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 .
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
1. Claims 1-4, 7, 9-12, 16 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Ailen et al US 2026/0004135 A1 (“Ailem”) in view of Ribeiro et al “Beyond Accuracy: Behavioral Testing of NLP Models with CheckList” (“Ribeiro” - IDS)
Per claim 1, Ailem discloses a method comprising:
accessing an input dataset comprising text data (The data analysis pipeline is described below with reference to an unstructured textual dataset (i.e., email dataset) …, para. [0030]-[0031]);
transforming the text data into a set of embedding vectors using a first large language model (LLM) of a pre-processing module (the data analysis pipeline accesses a dataset with data items and generates data item embeddings (“embeddings”). A data item embedding refers to a compact numerical representation of a data item, typically in the form of a vector of real numbers… text data embeddings computations involve transforming textual information into numerical representations, typically using techniques like Word2Vec (Word to Vector), GloVe (GloVe (Global Vectors for Word Representation) or BERT (Bidirectional Encoder Representations from Transformers). …, para. [0031]);
performing dimensional reduction on the set of embedding vectors using the pre- processing module to generate a set of lower-dimensional features, wherein each lower-dimensional feature represents text data of a data element of the input dataset (Dimensionality reduction with UMAP is performed on the embeddings. Dimensionality reduction using UMAP involves transforming high-dimensional data into a lower-dimensional space …, para. [0032]);
generating a first set of clusters from the set of lower-dimensional features using a clustering module, wherein each cluster of the first set of clusters is associated with a topic from a first set of topics (Reducing the dimensionality of the data enables applying unsupervised clustering algorithms to identify dense regions and assign data points to clusters in the reduced space, which helps in discovering meaningful patterns and structures in high-dimensional data efficiently. … para. [0032]-[0035]);
extracting a set of representative data elements from each cluster in the first set of clusters using an extraction module, wherein the set of representative data elements are a subset of the input dataset (Additionally, each cluster will be associated with a concise list of data items (i.e., extractive summary) that best represent it. For example, each email cluster may have a list of emails that are representative of the emails in the cluster. An LLM can be used to generate abstractive summaries and data feature assessments for each cluster. For example, abstractive summaries and risk assessments can be generated for each email cluster …, para. [0037]);
generating a set of cases by prompting a second LLM of a generator module based on the set of representative data elements (After clustering and topic modeling, outputs (i.e., clusters or instructive clusters) can be processed with LLMs. In particular, the clusters be processed to generate annotated clusters. For example, clusters can be processed using LLMs (e.g., LLM 140) to generate annotated clusters with categorization (e.g., category annotation) and abstractive summarization …, para. [0048]; para. [0063]);
generating a second set of clusters from the set of cases using a clustering module, wherein each cluster of the second set of clusters is associated with a topic from a second set of topics (para. [0048]; the downstream prompting 110B may apply the one or more of a categorization 110B-1 (e.g., content categorization) and summarization 110B-2 (e.g., subject summarization) …, para. [0062]-[0063]; In this way, the volume of the dataset 102B can be reduced to tangible correlated topics, and the summaries for each of these topics can provide answers and relevant information for the question of what was potentially taken from the dataset 102B. …, para. [0064]); and
testing a downstream machine learning model using one or more cases selected from a particular topic of the second set of topics (In this way, the combination of clustering, the application of the content categorization (i.e., categorization 110B-1) and subject summarization (i.e., summarization 110B-2), and AI agents of the reasoned knowledge graph-building engine 112B can essentially bucket topics of interest or grouped topics of interest for data having particularly relevant semantics and significance associated with a target or goal. For example, the reasoned knowledge graph-building engine 112B may generate vulnerability graphs or other types of graphs from the buckets of interest that point to related topics of interest, such that the generated graph(s) can focus on the clusters that relate to those topics for correlating specific topics for the downstream analysis …, para. [0065])
Ailem does not explicitly disclose generating a set of minimal functionality test (MFT) cases or the use of MFT cases
However, these features are taught by Ribeiro (As a test of the model’s Negation capability, we use a Minimum Functionality test (MFT), i.e. simple test cases designed to target a specific behavior (Figure 1A). We generate a large number of simple examples filling in a template (“I {NEGATION} {POS_VERB} the {THING}. “) with pre-built lexicons, and compute the model’s failure rate on such examples …, sec. 1)
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Ribeiro with the method of Ailem in arriving at the missing features of Ailem, because such combination would have resulted in determining the model’s failure rate (Ribeiro, sec. 1).
Per claim 2, Ailem in view of Ribeiro discloses the method of claim 1,
Ailem discloses wherein the pre-processing module comprises a uniform manifold approximation and projection (UMAP) tool for generating the set of lower-dimensional features (para. [0032]).
Per claim 3, Ailem in view of Ribeiro discloses the method of claim 1,
Ailem discloses wherein the clustering module comprises a hierarchical density-based spatial clustering of applications with noise based tool to generate the first set of clusters (para. [0033]).
Per claim 4, Ailem in view of Ribeiro discloses the method of claim 1,
Ailem discloses wherein the extraction module comprises a bidirectional encoder representation of transformers topic based tool and extracting the set of representative data elements from each cluster comprises analyzing a set of frequency values associated with text data appearing in each cluster (para. [0035]).
Per claim 7, Ailem in view of Ribeiro discloses the method of claim 1,
Ailem discloses wherein the MFT clustering module comprises a K-means clustering based tool for generating the second set of clusters (para. [0034]).
Per claim 9, Ailem in view of Ribeiro discloses the method of claim 1,
Ailem discloses wherein the second set of topics comprises a subset of the first set of topics (para. [0063]; In this way, the volume of the dataset 102B can be reduced to tangible correlated topics, and the summaries for each of these topics can provide answers and relevant information for the question of what was potentially taken from the dataset 102B.…, para. [0064]).
Per claim 10, Ailem discloses a system comprising:
one or more processors (para. [0124]);
a memory coupled to the one or more processors, the memory including instructions that, when executed by the one or more processors, cause the one or more processors to: access an input dataset comprising text data (The data analysis pipeline is described below with reference to an unstructured textual dataset (i.e., email dataset) …, para. [0030]-[0031]; para. [0124]);
transform the text data into a set of embedding vectors using a first large language model (LLM) of a pre-processing module (the data analysis pipeline accesses a dataset with data items and generates data item embeddings (“embeddings”). A data item embedding refers to a compact numerical representation of a data item, typically in the form of a vector of real numbers… text data embeddings computations involve transforming textual information into numerical representations, typically using techniques like Word2Vec (Word to Vector), GloVe (GloVe (Global Vectors for Word Representation) or BERT (Bidirectional Encoder Representations from Transformers). …, para. [0031]);
perform dimensional reduction on the set of embedding vectors using the pre- processing module to generate a set of lower-dimensional features, wherein each lower- dimensional feature represents text data of a data element of the input dataset (Dimensionality reduction with UMAP is performed on the embeddings. Dimensionality reduction using UMAP involves transforming high-dimensional data into a lower-dimensional space …, para. [0032]);
generate a first set of clusters from the set of lower-dimensional features using a clustering module, wherein each cluster of the first set of clusters is associated with a topic from a first set of topics (Reducing the dimensionality of the data enables applying unsupervised clustering algorithms to identify dense regions and assign data points to clusters in the reduced space, which helps in discovering meaningful patterns and structures in high-dimensional data efficiently. … para. [0032]-[0035]);
extract a set of representative data elements from each cluster in the first set of clusters using an extraction module, wherein the set of representative data elements are a subset of the input dataset (Additionally, each cluster will be associated with a concise list of data items (i.e., extractive summary) that best represent it. For example, each email cluster may have a list of emails that are representative of the emails in the cluster. An LLM can be used to generate abstractive summaries and data feature assessments for each cluster. For example, abstractive summaries and risk assessments can be generated for each email cluster …, para. [0037]);
generate a set of cases by prompting a second LLM of a generator module based on the set of representative data elements (After clustering and topic modeling, outputs (i.e., clusters or instructive clusters) can be processed with LLMs. In particular, the clusters be processed to generate annotated clusters. For example, clusters can be processed using LLMs (e.g., LLM 140) to generate annotated clusters with categorization (e.g., category annotation) and abstractive summarization …, para. [0048]; para. [0063]);
generate a second set of clusters from the set of cases using a clustering module, wherein each cluster of the second set of clusters is associated with a topic from a second set of topics (para. [0048]; the downstream prompting 110B may apply the one or more of a categorization 110B-1 (e.g., content categorization) and summarization 110B-2 (e.g., subject summarization) …, para. [0062]-[0063]; In this way, the volume of the dataset 102B can be reduced to tangible correlated topics, and the summaries for each of these topics can provide answers and relevant information for the question of what was potentially taken from the dataset 102B. …, para. [0064]); and
test a downstream machine learning model using one or more MFT cases selected from a particular topic of the second set of topics (In this way, the combination of clustering, the application of the content categorization (i.e., categorization 110B-1) and subject summarization (i.e., summarization 110B-2), and AI agents of the reasoned knowledge graph-building engine 112B can essentially bucket topics of interest or grouped topics of interest for data having particularly relevant semantics and significance associated with a target or goal. For example, the reasoned knowledge graph-building engine 112B may generate vulnerability graphs or other types of graphs from the buckets of interest that point to related topics of interest, such that the generated graph(s) can focus on the clusters that relate to those topics for correlating specific topics for the downstream analysis …, para. [0065])
Ailem does not explicitly disclose generating a set of minimal functionality test (MFT) cases or the use of MFT cases
However, these features are taught by Ribeiro (As a test of the model’s Negation capability, we use a Minimum Functionality test (MFT), i.e. simple test cases designed to target a specific behavior (Figure 1A). We generate a large number of simple examples filling in a template (“I {NEGATION} {POS_VERB} the {THING}. “) with pre-built lexicons, and compute the model’s failure rate on such examples …, sec. 1)
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Ribeiro with the system of Ailem in arriving at the missing features of Ailem, because such combination would have resulted in determining the model’s failure rate (Ribeiro, sec. 1).
Per claim 11, Ailem in view of Ribeiro discloses the system of claim 10,
Ailem discloses wherein the pre-processing module comprises a uniform manifold approximation and projection (UMAP) tool for generating the set of lower- dimensional features, and wherein the clustering module comprises a hierarchical density- based spatial clustering of applications with noise based tool to generate the first set of clusters (para. [0032]-[0033]).
Per claim 12, Ailem in view of Ribeiro discloses the system of claim 10,
Ailem discloses wherein the extraction module comprises a bidirectional encoder representation of transformers topic based tool and extracting the set of representative data elements from each cluster comprises analyzing a set of frequency values associated with text data appearing in each cluster (para. [0035]).
Per claim 16, Ailem discloses a non-transitory computer-readable medium embodying program code that is executable by one or more processors to cause the one or more processors to:
access an input dataset comprising text data (The data analysis pipeline is described below with reference to an unstructured textual dataset (i.e., email dataset) …, para. [0030]-[0031]);
transform the text data into a set of embedding vectors using a first large language model (LLM) of a pre-processing module (the data analysis pipeline accesses a dataset with data items and generates data item embeddings (“embeddings”). A data item embedding refers to a compact numerical representation of a data item, typically in the form of a vector of real numbers… text data embeddings computations involve transforming textual information into numerical representations, typically using techniques like Word2Vec (Word to Vector), GloVe (GloVe (Global Vectors for Word Representation) or BERT (Bidirectional Encoder Representations from Transformers). …, para. [0031]);
perform dimensional reduction on the set of embedding vectors using the pre- processing module to generate a set of lower-dimensional features, wherein each lower- dimensional feature represents text data of a data element of the input dataset (Dimensionality reduction with UMAP is performed on the embeddings. Dimensionality reduction using UMAP involves transforming high-dimensional data into a lower-dimensional space …, para. [0032]);
generate a first set of clusters from the set of lower-dimensional features using a clustering module, wherein each cluster of the first set of clusters is associated with a topic from a first set of topics (Reducing the dimensionality of the data enables applying unsupervised clustering algorithms to identify dense regions and assign data points to clusters in the reduced space, which helps in discovering meaningful patterns and structures in high-dimensional data efficiently. … para. [0032]-[0035]);
extract a set of representative data elements from each cluster in the first set of clusters using an extraction module, wherein the set of representative data elements are a subset of the input dataset (Additionally, each cluster will be associated with a concise list of data items (i.e., extractive summary) that best represent it. For example, each email cluster may have a list of emails that are representative of the emails in the cluster. An LLM can be used to generate abstractive summaries and data feature assessments for each cluster. For example, abstractive summaries and risk assessments can be generated for each email cluster …, para. [0037]);
generate a set of cases by prompting a second LLM of a generator module based on the set of representative data elements (After clustering and topic modeling, outputs (i.e., clusters or instructive clusters) can be processed with LLMs. In particular, the clusters be processed to generate annotated clusters. For example, clusters can be processed using LLMs (e.g., LLM 140) to generate annotated clusters with categorization (e.g., category annotation) and abstractive summarization …, para. [0048]; para. [0063]);
generate a second set of clusters from the set of cases using a clustering module, wherein each cluster of the second set of clusters is associated with a topic from a second set of topics (para. [0048]; the downstream prompting 110B may apply the one or more of a categorization 110B-1 (e.g., content categorization) and summarization 110B-2 (e.g., subject summarization) …, para. [0062]-[0063]; In this way, the volume of the dataset 102B can be reduced to tangible correlated topics, and the summaries for each of these topics can provide answers and relevant information for the question of what was potentially taken from the dataset 102B. …, para. [0064]); and
test a downstream machine learning model using one or more MFT cases selected from a particular topic of the second set of topics (In this way, the combination of clustering, the application of the content categorization (i.e., categorization 110B-1) and subject summarization (i.e., summarization 110B-2), and AI agents of the reasoned knowledge graph-building engine 112B can essentially bucket topics of interest or grouped topics of interest for data having particularly relevant semantics and significance associated with a target or goal. For example, the reasoned knowledge graph-building engine 112B may generate vulnerability graphs or other types of graphs from the buckets of interest that point to related topics of interest, such that the generated graph(s) can focus on the clusters that relate to those topics for correlating specific topics for the downstream analysis …, para. [0065])
Ailem does not explicitly disclose generating a set of minimal functionality test (MFT) cases or the use of MFT cases
However, these features are taught by Ribeiro (As a test of the model’s Negation capability, we use a Minimum Functionality test (MFT), i.e. simple test cases designed to target a specific behavior (Figure 1A). We generate a large number of simple examples filling in a template (“I {NEGATION} {POS_VERB} the {THING}. “) with pre-built lexicons, and compute the model’s failure rate on such examples …, sec. 1)
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Ribeiro with the medium of Ailem in arriving at the missing features of Ailem, because such combination would have resulted in determining the model’s failure rate (Ribeiro, sec. 1).
Per claim 17, Ailem in view of Ribeiro discloses the non-transitory computer-readable medium of claim 16,
Ailem discloses wherein: the pre-processing module comprises a uniform manifold approximation and projection (UMAP) tool for generating the set of lower-dimensional features (para. [0032]).
the clustering module comprises a hierarchical density-based spatial clustering of applications with noise based tool to generate the first set of clusters (para. [0033]; and
the extraction module comprises a bidirectional encoder representation of transformers topic based tool and extracting the set of representative data elements from each cluster comprises analyzing a set of frequency values associated with text data appearing in each cluster (para. [0035]).
2. Claims 5, 13 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Ailem in view of Ribeiro as applied to claims 1, 10 and 16 above, and further in view of Subraveti et al US 2024/0012844 A1 (“Subraveti”)
Per claim 5, Ailem in view of Ribeiro discloses the method of claim 1,
Ailem in view of Ribeiro does not explicitly disclose wherein the set of representative data elements for a particular cluster comprises a first representative data element and a second representative data element, and wherein the extraction module comprises a maximal marginal relevance (MMR) based tool with a predefined hyperparameter to compare the first representative data element to the second representative data element to determine a similarity score
However, this feature is taught by Subraveti (para. [0084]-[0086])
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Jain with the method of Ailem in view of Ribeiro in arriving at the missing features of Ailem in view of Ribeiro, because such combination would have resulted in controlling the diversity of key phrases (Subraveti, para. [0085]).
Per claim 13, Ailem in view of Ribeiro discloses the system of claim 10,
Ailem in view of Ribeiro does not explicitly disclose wherein the set of representative data elements for a particular cluster comprises a first representative data element and a second representative data element, and wherein the extraction module comprises a maximal marginal relevance (MMR) based tool with a predefined hyperparameter to compare the first representative data element to the second representative data element to determine a similarity score
However, this feature is taught by Subraveti (para. [0084]-[0086])
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Jain with the system of Ailem in view of Ribeiro in arriving at the missing features of Ailem in view of Ribeiro, because such combination would have resulted in controlling the diversity of key phrases (Subraveti, para. [0085]).
Per claim 18, Ailem in view of Ribeiro discloses the non-transitory computer-readable medium of claim 16,
Ailem in view of Ribeiro does not explicitly disclose wherein the set of representative data elements for a particular cluster comprises a first representative data element and a second representative data element, and wherein the extraction module comprises a maximal marginal relevance (MMR) based tool with a predefined hyperparameter to compare the first representative data element to the second representative data element to determine a similarity score
However, this feature is taught by Subraveti (para. [0084]-[0086])
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Jain with the medium of Ailem in view of Ribeiro in arriving at the missing features of Ailem in view of Ribeiro, because such combination would have resulted in controlling the diversity of key phrases (Subraveti, para. [0085]).
3. Claims 8, 15 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Ailem in view of Ribeiro as applied to claims 1, 10 and 16 above, and further in view of Jain et al US 2023/0267175 A1 (“Jain”)
Per claim 8, Ailem in view of Ribeiro discloses the method of claim 1,
Ailem in view of Ribeiro does not explicitly disclose labeling, using the first LLM, each data element of the input dataset with a positive sentiment label or a negative sentiment label based on the set of lower-dimensional features
However, this feature is taught by Jain (para. [0050]-[0053]; para. [0069])
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Jain with the method of Ailem in view of Ribeiro in arriving at the missing features of Ailem in view of Ribeiro, because such combination would have resulted in generating training data used to train an ML model (Jain, para. [0043]; para. [0047])
Per claim 15, Ailem in view of Ribeiro discloses the system of claim 10,
Ailem in view of Ribeiro does not explicitly disclose wherein the instructions further cause the one or more processors to label each data element of the input dataset with a positive sentiment label or a negative sentiment label based on the set of lower-dimensional features.
However, this feature is taught by Jain (para. [0050]-[0053]; para. [0069])
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Jain with the system of Ailem in view of Ribeiro in arriving at the missing features of Ailem in view of Ribeiro, because such combination would have resulted in generating training data used to train an ML model (Jain, para. [0043]; para. [0047])
Per claim 20, Ailem in view of Ribeiro discloses the non-transitory computer-readable medium of claim 16,
Ailem in view of Ribeiro does not explicitly disclose program code that is executable by the one or more processors to cause the one or more processors to label each data element of the input dataset with a positive sentiment label or a negative sentiment label based on the set of lower-dimensional features.
However, this feature is taught by Jain (para. [0050]-[0053]; para. [0069])
It would have been obvious to one of ordinary skill in the art before the effective filing of the invention to combine the teachings of Jain with the medium of Ailem in view of Ribeiro in arriving at the missing features of Ailem in view of Ribeiro, because such combination would have resulted in generating training data used to train an ML model (Jain, para. [0043]; para. [0047]).
Allowable Subject Matter
Claims 6, 14 and 19 are objected to as being dependent upon a rejected base claim but would be allowable (pending Applicant separately address the above 35 U.S.C. 101 issue) if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to OLUJIMI A ADESANYA whose telephone number is (571)270-3307. The examiner can normally be reached Monday-Friday 8:30-5:00pm.
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, Richemond Dorvil can be reached at 571-272-7602. 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.
/OLUJIMI A ADESANYA/Primary Examiner, Art Unit 2658
Prosecution Insights