DETAILED ACTION
This action is in response to the communications filed on 06/30/2026 in which claims 1-20 are amended, and claims 1-20 are pending.
--
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 .
Claim Objections
Claim 1 is objected to because of the following informalities:
In claim 1, “classify… a plurality of variances of an unlabled input…” should be “classify… a plurality of variances of an unlabeled input…” (a typo in “unlabled”)
Appropriate correction is required.
Claim Rejections - 35 USC § 101
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.
-
Claims 1-20 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more
Step 1: Claims 1-7 recite a processor comprising circuits. Claims 8-14 recite a method. Claims 15-20 recite a computer system comprising processors and memory. Therefore, claims 1-7 are directed to a machine, claims 8-14 are directed to a process, and claims 15-20 are directed to a machine.
With respect to claims 1, 8 and 15:
2A Prong 1: The claim recites a judicial exception.
classify… a plurality of variances of an unlabled input to the one or more pre-trained neural networks based, at least in part, on a prompt to the one or more pre-trained neural networks; (mental process – evaluation or judgement, classifying variances of an input based on a prompt)
update the prompt using a subset of the plurality of variances and the one or more pre-trained neural networks, the subset selected based, at least in part, on classification confidence scores for the plurality of variances; and (mental process – evaluation; updating/adjusting the prompt using variances and the pre-trained neural networks)
2A Prong 2: The judicial exception is not integrated into a practical application.
(claim 1) one or more circuits (claim 15) one or more processors and memory storing executable instructions that, if performed by the one or more processors, cause the one or more processors to (mere instructions to apply an exception, (2) Whether the claim invokes computers - MPEP 2106.05(f); generic computer components)
… using one or more pre-trained neural networks (mere instructions to apply an exception – MPEP 2106.05(f), (3) The particularity or generality of the application of the judicial exception; using pre-trained neural networks)
apply the updated prompt to the one or more pre-trained neural networks (mere instructions to apply an exception – MPEP 2106.05(f), (3) The particularity or generality of the application of the judicial exception; applying the prompt to the pre-trained neural networks)
Since the claim as a whole, looking at the additional elements individually and in combination, does not contain any other additional elements that are indicative of integration into a practical application, the claim is directed to an abstract idea.
2B: The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
(claim 1) one or more circuits (claim 15) one or more processors and memory storing executable instructions that, if performed by the one or more processors, cause the one or more processors to (mere instructions to apply an exception, (2) Whether the claim invokes computers - MPEP 2106.05(f); generic computer components)
… using one or more pre-trained neural networks (mere instructions to apply an exception – MPEP 2106.05(f), (3) The particularity or generality of the application of the judicial exception; using pre-trained neural networks)
apply the updated prompt to the one or more pre-trained neural networks (mere instructions to apply an exception – MPEP 2106.05(f), (3) The particularity or generality of the application of the judicial exception; applying the prompt to the pre-trained neural networks)
Considering the additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
With respect to claims 2, 9 and 16:
2A Prong 1: The claim recites a judicial exception.
wherein the unlabeled input to the neural network comprises an image (mental process – evaluation or judgement; claim 1 recites classifying variances of an input based on a prompt, which is an abstract idea. Specifying the details of the input does not change the scope of the claim.)
With respect to claim 3:
2A Prong 1: The claim recites a judicial exception.
wherein the prompt is an input to the one or more neural networks. (mental process – evaluation or judgement; claim 1 recites classifying variances of an input based on a prompt, which is an abstract idea. Specifying the details of the prompt does not change the scope of the claim.)
With respect to claims 4, 11 and 18:
2A Prong 2: The judicial exception is not integrated into a practical application.
wherein the one or more neural networks include a pre-trained vision language model (mere instructions to apply an exception – MPEP 2106.05(f), (3) The particularity or generality of the application of the judicial exception; Claim 1 recites using neural networks, which is mere instructions to apply an exception. Specifying the details of the neural networks does not cause the limitation to integrate the exception into a practical application.)
Since the claim as a whole, looking at the additional elements individually and in combination, does not contain any other additional elements that are indicative of integration into a practical application, the claim is directed to an abstract idea.
2B: The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception.
wherein the one or more neural networks include a pre-trained vision language model (mere instructions to apply an exception – MPEP 2106.05(f), (3) The particularity or generality of the application of the judicial exception; Claim 1 recites using neural networks, which is mere instructions to apply an exception. Specifying the details of the neural networks does not cause the limitation to be significantly more than the judicial exception.)
Considering the additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
With respect to claims 5 and 19:
2A Prong 1: The claim recites a judicial exception.
wherein the unlabeled input comprises an image and the plurality of variances comprise one or more randomly augmented views of the image (mental process – evaluation or judgement; claim 1 recites classifying variances of an input based on a prompt, which is an abstract idea. Specifying the details of the input and the variances does not change the scope of the claim.)
With respect to claims 6 and 20:
2A Prong 1: The claim recites a judicial exception.
wherein the prompt is tuned during inferencing based, at least in part, on minimizing entropy of the plurality of variances (mental process – evaluation or judgement, the prompt is adjusted based on filtering out views with high-entropy; in light of specification [0076] "... filter out views that generate high-entropy (i.e., low-confidence) predictions... and retain views that generate low-entropy (e.g., high-confidence) predictions")
With respect to claim 7:
2A Prong 1: The claim recites a judicial exception.
wherein the prompt is tuned based, at least in part, on classifying the plurality of variances based, at least in part, on removing one or more of the variances from the plurality of variances. (mental process – evaluation or judgement; adjusting the prompt based on classifying the variances based on removing variances)
With respect to claim 10:
2A Prong 1: The claim recites a judicial exception.
wherein the unlabeled input comprises a single image and wherein the prompt is based, at least in part, on content of the single image (mental process – evaluation or judgement; claim 8 recites classifying variances of an input based on the prompt, which is an abstract idea. Specifying the details of the input and the prompt does not change the scope of the claim.)
With respect to claim 12:
2A Prong 1: The claim recites a judicial exception.
further comprising: generating, randomly, the plurality of variances based, at least in part, on the unlabeled input (mental process – evaluation or judgement,--- randomly generating the variances based on the input)
With respect to claim 13:
2A Prong 1: The claim recites a judicial exception.
further comprising: generating one or more confidence metrics of the plurality of variances. (mental process – evaluation or judgement,--- generating confidence metrices of the variances)
With respect to claim 14:
2A Prong 1: The claim recites a judicial exception.
wherein the subset of the plurality of variances is based, at least in part, on an average value of the confidence scores of the plurality of variances. (mental process – evaluation or mathematical concept,--- updating the prompt using the subset of the variances based on an average value of the scores)
With respect to claim 17:
2A Prong 1: The claim recites a judicial exception.
wherein the prompt describes elements of the unlabeled input (mental process – evaluation or judgement; claim 15 recites classifying variances of an input based on a prompt, which is an abstract idea. Specifying the details of the prompt does not change the scope of the claim.)
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-20 rejected under 35 U.S.C. 103 as being unpatentable over Schulter (US 20230281858 A1, provisional date 20220307) in view of He ("Towards Open-vocabulary Scene Graph Generation with Prompt-based Finetuning" 20220907)
In regard to claims 1, 8 and 15, Schulter teaches: A processor, comprising: one or more circuits to: (Schulter, [0017]-[0020] "FIG. 1 is a block diagram showing an exemplary computing device 100... The computing device 100 is configured to perform unlabeled image mining with vision and language models for improving object detection… The processor 110 may be embodied as any type of processor capable of performing the functions described herein."; see Fig. 1)
classify, using one or more pre-trained neural networks, a plurality of variances of an unlabled input to the one or more pre-trained neural networks (Schulter, [0028] "FIG. 2 is a block diagram showing an exemplary system 200… The unlabeled data 201 [an unlabeled input] includes a set of RGB (Red, Green, Blue) images without any additional annotations… The proposal generator 210 is a class-agnostic proposal mechanism. Any existing mechanism can be used that takes an image and returns bounding boxes of potential objects. [a plurality of variances]"; [0033] "The vision and language model 230 is a pre-trained vision and language model, [pre-trained neural networks] which was trained with large-scale pairs of images and corresponding captions. The V & L model 230 includes an image encoder and a text encoder."; [0037] "Further regarding the vision and language model PLM 220, the same includes a region scorer 233 that applies the V & L model 230 on each of the regions to estimate a category name and a confidence, which then gives final pseudo labels 240. [classify, using V& L model, regions (bounding boxes) of an labeled image]"; [0075] "The present invention then crops image patches based on those regions [a plurality of variances] and feeds them into the CLIP image-encoder to obtain an embedding in the CLIP vision-and-language space. Using the corresponding CLIP text-encoder and template text prompts, the present invention generates embeddings for category names that are desired for the specific task."; classify, using the CLIP model, bounding boxes of an unlabeled image; classifying (or classification) corresponds to the category name or pseudo label) based, at least in part, on a prompt to the one or more pre-trained neural networks; (Schulter, [0043]-[0044] "For a given list of category names (like 'car', 'person', 'apple') 330, which can be defined arbitrarily for any desired task, a template sentence is created for each category name, c, as 'A photo of a <CLS>', where '<CLS>' is replaced with the category name c. [based on a (text) prompt] The sentences of all categories are put through the V & L text encoder 340, which returns a set of vectors yc, one for each category c, all of them of the same dimensionality as x. The image vector x is then compared with each text vector yc via cosine similarity by the scorer 350. A SoftMax function is then applied over all similarities between x and yc, which gives a probability (between 0 and 1) for a region r to be of category c. [classifying (category names) based on a (text) prompt] In an embodiment, the present invention takes the maximum probability..."; [0075] "Using the corresponding CLIP text-encoder and template text prompts [based on a (text) prompt], the present invention generates embeddings for category names [classifying (category
PNG
media_image1.png
440
1117
media_image1.png
Greyscale
names) based on a (text) prompt] that are desired for the specific task.")
Schulter does not teach, but He teaches: update the prompt using a subset of the plurality of variances and the one or more pre-trained neural networks, (He, p. 8, Prompt-based finetuning for Ov-SGG., "we propose two prompt-based finetuning strategies, hard prompt and soft visual-relation prompt (SVRP), for Ov-SGG based on our pre-trained VRM. [using the pre-trained neural networks]... thus supports predictions over unseen object labels at inference time. The core of prompt-based learning [update the prompt] is the design of templates to convert the input sequence, x_in, into a textual prompt X_pro with unfilled cloze-style slots. The VRM then makes a prediction by filling the slot with the candidate of the maximum probability in the label space. [using a subset of the variances (chosen by selecting the candidate with the max probability) based on classification confidence scores (the probability)]") the subset selected based, at least in part, on classification confidence scores for the plurality of variances; and (He, p. 7, Standard finetuning strategy. "Let r_so denote the union region of subject r_s and object r_o, and r_1, ..., r_m are the object proposals overlapping with r_so. We feed them into the pretrained image encoder as in Eq. (1) to produce visual embeddings h.... where h_r is the embedding of the union region of r_s and r_o... "; p. 8, Hard prompt based finetuning. "Since relations in SGG are represented as SPO triples, we could readily formulate a relation prompt as follows: X_pro = T(x_in) = [CLS]xs [MASK]xo[SEP] (5) [update the prompt] where xs and xo denote the label of the subject and object, and [MASK] is the slot for candidate predicate labels. Note that the labels of xs and xo are produced in an zero-shot manner describing in the second term of Eq. (4). Then, we could predict the relation label by: y^ = argmax P(f_fill(X_pro, M(y))|h_r; θ) (6) where f_fill(.) is a function that fills the slot [MASK] in Xpro with a label word M(r)∈V, hr = LN(hs, hso, ho) is defined as in Eq. (4), [using h_r, which is based on a subset of variances rs and ro] and θ are the parameters of the linear projection function LN (updated) and of VRM (frozen). For the prediction score P [classification confidence scores], we also use Cosine similarity to calculate it: P (...) = exp (cos.../Σ cos... )(7)"; see Fig. 2, Right: update X'_pro [update the prompt] using Εi pre-trained image encoder [using the pre-trained neural networks], and X_pro is updated based on the maximum probability [classification confidence scores] (the prediction score P of the predicate [MASK] for the triple subject-predicate-object), which is generated based on based on hr (which is based on rs and ro [the subset of variances]))
PNG
media_image2.png
536
540
media_image2.png
Greyscale
apply the updated prompt to the one or more pre-trained neural networks. (He, p. 9, Soft visual-relation prompt (SVRP) based finetuning. "we view X'_pro as an input of a masked language model and feed it into the pretrained TexTrans network [apply the updated prompt to the pre-trained neural networks] to find the token that maximizes the probability of [MASK].")
It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to have modified Schulter to incorporate the teachings of He by including two prompt-based techniques to finetune the pre-trained model. Doing so would achieve strong performance. (He, p. 1, "we propose a two-step method which firstly pre-trains on large amounts of coarse-grained region-caption data and then leverages two prompt-based techniques to finetune the pre-trained model without updating its parameters... On extensive experiments on three benchmark datasets, Visual Genome, GQA and Open-Image, our method significantly outperforms recent, strong SGG methods on the setting of Ov-SGG...")
Claims 8 and 15 recite substantially the same limitation as claim 1, therefore the rejection applied to claim 1 also apply to claims 8 and 15. In addition, Schulter teaches: (claim 15) A computer system, comprising: one or more processors and memory storing executable instructions that, if performed by the one or more processors, cause the one or more processors to: (Schulter, [0017]-[0020] "FIG. 1 is a block diagram showing an exemplary computing device 100... The computing device 100 is configured to perform unlabeled image mining with vision and language models for improving object detection… The processor 110 may be embodied as any type of processor capable of performing the functions described herein."; see Fig. 1)
In regard to claims 2, 9 and 16, Schulter teaches: wherein the unlabeled input to the neural network comprises an image. (Schulter, [0028] "FIG. 2 is a block diagram showing an exemplary system 200… The unlabeled data 201 [an unlabeled input] includes a set of RGB (Red, Green, Blue) images without any additional annotations.")
In regard to claim 3, Schulter teaches: wherein the prompt is an input to the one or more neural networks. (Schulter, [0043]-[0044] "For a given list of category names (like 'car', 'person', 'apple') 330, which can be defined arbitrarily for any desired task, a template sentence is created for each category name, c, as 'A photo of a <CLS>', where '<CLS>' is replaced with the category name c. The sentences of all categories are put through the V & L text encoder 340, [a (text) prompt is an input to the text encoder] which returns a set of vectors yc, one for each category c, all of them of the same dimensionality as x.")
In regard to claims 4, 11 and 18, Schulter teaches: wherein the one or more pre-trained neural networks include a pre-trained vision language model. (Schulter, [0033] "The vision and language model 230 is a pre-trained vision and language model, [a pre-trained vision language model] which was trained with large-scale pairs of images and corresponding captions. The V & L model 230 includes an image encoder and a text encoder.")
In regard to claims 5 and 19, Schulter teaches: wherein the unlabeled input comprises an image and the plurality of variances comprise one or more randomly augmented views of the image. (Schulter, [0028] "FIG. 2 is a block diagram showing an exemplary system 200… The unlabeled data 201 [an unlabeled input] includes a set of RGB (Red, Green, Blue) images without any additional annotations."; [0050] "At block 425, duplicate each region using into an original scale region and a larger scale region having same bounding box centers to obtain a first and a second image patch. In an embodiment, the larger scale is 1.5×. Of course, other scales can also be used [randomly augmented views], while maintaining the spirit of the present invention."; randomly cropped regions at different scales)
In regard to claims 6 and 20, Schulter does not teach, but He teaches: wherein the prompt is tuned during inferencing based, at least in part, on minimizing entropy of the plurality of variances. (He, p. 8, Prompt-based finetuning for Ov-SGG., "we propose two prompt-based finetuning [the prompt is tuned] strategies, hard prompt and soft visual-relation prompt (SVRP), for Ov-SGG based on our pre-trained VRM. A key advantage of our prompt-based tuning strategies is that they allow our pre-trained VRM to be optimized on task-specific data without updating its parameters. Doing so avoids altering the learned (open-vocabulary) knowledge during training, and thus supports predictions over unseen object labels at inference time. [during inferencing] The core of prompt-based learning is the design of templates to convert the input sequence, x_in, into a textual prompt X_pro with unfilled cloze-style slots. The VRM then makes a prediction by filling the slot with the candidate of the maximum probability in the label space. [selecting the candidate with the max probability (low entropy), i.e. minimizing entropy]"; p. 8, Hard prompt based finetuning. "we could predict the relation label by: y^ = argmax P(f_fill(X_pro, M(y))|h_r; θ) (6) where f_fill(.) is a function that fills the slot [MASK] in Xpro with a label word M(r)∈V, hr = LN(hs, hso, ho) is defined as in Eq. (4), and θ are the parameters of the linear projection function LN (updated) and of VRM (frozen). For the prediction score P, we also use Cosine similarity to calculate it: P (...) = exp (cos.../Σ cos... )(7) "; in light of specification [0076] "... filter out views that generate high-entropy (i.e., low-confidence) predictions... and retain views that generate low-entropy (e.g., high-confidence) predictions")
The rationale for combining the teachings of Schulter and He is the same as set forth in the rejection of claim 1.
In regard to claim 7, Schulter does not teach, but He teaches: wherein the prompt is tuned based, at least in part, on classifying the plurality of variances based, at least in part, on removing one or more of the variances from the plurality of variances. (He, p. 8, Prompt-based finetuning for Ov-SGG., "we propose two prompt-based finetuning [the prompt is tuned ] strategies, hard prompt and soft visual-relation prompt (SVRP), for Ov-SGG based on our pre-trained VRM... The core of prompt-based learning is the design of templates to convert the input sequence, x_in, into a textual prompt X_pro with unfilled cloze-style slots. The VRM then makes a prediction by filling the slot with the candidate of the maximum probability in the label space. [using a subset of the variances (chosen by selecting the candidate with the max probability, i.e. removing candidates with lower probability)]"; p. 8, Hard prompt based finetuning. "Since relations in SGG are represented as SPO triples, we could readily formulate a relation prompt as follows: X_pro = T(x_in) = [CLS]xs [MASK]xo[SEP] (5) [update the prompt] where xs and xo denote the label of the subject and object, and [MASK] is the slot for candidate predicate labels. [tuning based on classifying the variances (subject-predicate-object)] Note that the labels of xs and xo are produced in an zero-shot manner describing in the second term of Eq. (4). Then, we could predict the relation label by: y^ = argmax P(f_fill(X_pro, M(y))|h_r; θ) (6)")
The rationale for combining the teachings of Schulter and He is the same as set forth in the rejection of claim 1.
In regard to claim 10, Schulter teaches: wherein the unlabeled input comprises a single image and wherein the prompt is based, at least in part, on content of the single image. (Schulter, [0035] "Further regarding the vision and language model PLM 220, the same includes the region extractor 222 which applies the proposal generator 210 on the unlabeled image 221 [a single image] and returns a set of regions R."; [0043]-[0044] "For a given list of category names (like 'car', 'person', 'apple') 330, which can be defined arbitrarily for any desired task, a template sentence is created for each category name, c, as 'A photo of a <CLS>', where '<CLS>' is replaced with the category name c. [the (text) prompt] The sentences of all categories are put through the V & L text encoder 340, which returns a set of vectors yc, one for each category c, all of them of the same dimensionality as x. The image vector x is then compared with each text vector yc via cosine similarity by the scorer 350. A SoftMax function is then applied over all similarities between x and yc, which gives a probability (between 0 and 1) for a region r to be of category c. [content of the image] In an embodiment, the present invention takes the maximum probability...")
In regard to claim 12, Schulter teaches: further comprising: generating, randomly, the plurality of variances based, at least in part, on the unlabeled input. (Schulter, [0028] "FIG. 2 is a block diagram showing an exemplary system 200… The unlabeled data 201 [an unlabeled input] includes a set of RGB (Red, Green, Blue) images without any additional annotations."; [0050] "At block 425, duplicate each region using into an original scale region and a larger scale region having same bounding box centers to obtain a first and a second image patch. In an embodiment, the larger scale is 1.5×. Of course, other scales can also be used [randomly generating the variances], while maintaining the spirit of the present invention."; randomly cropped regions at different scales)
In regard to claim 13, Schulter teaches: further comprising: generating one or more confidence metrics of the plurality of variances. (Schulter, [0035] "Further regarding the vision and language model PLM 220, the same includes the region extractor 222 which applies the proposal generator 210 on the unlabeled image 221 and returns a set of regions R. Each region comes with a bounding box (4 coordinates in the 2D image space) and an 'objectness' score that indicates confidence of the proposal generator about its prediction."; [0048]-[0049] "At block 415, obtaining, by applying a proposal generator to a RGB image from the set, a set of regions [the variances] in the RGB image that include one or more of the potential objects, a bounding box for the one or more of the potential objections, and an objectness score indicating a region prediction confidence. [confidence metrics] At block 420, obtain, by a region scorer, a category name from a fixed set of category names and a confidence for the category name responsive to the objectness score.")
In regard to claim 14, Schulter does not teach, but He teaches: wherein the subset of the plurality of variances is based, at least in part, on an average value of the confidence scores of the plurality of variances. (He, p. 8, Prompt-based finetuning for Ov-SGG., "we propose two prompt-based finetuning strategies, hard prompt and soft visual-relation prompt (SVRP), for Ov-SGG based on our pre-trained VRM... The VRM then makes a prediction by filling the slot with the candidate of the maximum probability in the label space. [using a subset of the variances (chosen by selecting the candidate with the max probability)]"; p. 8, Hard prompt based finetuning. "we could predict the relation label by: y^ = argmax P(f_fill(X_pro, M(y))|h_r; θ) (6) where f_fill(.) is a function that fills the slot [MASK] in Xpro with a label word M(r)∈V, hr = LN(hs, hso, ho) is defined as in Eq. (4), and θ are the parameters of the linear projection function LN (updated) and of VRM (frozen). For the prediction score P, we also use Cosine similarity to calculate it: P (...) = exp (cos.../Σ cos... )(7) [an average value of the confidence scores of the variances]")
The rationale for combining the teachings of Schulter and He is the same as set forth in the rejection of claim 1.
In regard to claim 17, Schulter teaches: wherein the prompt describes elements of the unlabeled input. (Schulter, [0043]-[0044] "For a given list of category names (like 'car', 'person', 'apple') 330, which can be defined arbitrarily for any desired task, a template sentence is created for each category name, c, as 'A photo of a <CLS>', where '<CLS>' is replaced with the category name c. [the (text) prompt] The sentences of all categories are put through the V & L text encoder 340, which returns a set of vectors yc, one for each category c, all of them of the same dimensionality as x. The image vector x is then compared with each text vector yc via cosine similarity by the scorer 350. A SoftMax function is then applied over all similarities between x and yc, which gives a probability (between 0 and 1) for a region r to be of category c. [regions/elements of the unlabeled input] In an embodiment, the present invention takes the maximum probability...")
Response to Arguments
Applicant's amendments with respect to 35 USC 112(b) have been fully considered and are sufficient to overcome the objections. The 112(b) rejection has been withdrawn.
Applicant's arguments with respect to the rejection of the claims under 35 U.S.C. 103 have been fully considered but they are moot:
Argument: (p. 7-9) … Zhou in view of Zhang fails to teach… classify… update… apply…
Response: the arguments do not apply to the references (Schulter and He) being used in the current rejection.
Applicant's arguments with respect to the rejection of the claims under 35 U.S.C. 101 have been fully considered but they are not persuasive:
Argument: (p. 10) … Applicant's claims recite specific features that cannot be performed practically by the human mind. For instance, using a pre-trained neural network to classify a plurality of variances cannot reasonably be performed in the human mind. Additionally,
Response: The claim recites the limitation “classify… a plurality of variances.” The limitation is evaluated at step 2A prong one and is a mental process. Further, the claim recites “using a pre-trained neural network.” The limitation is evaluated at step 2A prong 2 and step 2B as an addition element, and amounts to mere instructions to apply an exception – MPEP 2106.05(f).
Argument: (p. 10) Furthermore, tuning a neural network prompt using a pre-trained neural network by classifying a plurality of variances, updating the prompt using a subset of the plurality of variances (based on classification confidence scores), and then applying the updated prompt to the pre-trained neural network provides a prompt producing better neural network predictions, esp. in zero shot situations where the neural network was not specifically trained on the particular types of input. For instance, such a tuned prompt aids the neural network to make more consistent predictions when classifying images with elements on which the neural network was not trained, as described in Applicant's specification. Thus, Applicant's recited subject matter adds "significantly more" than an abstract idea and provides a further improvement…
Response: The claim recites “update the prompt using a subset of the plurality of variances and the one or more pre-trained neural networks, the subset selected based, at least in part, on classification confidence scores for the plurality of variances; and.” The limitation is evaluated under step 2A prong 1 as an abstract idea describing a prompt text is updated using variances and the pre-trained neural networks. The claim can recite a mental process if they are being performed on a computer, and a mental process cannot provide improvements. See MPEP 2106.05(a): “It is important to note, the judicial exception alone cannot provide the improvement.”
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 SU-TING CHUANG whose telephone number is (408)918-7519. The examiner can normally be reached Monday - Thursday 8-5 PT.
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, Usmaan Saeed can be reached at (571) 272-4046. 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.
/S.C./Examiner, Art Unit 2146
/USMAAN SAEED/Supervisory Patent Examiner, Art Unit 2146
