DETAILED ACTION
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 .
Claims 1-11 are pending in this application. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
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.
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 1-11 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim(s) does/do not fall within at least one of the four categories of patent eligible subject matter because the claimed invention is directed to a judicial exception (i.e., abstract idea – an idea to itself/mathematical concept) without significantly more.
(1) Are the claims directed to a process, machine, manufacture or composition of matter;
(2A) Prong One: Are the claims directed to a judicially recognized exception, i.e., a law of nature, a natural phenomenon, or an abstract idea;
Prong Two: If the claims are directed to a judicial exception under Prong One, then is the judicial exception integrated into a practical application;
(2B) If the claims are directed to a judicial exception and do not integrate the judicial exception, do the claims provide an inventive concept.
With regard to (1), the instant claims recite a method and a non-transitory computer readable medium, therefore the answer is "YES".
With regard to (2A), Prong One: YES. When viewed under the broadest most reasonable interpretation, the instant claims are directed to a Judicial Exception – an abstract idea belonging to the group of mathematical concept and/or an idea of itself. The step of “obtaining a set of instances” is considered to be extra-solution activity directed to data gathering. The step of “processing the set of instances via an adaptive weighted uncertainty sampling (AWUS) methodology” is considered to be judicially recited mathematical concept/algorithm. The step of “assign weightings to unlabeled instances within the set of instances to generate weighted unlabeled instances” can be interpreted as either an overall mathematical concept. The step of “determining which of the weighted unlabeled instances should be processed further based on the assigned weightings” is also considered to be the application of the mathematical concept/algorithm. There is nothing in the claim that requires more than an operation that a human, armed with the appropriate apparatus executing a mathematical algorithm (in this case “values”) can perform.
With regard to (2A), Prong Two: NO. The instant claims do not apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception of “processing the set of instances via an adaptive weighted uncertainty sampling (AWUS) methodology” and therefore does not integrate the judicial exception into a practical application. In particular, Claim 1 and Claim 11 are independent claims the claim includes additional elements as follows and includes using a processing apparatus to perform the following:
a. 1. A method of active learning comprising; / 11. A non-transient computer readable medium containing program instructions for causing a computer to perform the method of:
b. 1. obtaining a set of instances; / 11. obtaining a set of instances;
c. 1. processing the set of instances via an adaptive weighted uncertainty sampling (AWUS) methodology to assign weightings to unlabeled instances within the set of instances to generate weighted unlabeled instances; / 11. processing the set of instances via an adaptive weighted uncertainty sampling (AWUS) methodology to assign weightings to unlabeled instances within the set of instances to generate weighted unlabeled instances;
d. 1. and determining which of the weighted unlabeled instances should be processed further based on the assigned weightings. / 11. and determining which of the weighted unlabeled instances should be processed further based on the assigned weightings
Step (a) is the preamble for both of the independent claims.
Step (b) for “obtaining a set of instances” is considered to be extra-solution activity directed to data gathering, and is recited at a high level of generality such that said “instances” can constitute any data, and can be used in the operation of the recited judicial exception (the data gathering step of “obtaining”). Supplying “instances” does not provide for “integration” of the abstract idea into a practical application, as said “instances” do not change the way in which said apparatus operates.
Step (c) has two parts, the first part is “processing the set of instances via an adaptive weighted uncertainty sampling (AWUS) methodology”, which is considered to be judicially recited mathematical concept/algorithm. This part is done either mathematically or via a learning algorithm, as there are no specifics about how the “processing” is conducted, and only requires the application of the AWUS methodology. The step of “assign weightings to unlabeled instances within the set of instances to generate weighted unlabeled instances” can be interpreted as either an overall mathematical concept. The manner in which weightings are determined and assigned is not qualified further, and it appears to only constitute the steps that would be present in the AWUS methodology, which is considered a mathematical calculation step.
Step (d) The step of “determining which of the weighted unlabeled instances should be processed further based on the assigned weightings” is also considered to be the application of the mathematical concept/algorithm. There is no further qualification within the claimed features in how this “determination” is performed. Further, it is unclear if this step can also be done mentally, and does not make the claim as a whole patent eligible because the claim as a whole of judicial exception does not integrate into a practical application.
In fact, there are no limits on the apparatus, which is recited at a high level of generality and thus said apparatus does nothing more than perform generic computing functions of “processing the set of instances via an adaptive weighted uncertainty sampling (AWUS) methodology” in the claim. With regard to (2B), the pending claims do not show what is more than a routine in the art presented in the claims, i.e., the additional elements are nothing more than routine and well-known steps. There is no improvement to technology here, and it has not been shown that the mathematical process allows the “technology” (whether it is computer technology or any other technology) to do something that it previously was not able to do.
Dependent claims 2-10 are rejected for the same reasons; claims are directed to a judicial exception and do not integrate the judicial exception.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1-5 and 8-11 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Zhang et al. (US PGPub US 20190370955, hereby referred to as “Zhang”).
Consider Claims 1 and 11.
Zhang teaches:
1. A method of active learning comprising; / 11. A non-transient computer readable medium containing program instructions for causing a computer to perform the method of: (Zhang: abstract; Methods and systems for performing active learning for defect classifiers are provided. One system includes one or more computer subsystems configured for performing active learning for training a defect classifier. The active learning includes applying an acquisition function to data points for the specimen. The acquisition function selects one or more of the data points based on uncertainty estimations associated with the data points. The active learning also includes acquiring labels for the selected one or more data points and generating a set of labeled data that includes the selected one or more data points and the acquired labels. The computer subsystem(s) are also configured for training the defect classifier using the set of labeled data. The defect classifier is configured for classifying defects detected on the specimen using the images generated by the imaging subsystem.[0041]-[0043], Figure 1)
1. obtaining a set of instances; / 11. obtaining a set of instances; (Zhang: Figure 2, [0059] As shown in FIG. 2, in one embodiment, data points 200 may be input to acquisition function 202. At the beginning of the process (in the first step of the first iteration), data points 200 may be considered the initial data to which results of one or more steps described herein may be added to thereby generate a dataset that includes at least some labeled data points. In one embodiment, the data points for the specimen consist of unlabeled data points. For example, the initial data may include no ground truth data (where “ground truth data” is generally defined as data that has been generated by a “ground truth” method such as a defect review process that is performed using a defect review tool capable of resolving images of the defects and/or user-provided information such as manual defect classifications). In another embodiment, the data points for the specimen include a combination of fewer than ten ground truth data points for any one defect type and unlabeled data. [0060] An objective of the adaptive discovery loop shown in FIG. 2 is to dynamically decide the candidate(s) to be verified based on the present “known” data, which may be performed as described further herein. For example, as shown in FIG. 2, acquisition function 202 is used to decide the batch of unknown candidates that are sent to labeling 204 next, which may be performed by a verifier to be verified. Labeling may include one or more of the ground truth methods described herein.)
1. processing the set of instances via an adaptive weighted uncertainty sampling (AWUS) methodology to assign weightings to unlabeled instances within the set of instances to generate weighted unlabeled instances; / 11. processing the set of instances via an adaptive weighted uncertainty sampling (AWUS) methodology to assign weightings to unlabeled instances within the set of instances to generate weighted unlabeled instances; (Zhang: [0061] In one embodiment, the acquisition function is configured to select the one or more of the data points that have the highest uncertainty of being any known defect type. For example, the acquisition function may be configured to select data points having the highest uncertainties of being any known defect type so that those data points can be sent for verification (labeling) as described further herein. By selecting the highest uncertainty data points for verification or labeling and then using those labeled data points for defect classifier training as described herein, the resulting trained defect classifier will have better performance for the function it is configured for (i.e., by selecting and then verifying the highest uncertainty data points and then training the defect classifier with those labeled data points, the uncertainty in the data points that may be input to the defect classifier can be essentially “trained out” of the defect classifier thereby rendering it capable of correctly classifying those previously uncertain data points). Depending on the type of acquisition function used in the embodiments described herein, the acquisition function can be configured to select the data points having the highest uncertainty of any known defect type as described further herein. [0062] In some embodiments, the acquisition function is defined as an adaptive sampling method, some suitable examples of which are described in U.S. Pat. No. 9,098,891 to Kulkarni et al., which is incorporated by reference as if fully set forth herein. The embodiments described herein may be further configured as described in this patent. The embodiments described herein may establish a (machine) learning-based approach to render an original rule based method capable of adaptive sampling on smaller design rules with higher process variations. The net result is better generalizability across multiple design rules and higher sampling efficiency under higher process variations (e.g., noise). The sampling may be adaptive in any other suitable manner (e.g., it may vary depending on the initial data points (or characteristics determined thereof) rather than some predetermined rules, it may vary depending on the labels produced as a result of labeling 204, etc.). [0063])
1. and determining which of the weighted unlabeled instances should be processed further based on the assigned weightings. / 11. and determining which of the weighted unlabeled instances should be processed further based on the assigned weightings. (Zhang: [0064]
The embodiments described herein may also perform adaptive labeling on deep learning/machine learning training datasets. In one example, the embodiments enable on-the-fly image labeling (i.e., marking the defects at pixel level accuracy for a semiconductor image) with the deep learning classification/detection model. This solution enables faster time-to-recipe-creation for new specimens or new design rules, which reduces recipe setup cost and enables relatively fast design of experiments and increases the value of the tool.
[0065]
In one embodiment, the acquisition function is defined as an unsupervised sampling method. For example, some possible methods to define the acquisition function are random (or weighted random) sampling and diversity sampling. Random (or weighted random) sampling may be performed in any suitable manner known in the art. Diversity sampling may include selecting two or more of the data points that are the most diverse (most different) in some manner (e.g., most diverse in a characteristic of the data points, which may include any suitable characteristic of the data points). The unsupervised sampling method may be unsupervised in that the data points that are being sampled are not labeled and/or the sampling is not performed based on any labels that are available for the data points being sampled.)
2. The method of active learning of Claim 1 further comprising, after processing the set of instances: annotating at least one of the weighted unlabeled instances. (Examiner Note: the embodiment of including (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data is analogous in scope to annotating the weighted unlabeled instances; Zhang: [0059] As shown in FIG. 2, in one embodiment, data points 200 may be input to acquisition function 202. At the beginning of the process (in the first step of the first iteration), data points 200 may be considered the initial data to which results of one or more steps described herein may be added to thereby generate a dataset that includes at least some labeled data points. In one embodiment, the data points for the specimen consist of unlabeled data points. For example, the initial data may include no ground truth data (where “ground truth data” is generally defined as data that has been generated by a “ground truth” method such as a defect review process that is performed using a defect review tool capable of resolving images of the defects and/or user-provided information such as manual defect classifications). In another embodiment, the data points for the specimen include a combination of fewer than ten ground truth data points for any one defect type and unlabeled data. For example, the initial data may include more than one defect data points. In one particular example, the data points may include only 1 or 2 labeled examples of any one defect type, possibly with 1 or 2 labeled examples of multiple defect types (e.g., 1 or 2 bridge defects, 1 or 2 particle defects, 1 or 2 3D embedded defects, and so on). In this manner, the workflow can start from either (1) no ground truth data, i.e., a pool of unlabeled data, or (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data. [0060] An objective of the adaptive discovery loop shown in FIG. 2 is to dynamically decide the candidate(s) to be verified based on the present “known” data, which may be performed as described further herein. For example, as shown in FIG. 2, acquisition function 202 is used to decide the batch of unknown candidates that are sent to labeling 204 next, which may be performed by a verifier to be verified. Labeling may include one or more of the ground truth methods described herein. [0061] In one embodiment, the acquisition function is configured to select the one or more of the data points that have the highest uncertainty of being any known defect type. For example, the acquisition function may be configured to select data points having the highest uncertainties of being any known defect type so that those data points can be sent for verification (labeling) as described further herein. By selecting the highest uncertainty data points for verification or labeling and then using those labeled data points for defect classifier training as described herein, the resulting trained defect classifier will have better performance for the function it is configured for (i.e., by selecting and then verifying the highest uncertainty data points and then training the defect classifier with those labeled data points, the uncertainty in the data points that may be input to the defect classifier can be essentially “trained out” of the defect classifier thereby rendering it capable of correctly classifying those previously uncertain data points). Depending on the type of acquisition function used in the embodiments described herein, the acquisition function can be configured to select the data points having the highest uncertainty of any known defect type as described further herein.)
3. The method of active learning of Claim 1 further comprising: processing the determined weighted unlabeled instances. (Examiner Note: the embodiment of including (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data is analogous in scope to annotating the weighted unlabeled instances; Zhang: [0059] As shown in FIG. 2, in one embodiment, data points 200 may be input to acquisition function 202. At the beginning of the process (in the first step of the first iteration), data points 200 may be considered the initial data to which results of one or more steps described herein may be added to thereby generate a dataset that includes at least some labeled data points. In one embodiment, the data points for the specimen consist of unlabeled data points. For example, the initial data may include no ground truth data (where “ground truth data” is generally defined as data that has been generated by a “ground truth” method such as a defect review process that is performed using a defect review tool capable of resolving images of the defects and/or user-provided information such as manual defect classifications). In another embodiment, the data points for the specimen include a combination of fewer than ten ground truth data points for any one defect type and unlabeled data. For example, the initial data may include more than one defect data points. In one particular example, the data points may include only 1 or 2 labeled examples of any one defect type, possibly with 1 or 2 labeled examples of multiple defect types (e.g., 1 or 2 bridge defects, 1 or 2 particle defects, 1 or 2 3D embedded defects, and so on). In this manner, the workflow can start from either (1) no ground truth data, i.e., a pool of unlabeled data, or (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data. [0060] An objective of the adaptive discovery loop shown in FIG. 2 is to dynamically decide the candidate(s) to be verified based on the present “known” data, which may be performed as described further herein. For example, as shown in FIG. 2, acquisition function 202 is used to decide the batch of unknown candidates that are sent to labeling 204 next, which may be performed by a verifier to be verified. Labeling may include one or more of the ground truth methods described herein. [0061] In one embodiment, the acquisition function is configured to select the one or more of the data points that have the highest uncertainty of being any known defect type. For example, the acquisition function may be configured to select data points having the highest uncertainties of being any known defect type so that those data points can be sent for verification (labeling) as described further herein. By selecting the highest uncertainty data points for verification or labeling and then using those labeled data points for defect classifier training as described herein, the resulting trained defect classifier will have better performance for the function it is configured for (i.e., by selecting and then verifying the highest uncertainty data points and then training the defect classifier with those labeled data points, the uncertainty in the data points that may be input to the defect classifier can be essentially “trained out” of the defect classifier thereby rendering it capable of correctly classifying those previously uncertain data points). Depending on the type of acquisition function used in the embodiments described herein, the acquisition function can be configured to select the data points having the highest uncertainty of any known defect type as described further herein.)
4. The method of active learning of Claim 3 further comprising: transmitting information associated with processing the determined weighted unlabeled instances.(Zhang: [0061] In one embodiment, the acquisition function is configured to select the one or more of the data points that have the highest uncertainty of being any known defect type. For example, the acquisition function may be configured to select data points having the highest uncertainties of being any known defect type so that those data points can be sent for verification (labeling) as described further herein. By selecting the highest uncertainty data points for verification or labeling and then using those labeled data points for defect classifier training as described herein, the resulting trained defect classifier will have better performance for the function it is configured for (i.e., by selecting and then verifying the highest uncertainty data points and then training the defect classifier with those labeled data points, the uncertainty in the data points that may be input to the defect classifier can be essentially “trained out” of the defect classifier thereby rendering it capable of correctly classifying those previously uncertain data points). Depending on the type of acquisition function used in the embodiments described herein, the acquisition function can be configured to select the data points having the highest uncertainty of any known defect type as described further herein. [0065] In one embodiment, the acquisition function is defined as an unsupervised sampling method. For example, some possible methods to define the acquisition function are random (or weighted random) sampling and diversity sampling. Random (or weighted random) sampling may be performed in any suitable manner known in the art. Diversity sampling may include selecting two or more of the data points that are the most diverse (most different) in some manner (e.g., most diverse in a characteristic of the data points, which may include any suitable characteristic of the data points). The unsupervised sampling method may be unsupervised in that the data points that are being sampled are not labeled and/or the sampling is not performed based on any labels that are available for the data points being sampled.)
5. The method of active learning of Claim 1 wherein obtaining a set of instances comprises: receiving a set of images generated by a data generating system. (Zhang: [0061] In one embodiment, the acquisition function is configured to select the one or more of the data points that have the highest uncertainty of being any known defect type. For example, the acquisition function may be configured to select data points having the highest uncertainties of being any known defect type so that those data points can be sent for verification (labeling) as described further herein. By selecting the highest uncertainty data points for verification or labeling and then using those labeled data points for defect classifier training as described herein, the resulting trained defect classifier will have better performance for the function it is configured for (i.e., by selecting and then verifying the highest uncertainty data points and then training the defect classifier with those labeled data points, the uncertainty in the data points that may be input to the defect classifier can be essentially “trained out” of the defect classifier thereby rendering it capable of correctly classifying those previously uncertain data points). Depending on the type of acquisition function used in the embodiments described herein, the acquisition function can be configured to select the data points having the highest uncertainty of any known defect type as described further herein. [0062] In some embodiments, the acquisition function is defined as an adaptive sampling method, some suitable examples of which are described in U.S. Pat. No. 9,098,891 to Kulkarni et al., which is incorporated by reference as if fully set forth herein. The embodiments described herein may be further configured as described in this patent. The embodiments described herein may establish a (machine) learning-based approach to render an original rule based method capable of adaptive sampling on smaller design rules with higher process variations. The net result is better generalizability across multiple design rules and higher sampling efficiency under higher process variations (e.g., noise). The sampling may be adaptive in any other suitable manner (e.g., it may vary depending on the initial data points (or characteristics determined thereof) rather than some predetermined rules, it may vary depending on the labels produced as a result of labeling 204, etc.). [0063])
8. The method of active learning of Claim 6 wherein processing the set of unlabeled instances via an AWUS methodology further comprises: calculating a probability mass function (pmf) value for each of the set of unlabeled instances. (Zhang: [0067] In one embodiment, the acquisition function is defined as a sampling method based on Maximum Entropy. For example, one possible method to define the acquisition function is uncertainty sampling such as Maximum Entropy. In this manner, the acquisition function may be implemented via entropy. Maximum Entropy may include evaluating a number of different probability distributions for any data set such as the set of data points described herein to find the probability distribution that has the maximum entropy (the largest uncertainty). The probability distributions may include any suitable probability distributions, and Maximum Entropy may be performed in any suitable manner known in the art. Sampling based on Maximum Entropy may include sampling the data points based on the probability distribution having the largest entropy. [0076]-[0077], [0080] In another such embodiment, the one or more probability distributions include a supervised or semi-supervised estimation of model posterior and its derived uncertainty distribution. For example, the one or more probability distributions may include a supervised or semi-supervised estimation of model posterior p(w|D) and its derived uncertainty distribution, where w are the model parameters, D is the labeled dataset for supervised methods, and D is the labeled and unlabeled dataset for the semi-supervised case)
9. The method of active learning of Claim 1 further comprising training a machine learning model on the processed set of unlabeled instances. (Examiner Note: the embodiment of including (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data is analogous in scope to annotating the weighted unlabeled instances; Zhang: [0059] As shown in FIG. 2, in one embodiment, data points 200 may be input to acquisition function 202. At the beginning of the process (in the first step of the first iteration), data points 200 may be considered the initial data to which results of one or more steps described herein may be added to thereby generate a dataset that includes at least some labeled data points. In one embodiment, the data points for the specimen consist of unlabeled data points. For example, the initial data may include no ground truth data (where “ground truth data” is generally defined as data that has been generated by a “ground truth” method such as a defect review process that is performed using a defect review tool capable of resolving images of the defects and/or user-provided information such as manual defect classifications). In another embodiment, the data points for the specimen include a combination of fewer than ten ground truth data points for any one defect type and unlabeled data. For example, the initial data may include more than one defect data points. In one particular example, the data points may include only 1 or 2 labeled examples of any one defect type, possibly with 1 or 2 labeled examples of multiple defect types (e.g., 1 or 2 bridge defects, 1 or 2 particle defects, 1 or 2 3D embedded defects, and so on). In this manner, the workflow can start from either (1) no ground truth data, i.e., a pool of unlabeled data, or (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data. [0060] An objective of the adaptive discovery loop shown in FIG. 2 is to dynamically decide the candidate(s) to be verified based on the present “known” data, which may be performed as described further herein. For example, as shown in FIG. 2, acquisition function 202 is used to decide the batch of unknown candidates that are sent to labeling 204 next, which may be performed by a verifier to be verified. Labeling may include one or more of the ground truth methods described herein. [0061] In one embodiment, the acquisition function is configured to select the one or more of the data points that have the highest uncertainty of being any known defect type. For example, the acquisition function may be configured to select data points having the highest uncertainties of being any known defect type so that those data points can be sent for verification (labeling) as described further herein. By selecting the highest uncertainty data points for verification or labeling and then using those labeled data points for defect classifier training as described herein, the resulting trained defect classifier will have better performance for the function it is configured for (i.e., by selecting and then verifying the highest uncertainty data points and then training the defect classifier with those labeled data points, the uncertainty in the data points that may be input to the defect classifier can be essentially “trained out” of the defect classifier thereby rendering it capable of correctly classifying those previously uncertain data points). Depending on the type of acquisition function used in the embodiments described herein, the acquisition function can be configured to select the data points having the highest uncertainty of any known defect type as described further herein.)
10. The method of active learning of Claim 9 further comprising: obtaining a further set of unlabeled instances based on the training of the machine learning model on the weighted unlabeled instances. (Examiner Note: the embodiment of including (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data is analogous in scope to annotating the weighted unlabeled instances; Zhang: [0059] As shown in FIG. 2, in one embodiment, data points 200 may be input to acquisition function 202. At the beginning of the process (in the first step of the first iteration), data points 200 may be considered the initial data to which results of one or more steps described herein may be added to thereby generate a dataset that includes at least some labeled data points. In one embodiment, the data points for the specimen consist of unlabeled data points. For example, the initial data may include no ground truth data (where “ground truth data” is generally defined as data that has been generated by a “ground truth” method such as a defect review process that is performed using a defect review tool capable of resolving images of the defects and/or user-provided information such as manual defect classifications). In another embodiment, the data points for the specimen include a combination of fewer than ten ground truth data points for any one defect type and unlabeled data. For example, the initial data may include more than one defect data points. In one particular example, the data points may include only 1 or 2 labeled examples of any one defect type, possibly with 1 or 2 labeled examples of multiple defect types (e.g., 1 or 2 bridge defects, 1 or 2 particle defects, 1 or 2 3D embedded defects, and so on). In this manner, the workflow can start from either (1) no ground truth data, i.e., a pool of unlabeled data, or (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data. [0060] An objective of the adaptive discovery loop shown in FIG. 2 is to dynamically decide the candidate(s) to be verified based on the present “known” data, which may be performed as described further herein. For example, as shown in FIG. 2, acquisition function 202 is used to decide the batch of unknown candidates that are sent to labeling 204 next, which may be performed by a verifier to be verified. Labeling may include one or more of the ground truth methods described herein. [0061] In one embodiment, the acquisition function is configured to select the one or more of the data points that have the highest uncertainty of being any known defect type. For example, the acquisition function may be configured to select data points having the highest uncertainties of being any known defect type so that those data points can be sent for verification (labeling) as described further herein. By selecting the highest uncertainty data points for verification or labeling and then using those labeled data points for defect classifier training as described herein, the resulting trained defect classifier will have better performance for the function it is configured for (i.e., by selecting and then verifying the highest uncertainty data points and then training the defect classifier with those labeled data points, the uncertainty in the data points that may be input to the defect classifier can be essentially “trained out” of the defect classifier thereby rendering it capable of correctly classifying those previously uncertain data points). Depending on the type of acquisition function used in the embodiments described herein, the acquisition function can be configured to select the data points having the highest uncertainty of any known defect type as described further herein.)
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 may not be obtained though the invention is not identically disclosed or described as set forth in section 102 of this title, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter pertains. Patentability shall not be negatived by the manner in which the invention was made.
Claims 6-7 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (US PGPub US 20190370955, hereby referred to as “Zhang”), in view of Houlsby et al. (Bayesian Active Learning for Classification and Preference Learning,” Houlsby et al., arXiv:1112.5745, 201, incorporated by reference), hereby referred to as “Houlsby”.
Consider Claim 6.
Zhang teaches:
6. The method of active learning of Claim 1 wherein processing the set of instances via an AWUS methodology comprises: selecting a set of unlabeled instances from the set of instances. (Examiner Note: the embodiment of including (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data is analogous in scope to annotating the weighted unlabeled instances; Zhang: [0059] As shown in FIG. 2, in one embodiment, data points 200 may be input to acquisition function 202. At the beginning of the process (in the first step of the first iteration), data points 200 may be considered the initial data to which results of one or more steps described herein may be added to thereby generate a dataset that includes at least some labeled data points. In one embodiment, the data points for the specimen consist of unlabeled data points. For example, the initial data may include no ground truth data (where “ground truth data” is generally defined as data that has been generated by a “ground truth” method such as a defect review process that is performed using a defect review tool capable of resolving images of the defects and/or user-provided information such as manual defect classifications). In another embodiment, the data points for the specimen include a combination of fewer than ten ground truth data points for any one defect type and unlabeled data. For example, the initial data may include more than one defect data points. In one particular example, the data points may include only 1 or 2 labeled examples of any one defect type, possibly with 1 or 2 labeled examples of multiple defect types (e.g., 1 or 2 bridge defects, 1 or 2 particle defects, 1 or 2 3D embedded defects, and so on). In this manner, the workflow can start from either (1) no ground truth data, i.e., a pool of unlabeled data, or (2) a few (>1) ground truth data optionally in combination with a pool of unlabeled data. [0060] An objective of the adaptive discovery loop shown in FIG. 2 is to dynamically decide the candidate(s) to be verified based on the present “known” data, which may be performed as described further herein. For example, as shown in FIG. 2, acquisition function 202 is used to decide the batch of unknown candidates that are sent to labeling 204 next, which may be performed by a verifier to be verified. Labeling may include one or more of the ground truth methods described herein. [0061], [0068] In another embodiment, the acquisition function is defined as a sampling method based on Bayesian Active Learning. For example, the acquisition function may be defined as a Bayesian method. One possible method for defining the acquisition function is Bayesian Active Learning by Disagreement (BALD). Some examples of BALD that may be used to define the acquisition function used in the embodiments described herein are described in “Bayesian Active Learning for Classification and Preference Learning,” Houlsby et al., arXiv:1112.5745, 2011, which is incorporated by reference as if fully set forth herein. In this manner, the acquisition function may be implemented via BALD.)
Zhang does not explicitly state: calculating an exponential value for each of the set of unlabeled instances.
Houlsby teaches:
6. The method of active learning of Claim 1 wherein processing the set of instances via an AWUS methodology comprises: selecting a set of unlabeled instances from the set of instances; and calculating an exponential value for each of the set of unlabeled instances. (Houlsby: section 3 Gaussian Processes for Classification and Preference Learning. In this section we derive the BALD algorithm for Gaussian Process classification (GPC). GPs are a powerful and popular non-parametric tool for regression and classification. GPC appears to be an especially challenging problem for information-theoretic active learning because the parameter space is infinite, however, by using (2) we are able to calculate fully the relevant information quantities without having to work out entropies of infinite dimensional object. Page 5
PNG
media_image1.png
336
584
media_image1.png
Greyscale
PNG
media_image2.png
366
596
media_image2.png
Greyscale
page 9 section 5 Experiments Quantifying Approximation Losses: To obtain (5) we made two approximations: we perform approximate inference (1≈), and we approximated the
binary entropy of the Gaussian CDF by a squared exponential ( 2≈). Both of these can be substituted with Monte Carlo sampling, enabling us to compute an asymptotically unbiased estimate of the expected information gain.)
It would have been obvious before the effective filing date of the claimed invention was made to enhance Zhang’s overall active machine learning algorithm with the improvements of Houlsby for “defining the acquisition function is Bayesian Active Learning by Disagreement (BALD)”, as it was incorporated by reference and would serve as the paradigm upon which the overall learned classifier is built. Zhang teaches an overall active learning classifier that includes acquiring labels for data points and generation labels, and clearly suggest that “the acquisition function may be implemented via BALD” in paragraph [0068]. The determination of obviousness is predicated upon the following findings: One skilled in the art would have been motivated to modify Zhang in order to further understand and optimize the algorithm upon which the active learning model is built. Furthermore, the prior art collectively includes each element claimed (though not all in the same reference), and one of ordinary skill in the art could have combined the elements in the manner explained above using known engineering design, interface and programming techniques, without changing a “fundamental” operating principle of Zhang, while the teaching of Houlsby continues to perform the same function as originally taught prior to being combined, in order to produce the repeatable and predictable result of optimizing the overall architecture for active learning algorithms. It is for at least the aforementioned reasons that the examiner has reached a conclusion of obviousness with respect to the claim in question.
Consider Claim 7.
The combination of Zhang and Houlsby teaches:
7. The method of active learning of Claim 6 wherein calculating an exponential value for each of the set of unlabeled instances comprising: calculating the exponential value based on a similarity metric. Zhang: [0059]-[0061], [0068] In another embodiment, the acquisition function is defined as a sampling method based on Bayesian Active Learning. For example, the acquisition function may be defined as a Bayesian method. One possible method for defining the acquisition function is Bayesian Active Learning by Disagreement (BALD). Some examples of BALD that may be used to define the acquisition function used in the embodiments described herein are described in “Bayesian Active Learning for Classification and Preference Learning,” Houlsby et al., arXiv:1112.5745, 2011, which is incorporated by reference as if fully set forth herein. In this manner, the acquisition function may be implemented via BALD. Houlsby: section 3 Gaussian Processes for Classification and Preference Learning. In this section we derive the BALD algorithm for Gaussian Process classification (GPC). GPs are a powerful and popular non-parametric tool for regression and classification. GPC appears to be an especially challenging problem for information-theoretic active learning because the parameter space is infinite, however, by using (2) we are able to calculate fully the relevant information quantities without having to work out entropies of infinite dimensional object. Page 5
PNG
media_image1.png
336
584
media_image1.png
Greyscale
PNG
media_image2.png
366
596
media_image2.png
Greyscale
page 9 section 5 Experiments Quantifying Approximation Losses: To obtain (5) we made two approximations: we perform approximate inference (1≈), and we approximated the
binary entropy of the Gaussian CDF by a squared exponential ( 2≈). Both of these can be substituted with Monte Carlo sampling, enabling us to compute an asymptotically unbiased estimate of the expected information gain.)
Conclusion
The prior art made of record in form PTO-892 and not relied upon is considered pertinent to applicant's disclosure.
PNG
media_image3.png
162
904
media_image3.png
Greyscale
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TAHMINA ANSARI whose telephone number is 571-270-3379. The examiner can normally be reached on IFP Flex - Monday through Friday 9 to 5.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, O’NEAL MISTRY can be reached on 313-446-4912. The fax phone numbers for the organization where this application or proceeding is assigned are 571-273-8300 for regular communications and 571-273-8300 for After Final communications. TC 2600’s customer service number is 571-272-2600.
Any inquiry of a general nature or relating to the status of this application or proceeding should be directed to the receptionist whose telephone number is 571-272-2600.
2674
/Tahmina Ansari/
/TAHMINA N ANSARI/Primary Examiner, Art Unit 2674