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 .
Detailed Action
The following action is in response to the communication(s) received on 06/30/2026.
As of the claims filed 06/30/2026:
Claims 1-20 are pending.
Claims 1, 11, and 20 have been amended.
Claims 1, 11, and 20 are independent claims.
Response to Arguments
Applicant’s arguments filed 06/30/2026 have been fully considered, but are not fully persuasive.
With respect to the rejection under 35 USC § 101:
Applicant asserts that constructing the Boolean rules using a trainable, differentiable neural logic model is a functional role, not mere math, in expressing how the AI model combined feature vectors to make the inference, and thus does not recite a mathematical formula, equation, or calculation (p.10 last ¶). Examiner respectfully submits that constructing Boolean rules was identified as an abstract idea and not math, since in the broadest reasonable interpretation, it involves observing the patterns of a trained neural network. The trained neural network is performing the construction of the Boolean rules, as opposed to the training method of a neural network involving Boolean rule construction.
Applicant further asserts that constructing the Boolean rules requires using a trainable, differentiable neural logic model and thus cannot be practically performed in the human mind (p.11 ¶1). As explained above, the differentiable neural logic model is merely being applied to perform the construction of the Boolean rules, which is an abstract idea in the broadest reasonable interpretation.
Applicant further asserts that, in the “capturing” step, a human cannot observe the operation of the internal layers of a neural network (p.11 ¶1b). Examiner respectfully disagrees: although extracting feature vectors is not positively recited in the claims, it merely involves gathering the information of the layers of a trained neural network, of which is merely an algorithm. As currently recited, the claimed invention involves constructing Boolean rules that correspond to an algorithm step (a feature vector of an artificial intelligence model). As this “capturing” step does not involve training the neural network, this is not impractical to perform in the human mind.
Applicant further asserts that the additional elements identified in the eligibility analysis provide an improvement in producing AI explanations that are deterministic, grounded in the internal computation of the model, and understandable to a user (p.13 ¶2). Examiner respectfully submits that, as explained above, producing an explanation behind a neural network's decision is merely an abstract idea being improved upon and does not indicate an improvement in a particular technology, such as training a neural network.
Applicant further asserts that the claims do not recite a generic computer, as they recite a particular mechanism and a particular relationship between the output and the AI model’s reasoning (p.13 ¶3). Examiner respectfully submits that the analysis requires deliberation on whether the role of the computer produces a practical application to a particular technology. However, as explained above, the construction of the Boolean rule involves merely observing the trained neural network, which is an abstract idea, instead of a particular technology, such as training a neural network.
Applicant further asserts that displaying the Boolean rules corresponds to the culmination of the claimed technical pipeline of delivering a model-grounded explanation to the user (p.13 last ¶). Examiner respectfully disagrees: displaying the result of the Boolean rules extracted from the observed feature vectors to the user does not provide significantly more than the abstract idea, as outputting data is not the particular technology being improved upon. Instead, the crux of the invention involves generating the explanation behind the neural network's decision, which is an abstract idea.
Applicant further asserts that the ordered combination of elements provides something significantly more than the abstract idea itself, as the claimed approach recites an inventive concept by linking training-time structure and inference-time explanation via a particular model (p.14 ¶1). Examiner respectfully submits that, as explained above, the identified additional elements are not the crux of the invention. For example, the claimed invention is not about a particular method of training a machine learning model or outputting data; instead, the invention is merely directed to generating an explanation for a model’s decision, which is an abstract idea.
Thus, the instant claims remain directed to abstract ideas, which are not patentable.
With respect to the rejection under 35 USC § 103,
Applicant asserts that Payani’s predicates merely describe what is present in an image and not “how the artificial intelligence model combined” features to arrive at a particular inference or classification (p.15¶3). Examiner respectfully submits that the argument has been carefully considered but is moot in view of the new art taught by Riley in the combination of Chen/Payani/Riley. More specifically, although Chen/Payani does not teach using the predicate conversions to explain how the artificial intelligence made a particular classification, Riley further teaches that the formalized logic trained obtained from the feature maps can be used to explain the reasoning behind the classification (Riley [p.13 right ¶2] When asked to explain this answer (“Please explain this answer”),… The extracted features were processed using the trained decision tree… The decision tree output, image feature vector, and input question, were processed by the previously trained RNN to provide the answer type and the particular answer described above)
Applicant further asserts that the prior art does not teach concurrently displaying the representative sample, the inference, and the Boolean rules (p.16¶2). Examiner respectfully submits that Chen’s method shows displaying the explanation of the classification decision through showing the feature maps and the inference, and Riley, via the combination of Chen/Payani/Riley as explained above, further shows that the displaying can further include the Boolean representations extracted from the trained neural network, thus teaching that the explanation can have a concurrent display of these three elements.
Applicant further asserts that the additional independent claims 11 and 20 are patentable for the same reasons as claim 1. Examiner respectfully submits that the claims remain rejected for the same reasons provided above.
Applicant further asserts that the dependent claims 2-10 and 12-19 are patentable by virtue of dependency to their respective parent claims. Examiner respectfully submits that the claims remain rejected at least by virtue of dependency to their rejected parent claims.
Thus, Chen/Payani/Riley teaches the methods and limitations recited in the claims.
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 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…method, thus a process, one of the four statutory categories of patentable subject matter (Step 1). However, Claim 1 further recites:
making an inference regarding input data, which is an evaluation or judgement that can be performed in the human mind;
selecting, based at least in part on the one or more feature vectors, a representative sample from a training dataset used to train the artificial intelligence model, which is an evaluation or judgement that can be performed in the human mind;
wherein selecting comprises comparing the one or more feature vectors to stored feature vectors…, which is an evaluation or judgement that can be performed in the human mind;
constructing…one or more Boolean rules in disjunctive normal form that correspond to the one or more feature vectors, respectively, wherein the one or more Boolean rules express how the artificial intelligence model combined the one or more feature vectors to make the inference regarding the input data, which is an evaluation or judgement that can be performed in the human mind.
Thus, the claim recites an abstract idea under Step 2A Prong 1.
Under Step 2A Prong 2, the claim does not include any additional elements which integrate the abstract idea into a practical application, since the additional elements consist of:
…computer-implemented; …using an artificial intelligence model, as the performance of an abstract idea on a computer is not more than instructions to "apply it" on a computer, which by MPEP 2106.05(f) cannot integrate an abstract idea into a practical application;
receiving input data from a user, which is merely an insignificant extra-solution activity of data gathering, which by MPEP 2106.05(g) cannot integrate an abstract idea into a practical application;
capturing one or more feature vectors... to make the inference, which is merely an insignificant extra-solution activity of data gathering, which by MPEP 2106.05(g) cannot integrate an abstract idea into a practical application;
…from a predetermined feature layer of the artificial intelligence model that were used by the artificial intelligence model, which merely specifies the particular field of use or particular technological environment in which the abstract idea is to be performed, which by MPEP 2106.05(h) cannot integrate the abstract idea into a practical application.
causing the representative sample, the inference, and the one or more Boolean rules expressing how the artificial intelligence model combined the one or more feature vectors to make the inference to be concurrently displayed on a display device associated with the user, which is merely an insignificant extra-solution activity of data output, which by MPEP 2106.05(g) cannot integrate an abstract idea into a practical application.
associated with respective samples of the training dataset in a feature-vector store generated from the training dataset, which merely specifies the particular field of use or particular technological environment in which the abstract idea is to be performed, which by MPEP 2106.05(h) cannot integrate the abstract idea into a practical application.
and selecting the representative sample based on which of the stored feature vectors the artificial intelligence model considers to be similar to the one or more feature vectors, which merely specifies the particular field of use or particular technological environment in which the abstract idea is to be performed, which by MPEP 2106.05(h) cannot integrate the abstract idea into a practical application.
using a trainable, differentiable neural logic model trained via gradient optimization alongside the artificial intelligence model, as the performance of an abstract idea on a computer is not more than instructions to "apply it" on a computer, which by MPEP 2106.05(f) cannot integrate an abstract idea into a practical application;
Thus, the claim is directed towards an abstract idea.
Further, the additional elements, alone or in combination, do not provide significantly more than the abstract idea itself, because the particular field of use or particular technological environment (MPEP 2106.05(h)), implementation on a computer (MPEP 2106.05(f)),and the activity of data gathering/output(MPEP 2106.05(g)) cannot provide significantly more, as storing and retrieving information in memory is well understood, routine, and conventional (MPEP 2106.05(d)(II)(iv)) as receiving or transmitting data over a network is well understood, routine, and conventional (MPEP 2106.05(d)(II)(i)) and the combination of additional elements does not provide an inventive concept. Thus, the claim is ineligible.
Claim 2, dependent upon Claim 1, further recites no additional abstract ideas. However:
Under Step 2A Prong 2, the claim does not include any additional elements which integrate the abstract idea into a practical application, since the additional elements consist of:
the artificial intelligence model comprises a classifier, which are mere details of the performance of an abstract idea on a computer, which is not more than instructions to "apply it" on a computer, which by MPEP 2106.05(f) cannot integrate an abstract idea into a practical application.
Thus, the claim is directed towards an abstract idea.
Further, under Step 2B, the additional element does not provide significantly more than the abstract idea itself, because implementation on a computer (MPEP 2106.05(f)) cannot provide significantly more. Thus, the claim is ineligible.
Claim 3, dependent upon Claim 1, further recites
the input data comprises an image, which is merely a detail of an abstract idea (making an inference regarding input data).
Thus, the claim recites an abstract idea under Step 2A Prong 1.
Under Step 2A Prong 2 and 2B, the claim does not recite any new additional elements which could integrate the abstract idea into a practical application or provide significantly more than the abstract idea itself. Thus, the claim is ineligible.
Claim 4, dependent upon Claim 1, further recites no additional abstract ideas. However:
Under Step 2A Prong 2, the claim does not include any additional elements which integrate the abstract idea into a practical application, since the additional elements consist of:
the artificial intelligence model comprises a neural network, which are mere details of the performance of an abstract idea on a computer, which is not more than instructions to "apply it" on a computer, which by MPEP 2106.05(f) cannot integrate an abstract idea into a practical application.
Thus, the claim is directed towards an abstract idea.
Further, under Step 2B, the additional element does not provide significantly more than the abstract idea itself, because implementation on a computer (MPEP 2106.05(f)) cannot provide significantly more. Thus, the claim is ineligible.
Claim 5, dependent upon Claim 1, further recites
selecting the representative sample comprises: determining a distance between one or more feature vectors associated with the representative sample to the one or more feature vectors used by the artificial intelligence model to make the inference, which is an evaluation or judgement that can be performed in the human mind.
Thus, the claim recites an abstract idea under Step 2A Prong 1.
Under Step 2A Prong 2 and 2B, the claim does not recite any new additional elements which could integrate the abstract idea into a practical application or provide significantly more than the abstract idea itself. Thus, the claim is ineligible.
Claim 6, dependent upon Claim 5, further recites
clustering feature vectors associated with the training dataset, which is an evaluation or judgement that can be performed in the human mind.
Thus, the claim recites an abstract idea under Step 2A Prong 1.
Under Step 2A Prong 2, the claim does not include any additional elements which integrate the abstract idea into a practical application, since the additional elements consist of:
the one or more feature vectors associated with the representative sample are captured during training of the artificial intelligence model, which is merely an insignificant extra-solution activity of data gathering, which by MPEP 2106.05(g) cannot integrate an abstract idea into a practical application.
Thus, the claim is directed towards an abstract idea.
Further, under Step 2B, the additional element does not provide significantly more than the abstract idea itself, because the activity of data gathering (MPEP 2106.05(g)) cannot provide significantly more, as storing and retrieving information in memory is well understood, routine, and conventional (MPEP 2106.05(d)(II)(iv)). Thus, the claim is ineligible.
Claim 7, dependent upon Claim 5, further recites no additional abstract ideas. However:
Under Step 2A Prong 2, the claim does not include any additional elements which integrate the abstract idea into a practical application, since the additional elements consist of:
configuring one or more neural network layers of the artificial intelligence model to capture them when the representative sample as used as input to the artificial intelligence model, as the performance of an abstract idea on a computer is not more than instructions to "apply it" on a computer, which by MPEP 2106.05(f) cannot integrate an abstract idea into a practical application;
the one or more feature vectors associated with the representative sample are captured during training of the artificial intelligence model, which is merely an insignificant extra-solution activity of data gathering, which by MPEP 2106.05(g) cannot integrate an abstract idea into a practical application.
Thus, the claim is directed towards an abstract idea.
Further, the additional elements, alone or in combination, do not provide significantly more than the abstract idea itself, because implementation on a computer (MPEP 2106.05(f)), and the activity of data gathering (MPEP 2106.05(g)) cannot provide significantly more, as storing and retrieving information in memory is well understood, routine, and conventional (MPEP 2106.05(d)(II)(iv)) and the combination of additional elements does not provide an inventive concept. Thus, the claim is ineligible.
Claim 8, dependent upon Claim 1, further recites no additional abstract ideas. However:
Under Step 2A Prong 2, the claim does not include any additional elements which integrate the abstract idea into a practical application, since the additional elements consist of:
the trainable, differentiable neural logic model comprises one or more conjunction neurons, as the performance of an abstract idea on a computer is not more than instructions to "apply it" on a computer, which by MPEP 2106.05(f) cannot integrate an abstract idea into a practical application.
Thus, the claim is directed towards an abstract idea.
Further, under Step 2B, the additional element does not provide significantly more than the abstract idea itself, because implementation on a computer (MPEP 2106.05(f)) cannot provide significantly more. Thus, the claim is ineligible.
Claim 9, dependent upon Claim 8, further recites
construct a conjunction of a subset of the one or more feature vectors, which is an evaluation or judgement that can be performed in the human mind.
Thus, the claim recites an abstract idea under Step 2A Prong 1.
Under Step 2A Prong 2, the claim does not include any additional elements which integrate the abstract idea into a practical application, since the additional elements consist of:
the one or more conjunction neurons are configured to, as the performance of an abstract idea on a computer is not more than instructions to "apply it" on a computer, which by MPEP 2106.05(f) cannot integrate an abstract idea into a practical application.
Thus, the claim is directed towards an abstract idea.
Further, under Step 2B, the additional element does not provide significantly more than the abstract idea itself, because implementation on a computer (MPEP 2106.05(f)) cannot provide significantly more. Thus, the claim is ineligible.
Claim 10, dependent upon Claim 8, further recites no additional abstract ideas. However:
Under Step 2A Prong 2, the claim does not include any additional elements which integrate the abstract idea into a practical application, since the additional elements consist of:
the trainable, differentiable neural logic model further comprises at least one disjunction neuron, as the performance of an abstract idea on a computer is not more than instructions to "apply it" on a computer, which by MPEP 2106.05(f) cannot integrate an abstract idea into a practical application.
Thus, the claim is directed towards an abstract idea.
Further, under Step 2B, the additional element does not provide significantly more than the abstract idea itself, because implementation on a computer (MPEP 2106.05(f)) cannot provide significantly more. Thus, the claim is ineligible.
Claims 11-19 recite An apparatus, thus a machine, one of the four statutory categories of patentable subject matter. However, Claims 11-20 recite comprising: one or more network interfaces; a processor coupled to the one or more network interfaces and configured to execute one or more processes; and a memory configured to store a process that is executable by the processor, the process when executed configured to precisely perform the abstract ideas and additional elements of Claims 1-9, respectively. Therefore, Step 2A Prong 1 analysis remains the same. As for Step 2A Prong 2 and Step 2B: performance on a computer cannot integrate an abstract idea into a practical application (Step 2A Prong 2) nor provide significantly more than the abstract idea itself (Step 2B) (MPEP 2106.05(f)), Claims 11-19 are rejected as subject-matter ineligible for reasons set forth in the rejections of Claims 1-9, respectively.
Claim 20 recites A tangible, non-transitory, computer-readable medium, thus an article of manufacture, one of the four statutory categories of patentable subject matter. However, Claim 20 recites storing program instructions that cause a device to execute a process comprising precisely the abstract ideas and additional elements of Claim 1. Therefore, Step 2A Prong 1 analysis remains the same. As for Step 2A Prong 2 and Step 2B: performance on a computer cannot integrate an abstract idea into a practical application (Step 2A Prong 2) nor provide significantly more than the abstract idea itself (Step 2B) (MPEP 2106.05(f)), and thus Claim 20 is rejected as subject-matter ineligible for reasons set forth in the rejections of Claim 1.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al., “This Looks Like That: Deep Learning for Interpretable Image Recognition” (hereinafter Chen), in view of Payani et al., “Incorporating Relational Background Knowledge into Reinforcement Learning via Differentiable Inductive Logic Programming” (hereinafter Payani), further in view of Riley et al., "Integrating Non-monotonic Logical Reasoning and Inductive Learning With Deep Learning for Explainable Visual Question Answering" (hereinafter Riley).
Regarding Claim 1, Chen teaches:
A computer-implemented method comprising: receiving input data from a user; making an inference regarding the input data using an artificial intelligence model; (Chen [Abstract] The mounting evidence for each of the classes helps us make our final decision. In this work, we introduce a deep network architecture – prototypical part network (ProtoPNet), that reasons in a similar way: the network dissects the image by finding prototypical parts, and combines evidence from the prototypes to make a final classification.
[p.7 4th ¶] Figure 3 shows the reasoning process of our ProtoPNet in reaching a classification decision on a test image of a red-bellied woodpecker at the top of the figure.
PNG
media_image1.png
484
959
media_image1.png
Greyscale
) (Note: the image of the bird classified as a red-bellied woodpecker corresponds to the input data received from the user; reaching a classification decision corresponds to making an inference)
capturing one or more feature vectors from a predetermined feature layer of the artificial intelligence model that were used by the artificial intelligence model to make the inference; (Chen [Abstract] The mounting evidence for each of the classes helps us make our final decision. In this work, we introduce a deep network architecture – prototypical part network (ProtoPNet), that reasons in a similar way: the network dissects the image by finding prototypical parts, and combines evidence from the prototypes to make a final classification.) (Note: the prototypical parts correspond to the feature vectors used by the predetermined feature layers of the artificial intelligence model used to make the inference)
selecting, based at least in part on the one or more feature vectors, a representative sample from a training dataset used to train the artificial intelligence model, (Chen, fig. 3
PNG
media_image1.png
484
959
media_image1.png
Greyscale
)
wherein selecting comprises comparing the one or more feature vectors to stored feature vectors, (Chen, p.7 ¶4, “Figure 3 shows the reasoning process of our ProtoPNet in reaching a classification decision on a test image of a red-bellied woodpecker at the top of the figure. Given this test image x, our model compares its latent features f(x) against the learned prototypes.”)
and selecting the representative sample based at least in part on which of the stored feature vectors the artificial intelligence model considers to be similar to the one or more feature vectors; (Chen
PNG
media_image1.png
484
959
media_image1.png
Greyscale
PNG
media_image2.png
348
783
media_image2.png
Greyscale
[p.5 ¶1] This activation map preserves the spatial relation of the convolutional output, and can be upsampled to the size of the input image to produce a heat map that identifies which part of the input image is most similar to the learned prototype.) (Note: the training image where the prototype comes from corresponds to the representative sample from the training dataset; each label in the output logits corresponds to each representative sample (in this case, the clay colored sparrow is the selected representative sample))
Chen does not teach, but Payani further teaches:
constructing, using a trainable, differentiable neural logic model trained via gradient optimization alongside the artificial intelligence model, one or more Boolean rules in disjunctive form that correspond to the one or more feature vectors, respectively,
(Payani [p.1 right last ¶] In (Payani & Fekri, 2019) a novel ILP solver was introduced which uses Neural-Logical Network (NLN) (Payani & Fekri, 2018) for constructing a differentiable neural-logic ILP solver (dNL-ILP).
[p.2 left last ¶] Each atom is created by applying an n-ary Boolean function called predicate to some constants or variables. A predicate states the relation between some variables or constants in the logic program.
[p.8 right ¶1] We use the same CNN network and similar to the GridWorld experiment, we learn the state representation using predicate color(X,Y,C) (the color of each cell in the grid) as well as isCircle(X,Y) which learn if the shape of an object is circle or not.
[p.3 left ¶1] Likewise, a neural disjunction function fdisj(xn) can be defined using the auxiliary function Fd with the truth table as in Fig. 2b. By cascading a layer of N neural conjunction functions with a layer of N neural disjunction functions, we can construct a differentiable function to be used for representing and learning a Boolean Disjunctive Normal Form (DNF).
[p.10¶3] In the other section, 4 groups of predicates are presented: (i) state representation predicates that their groundings are learned from image, (ii) extensional predicates which are solely defined by background facts, (iii) auxiliary predicates and, (iv) the signature for the target predicate that is used to represent the actions in the policy gradient scheme. To learn the policy, we used discount factor of 0.7, and we use ADAM optimizer with learning rate of 0.002. We set the maximum number of steps for each episode to 20.) (Note: learning the predicates color and isCircle using the CNN network corresponds to constructing Boolean rules that correspond to the feature vectors; the policy gradient scheme using the ADAM optimizer corresponds to the gradient optimization)
Payani and Chen are analogous to the present invention because both are from the same field of endeavor of CNN-based interpretation of feature vectors. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to implement the differentiable neural-logic ILP solver from Payani into Chen’s method of selecting feature identifiers. The motivation would be to “effectively learn relational information from image and present the state of the environment as first order logic predicates” (Payani [abstract]).
Chen, via Chen/Payani, further teaches:
and causing the representative sample, the inference…to be concurrently displayed on a display device associated with the user.
(Chen
PNG
media_image1.png
484
959
media_image1.png
Greyscale
PNG
media_image3.png
432
1068
media_image3.png
Greyscale
) (Note: fig. 4c corresponds to human-interpretable feature identifiers; the training image where the prototype comes from corresponds to the representative sample from the training dataset; the document containing the figure corresponds to the display device)
Chen/Payani does not teach, but Riley further teaches:
wherein the one or more Boolean rules express how the artificial intelligence model combined the one or more feature vectors to make the inference regarding the input data;
(Riley [p.5 right ¶1] 3. A component that answers explanatory questions. If non monotonic logical reasoning is used for classification, it is also used to provide answers to these questions. If a decision tree is instead used for classification, an RNN is trained to map the decision tree’s output, the image features, and the question, to the corresponding answer.)
and the one or more Boolean rules expressing how the artificial intelligence model combined the one or more feature vectors to make the inference to be concurrently displayed on a display device associated with the user
(Riley [p.13 right ¶2] • The classification question posed was: “what is the sign’s message?” The architecture’s answer: “uneven surfaces ahead”. • When asked to explain this answer (“Please explain this answer”), the architecture identified that the CNNs extracted the following features of the sign in the image: (i) it is triangle shaped; (ii) main color is white and other (i.e., border) color is red; (iii) it has no background image; (iv) it has a bumpy-road symbol and no secondary symbol; and (v) it has no cross. • These features were converted to literals and used in ASP based inference based on existing knowledge in the TS domain. ASP-based inference is unable to provide an answer, i.e., unable to classify the sign. • The extracted features were processed using the trained decision tree, which only used the colors in the sign to assign the class label. The main (or border) color is normally insufficient to accurately classify signs. However, recall that the decision tree is trained to classify signs that cannot be classified by reasoning with existing knowledge. • The decision tree output, image feature vector, and input question, were processed by the previously trained RNN to provide the answer type and the particular answer described above.)
Riley and Chen/Payani are analogous to the present invention because both are from the same field of endeavor of using formal model conversions of feature maps for explanations. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to implement the execution trace using the generated formal logic to explain the classification decisions from Riley into Chen/Payani’s method of using predicates from the formalized feature maps. The motivation would be to “The architecture takes advantage of (and perform non-monotonic logical inference with) the existing commonsense domain knowledge to reliably and efficiently address the decision-making problem (classification in the examples above) when possible. In such cases, it is also able to answer explanatory questions about the classification decision and the underlying scene.” (Riley [p.13 right 3rd last ¶]).
Regarding Claim 2, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 1. Chen, via Chen/Payani/Riley, further teaches:
The computer-implemented method as in claim 1, wherein the artificial intelligence model comprises a classifier. (Chen [Abstract] The mounting evidence for each of the classes helps us make our final decision. In this work, we introduce a deep network architecture – prototypical part network (ProtoPNet), that reasons in a similar way: the network dissects the image by finding prototypical parts, and combines evidence from the prototypes to make a final classification.) (Note: the ProtoPNet making a final classification corresponds to comprising a classifier.)
Regarding Claim 3, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 1. Chen, via Chen/Payani/Riley, further teaches:
The computer-implemented method as in claim 1, wherein the input data comprises an image. (Chen [Abstract] The mounting evidence for each of the classes helps us make our final decision. In this work, we introduce a deep network architecture – prototypical part network (ProtoPNet), that reasons in a similar way: the network dissects the image by finding prototypical parts, and combines evidence from the prototypes to make a final classification.)
Regarding Claim 4, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 1. Chen, via Chen/Payani/Riley, further teaches:
The computer-implemented method as in claim 1, wherein the artificial intelligence model comprises a neural network. (Chen [Abstract] The mounting evidence for each of the classes helps us make our final decision. In this work, we introduce a deep network architecture – prototypical part network (ProtoPNet), that reasons in a similar way: the network dissects the image by finding prototypical parts, and combines evidence from the prototypes to make a final classification.
[p.2 2nd ¶] Our work relates to (but contrasts with) those that perform posthoc interpretability analysis for a trained convolutional neural network (CNN).)
Regarding Claim 5, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 1. Chen, via Chen/Payani/Riley, further teaches:
The computer-implemented method as in claim 1, wherein selecting the representative sample comprises: determining a distance between one or more feature vectors associated with the representative sample to the one or more feature vectors used by the artificial intelligence model to make the inference. (Chen [p.3 last ¶] Given a convolutional output z = f(x), the j-th prototype unit gpj in the prototype layer gp computes the squared L 2 distances between the j-th prototype pj and all patches of z that have the same shape as pj , and inverts the distances into similarity scores. The result is an activation map of similarity scores whose value indicates how strong a prototypical part is present in the image.)
Regarding Claim 6, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 5. Chen, via Chen/Payani/Riley, further teaches:
The computer-implemented method as in claim 5, wherein the one or more feature vectors associated with the representative sample are captured during training of the artificial intelligence model in part by clustering feature vectors associated with the training dataset. (Chen [p.5 6th ¶] In the first training stage, we aim to learn a meaningful latent space, where the most important patches for classifying images are clustered (in L 2 -distance) around semantically similar prototypes of the images’ true classes, and the clusters that are centered at prototypes from different classes are well-separated. To achieve this goal, we jointly optimize the convolutional layers’ parameters wconv and the prototypes P = {pj} m j=1 in the prototype layer gp using SGD, while keeping the last layer weight matrix wh fixed.)
Regarding Claim 7, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 5. Chen, via Chen/Payani/Riley, further teaches:
The computer-implemented method as in claim 5, wherein the one or more feature vectors associated with the representative sample are captured during training of the artificial intelligence model in part by configuring one or more neural network layers of the artificial intelligence model to capture them when the representative sample as used as input to the artificial intelligence model. (Chen [p.3 last ¶] The network learns m prototypes P = {pj} m j=1, whose shape is H1 × W1 × D with H1 ≤ H and W1 ≤ W. In our experiments, we used H1 = W1 = 1. Since the depth of each prototype is the same as that of the convolutional output but the height and the width of each prototype is smaller than those of the whole convolutional output, each prototype will be used to represent some prototypical activation pattern in a patch of the convolutional output, which in turn will correspond to some prototypical image patch in the original pixel space. Hence, each prototype pj can be understood as the latent representation of some prototypical part of some bird image in this case study.) (Note: the prototypical part corresponds to the representative sample.)
Regarding Claim 8, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 1. Payani, via Chen/Payani, further teaches:
The computer-implemented method as in claim 1, wherein the trainable, differentiable neural logic model comprises one or more conjunction neurons.
(Payani [p.3 left ¶1] Likewise, a neural disjunction function fdisj(xn) can be defined using the auxiliary function Fd with the truth table as in Fig. 2b. By cascading a layer of N neural conjunction functions with a layer of N neural disjunction functions, we can construct a differentiable function to be used for representing and learning a Boolean Disjunctive Normal Form (DNF).)
Regarding Claim 9, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 8. Payani, via Chen/Payani, further teaches:
The computer-implemented method as in claim 8, wherein the one or more conjunction neurons are configured to construct a conjunction of a subset of the one or more feature vectors. (Payani [p.2 right last ¶] In (Payani &Fekri, 2019), a novel approach was introduced to alleviate the above limitation and to allow for learning arbitrary complex predicate formulas. The main idea behind this approach is to use multiplicative neurons (Payani & Fekri, 2018) that are capable of learning and representing Boolean logic.
[p.8 right ¶1] We use the same CNN network and similar to the GridWorld experiment, we learn the state representation using predicate color(X,Y,C) (the color of each cell in the grid) as well as isCircle(X,Y) which learn if the shape of an object is circle or not.) (Note: learning the predicates color and isCircle using the CNN network corresponds to constructing Boolean rules that correspond to the feature vectors)
Regarding Claim 10, Chen/Payani/Riley respectively teaches and incorporates the claimed limitations and rejections of Claim 8. Payani, via Chen/Payani, further teaches:
The computer-implemented method as in claim 8, wherein the trainable, differentiable neural logic model further comprises at least one disjunction neuron. (Payani [p.3 left ¶1] Likewise, a neural disjunction function fdisj(xn) can be defined using the auxiliary function Fd with the truth table as in Fig. 2b. By cascading a layer of N neural conjunction functions with a layer of N neural disjunction functions, we can construct a differentiable function to be used for representing and learning a Boolean Disjunctive Normal Form (DNF).)
Independent Claim 11 recites An apparatus, comprising: one or more network interfaces; a processor coupled to the one or more network interfaces and configured to execute one or more processes; and a memory configured to store a process that is executable by the processor, the process when executed configured to (Chen [p.3 8th line] In contrast, our ProtoPNet uses a specialized neural network architecture for feature extraction and prototype learning, and can be trained in an end-to-end fashion. [p.9 4th to last line] Supplementary Material and Code: The supplementary material and code are available at https://github.com/cfchen-duke/ProtoPNet.) to perform precisely the methods of Claim 1. Thus, Claim 11 is rejected for reasons set forth in Claim 1. (Note: training ProtoPNet requires a processor and memory; running the code stored on GitHub requires a network interface.)
Claim(s) 12-19, dependent on Claim 11 also recite the apparatus configured to perform precisely the methods of Claims 2-9, respectively, and thus are rejected for reasons set forth in these claims.
Independent Claim 20 recites A tangible, non-transitory, computer-readable medium storing program instructions that cause a device to execute a process comprising (Chen [p.3 8th line] In contrast, our ProtoPNet uses a specialized neural network architecture for feature extraction and prototype learning, and can be trained in an end-to-end fashion.) to perform precisely the methods of Claim 1. Thus, Claim 20 is rejected for reasons set forth in Claim 1. (Note: training ProtoPNet requires a processor and memory)
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 JOSEP HAN whose telephone number is (703)756-1346. The examiner can normally be reached Mon-Fri 9am-5pm.
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 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.
/J.H./Examiner, Art Unit 2122
/KAKALI CHAKI/Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights