB NO. 20190236482A1 [hereafter Desjardins] shows training a machine learning model on a second task after it has already been trained on a first task, while protecting its performance on the first task.
c. Annapureddy et al., US-PGPUB NO. 20160217369A1 [hereafter Annapureddy] shows compressing a neural network and then fine-tuning either the whole network or only a selected subset of layers. It also shows choosing parameters layer by layer, by measuring accuracy on a validation set and keeping the value for which the accuracy drop stays within a threshold, including by a greedy search.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TSU-CHANG LEE whose telephone number is 571-272-3567. The fax number is 571-273-3567.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Omar Fernandez Rivas, can be reached 571-272-2589.
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.
/TSU-CHANG LEE/
Primary Examiner, Art Unit 2128
