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 instant application having Application No. 18982689 has a total of 25 claims pending in the application, of which claims 4, 11, 18, and 21-22 have been cancelled.
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 1-3, 5-10, 12-17, 19-20, and 23-25 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1: Claim 1 is a process type claim. Claim 8 is a machine type claim. Claim 15 is a manufacture type claim. Therefore, claims 1-3, 5-10, 12-17, 19-20 and 23-25 are directed to either a process, machine, manufacture or composition of matter.
As per claim 1,
2A Prong 1:
“grouping the data instance having common features into a single prompt based at least in part on a character limit…, wherein the common features includes one or both of amounts and data instance types” An Accountant mentally or with pencil and paper looks at the financial data and groups it into a single prompt based on the data within and the limits of the amount of data they want to deal with at one time.
“providing … the single prompt that comprises instructions to classify the data instances in one or more classes, the one or more classes comprising at least one of: an anomalous class, a typical class, and an indeterminate class… wherein the anomalous class represents data instances that were performed without consent of a user, the typical class represents data instances that conform with other data instances, and the indeterminate class represents data instances for which the … is unable to classify” The accountant mentally or with pencil and paper classifies the data into anomalous, typical, or indeterminate classes based on what is in the financial data.
“Based at least in part on providing the single prompt … obtaining… first output data” The Accountant mentally or with pencil and paper determines the classifications of the financial data.
“determining … that a specific type of class for a first data instance and a second data instance of the data instances in the first output data matches the indeterminate class” The Accountant mentally or with pencil and paper determines that the output includes at least one set of financial data that they can’t determine the class.
“generating … a subsequent prompt that includes the data instance, the indeterminate class, and user specific account history associated with the user that was not provided in the single prompt” The Accountant mentally or with pencil and paper considers additional data about the unknown data item in order to classify it.
“Based at least in on providing the subsequent prompt… obtaining, …. Second output data identifying that the first data instance is labeled with the typical class, and the second data item is labeled with the anomalous class wherein the first data instance comprises a first financial transaction and the second data instance comprises a second financial transaction” The Accountant mentally or with pencil and paper considers the additional data and determines that the unknown financial data is actually anomalous!
“Classifying the first financial transaction with the typical class based at least in part on the first output data or the second output data…” The Accountant mentally or with pencil and paper determines that the first financial transaction is normal based on the data he is examining.
“Causing … to process the first financial transaction based at least in part on the typical class” The Accountant mentally or with pencil and paper processes the typical data because it is normal financial work.
“classifying the second financial transaction with the anomalous class based at least in part on the first output data or the second output data obtained from the LLM” The Accountant mentally or with pencil and paper classifies the second financial transaction as anomalous based on his examination of the data both before and after the new information he considered.
“Causing the … to refrain from processing the second financial transaction based at least in part on the anomalous class” The Accountant mentally or with pencil and paper decides not to process the financial transaction due to the anomalous classification.
“Determining … one or more labels for the first data instance and the second data instance based at least in part on the first output or the second output data…” The Accountant mentally or with pencil and paper labels the financial transactions based on his determinations of their class.
“Executing … one or more operations based at least in part on classifying the first financial transaction and the second financial transaction, wherein the one or more operations comprise” The Accountant mentally or with pencil and paper takes the appropriate actions to implement the financial transactions as needed.
“adjusting or initiating … a monitoring process for monitoring subsequent data instances that are associated with the user that initiated the second financial transaction” The Accountant mentally or with pencil and paper makes a note to further examining subsequent transactions from the user who made the anomalous transaction
2A Prong 2: This judicial exception is not integrated into a practical application.
Additional elements:
A computing device, a receiver computer (mere instructions to apply the exception using a generic computer component);
“an LLM”, “the LLM”, “the LLM being previously trained to classify instances of input data based at least in part on the anomalous class, the typical class, and the indeterminate class” (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: Claims require a generic LLM with no additional limitations or details that make it anything more than a generic, off the shelf LLM model.
“collecting, by a computing device, data instances that meet a set of criteria based at least in part on historical data and a prioritized performance, wherein the set of criteria comprises at least one of: an amount that exceeds a value for an account associated with a user or a previous data instance that is anomalous that is associated with the user”, “displaying, by a graphical user interface, the second data instance and the one or more labels” Adding insignificant extra-solution activity to the judicial exception - see MPEP 2106.05(g)).
2B: The claim does not include additional elements individually or in combination that are sufficient to amount to significantly more than the judicial exception.
Additional elements:
A computing device, a receiver computer (mere instructions to apply the exception using a generic computer component)
“an LLM”, “the LLM”, “the LLM being previously trained to classify instances of input data based at least in part on the anomalous class, the typical class, and the indeterminate class” (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: Claims require a generic LLM with no additional limitations or details that make it anything more than a generic, off the shelf LLM model.
“collecting, by a computing device, data instances that meet a set of criteria based at least in part on historical data and a prioritized performance, wherein the set of criteria comprises at least one of: an amount that exceeds a value for an account associated with a user or a previous data instance that is anomalous that is associated with the user”, “displaying, by a graphical user interface, the second data instance and the one or more labels” (MPEP 2106.05(d)(II) indicate that merely “receiving or transmitting data” is a well‐understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed collecting and displaying steps are well-understood, routine, conventional activity is supported under Berkheimer).
As per claims 2-3, and 5-6, these claims describe similar mental steps to claim 1, and are rejected for similar reasons to claim 1.
As per claims 7 and 24, these claims describe similar mental steps and similar generic machine learning models similar to claim 1, and is rejected for similar reasons to claim 1.
As per claim 25, this claim describes similar transmitting and receiving of data as described in claim 1, and is rejected for similar reasons to claim 1.
As per claim 8,
2A Prong 1:
“group the data instance having common features into a single prompt based at least in part on a character limit…, wherein the common features includes one or both of amounts and data instance types” An Accountant mentally or with pencil and paper looks at the financial data and groups it into a single prompt based on the data within and the limits of the amount of data they want to deal with at one time.
“provide … the single prompt that comprises instructions to classify the data instances in one or more classes, the one or more classes comprising at least one of: an anomalous class, a typical class, and an indeterminate class… wherein the anomalous class represents data instances that were performed without consent of a user, the typical class represents data instances that conform with other data instances, and the indeterminate class represents data instances for which the … is unable to classify” The accountant mentally or with pencil and paper classifies the data into anomalous, typical, or indeterminate classes based on what is in the financial data.
“Based at least in part on providing the single prompt … obtaining… first output data” The Accountant mentally or with pencil and paper determines the classifications of the financial data.
“determine … that a specific type of class for a first data instance and a second data instance of the data instances in the first output data matches the indeterminate class” The Accountant mentally or with pencil and paper determines that the output includes at least one set of financial data that they can’t determine the class.
“generate … a subsequent prompt that includes the data instance, the indeterminate class, and user specific account history associated with the user that was not provided in the single prompt” The Accountant mentally or with pencil and paper considers additional data about the unknown data item in order to classify it.
“Based at least in on providing the subsequent prompt… obtain, …. Second output data identifying that the first data instance is labeled with the typical class, and the second data item is labeled with the anomalous class wherein the first data instance comprises a first financial transaction and the second data instance comprises a second financial transaction” The Accountant mentally or with pencil and paper considers the additional data and determines that the unknown financial data is actually anomalous!
“Classify the first financial transaction with the typical class based at least in part on the first output data or the second output data…” The Accountant mentally or with pencil and paper determines that the first financial transaction is normal based on the data he is examining.
“Cause … to process the first financial transaction based at least in part on the typical class” The Accountant mentally or with pencil and paper processes the typical data because it is normal financial work.
“classify the second financial transaction with the anomalous class based at least in part on the first output data or the second output data obtained from the LLM” The Accountant mentally or with pencil and paper classifies the second financial transaction as anomalous based on his examination of the data both before and after the new information he considered.
“Cause the … to refrain from processing the second financial transaction based at least in part on the anomalous class” The Accountant mentally or with pencil and paper decides not to process the financial transaction due to the anomalous classification.
“Determine … one or more labels for the first data instance and the second data instance based at least in part on the first output or the second output data…” The Accountant mentally or with pencil and paper labels the financial transactions based on his determinations of their class.
“Execute … one or more operations based at least in part on classifying the first financial transaction and the second financial transaction, wherein the one or more operations comprise” The Accountant mentally or with pencil and paper takes the appropriate actions to implement the financial transactions as needed.
“adjust or initiate … a monitoring process for monitoring subsequent data instances that are associated with the user that initiated the second financial transaction” The Accountant mentally or with pencil and paper makes a note to further examining subsequent transactions from the user who made the anomalous transaction
2A Prong 2: This judicial exception is not integrated into a practical application.
Additional elements:
A computing device, one or more processors, one or more memories, a receiver computer (mere instructions to apply the exception using a generic computer component);
“an LLM”, “the LLM”, “the LLM being previously trained to classify instances of input data based at least in part on the anomalous class, the typical class, and the indeterminate class” (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: Claims require a generic LLM with no additional limitations or details that make it anything more than a generic, off the shelf LLM model.
“collect, by a computing device, data instances that meet a set of criteria based at least in part on historical data and a prioritized performance, wherein the set of criteria comprises at least one of: an amount that exceeds a value for an account associated with a user or a previous data instance that is anomalous that is associated with the user”, “displaying, by a graphical user interface, the second data instance and the one or more labels” Adding insignificant extra-solution activity to the judicial exception - see MPEP 2106.05(g)).
2B: The claim does not include additional elements individually or in combination that are sufficient to amount to significantly more than the judicial exception.
Additional elements:
A computing device, one or more processors, one or more memories, a receiver computer (mere instructions to apply the exception using a generic computer component)
“an LLM”, “the LLM”, “the LLM being previously trained to classify instances of input data based at least in part on the anomalous class, the typical class, and the indeterminate class” (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: Claims require a generic LLM with no additional limitations or details that make it anything more than a generic, off the shelf LLM model.
“collect, by a computing device, data instances that meet a set of criteria based at least in part on historical data and a prioritized performance, wherein the set of criteria comprises at least one of: an amount that exceeds a value for an account associated with a user or a previous data instance that is anomalous that is associated with the user”, “displaying, by a graphical user interface, the second data instance and the one or more labels” (MPEP 2106.05(d)(II) indicate that merely “receiving or transmitting data” is a well‐understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed collecting and displaying steps are well-understood, routine, conventional activity is supported under Berkheimer).
As per claims 9-10, and 12-13, these claims describe similar mental steps to claim 8, and are rejected for similar reasons to claim 8.
As per claim 14, this claim describes similar mental steps and similar generic machine learning models similar to claim 8, and is rejected for similar reasons to claim 8.
As per claim 15,
2A Prong 1:
“group the data instance having common features into a single prompt based at least in part on a character limit…, wherein the common features includes one or both of amounts and data instance types” An Accountant mentally or with pencil and paper looks at the financial data and groups it into a single prompt based on the data within and the limits of the amount of data they want to deal with at one time.
“provide … the single prompt that comprises instructions to classify the data instances in one or more classes, the one or more classes comprising at least one of: an anomalous class, a typical class, and an indeterminate class… wherein the anomalous class represents data instances that were performed without consent of a user, the typical class represents data instances that conform with other data instances, and the indeterminate class represents data instances for which the … is unable to classify” The accountant mentally or with pencil and paper classifies the data into anomalous, typical, or indeterminate classes based on what is in the financial data.
“Based at least in part on providing the single prompt … obtaining,… first output data” The Accountant mentally or with pencil and paper determines the classifications of the financial data.
“determine … that a specific type of class for a first data instance and a second data instance of the data instances in the first output data matches the indeterminate class” The Accountant mentally or with pencil and paper determines that the output includes at least one set of financial data that they can’t determine the class.
“generate … a subsequent prompt that includes the data instance, the indeterminate class, and user specific account history associated with the user that was not provided in the single prompt” The Accountant mentally or with pencil and paper considers additional data about the unknown data item in order to classify it.
“Based at least in on providing the subsequent prompt… obtain, …. Second output data identifying that the first data instance is labeled with the typical class, and the second data item is labeled with the anomalous class wherein the first data instance comprises a first financial transaction and the second data instance comprises a second financial transaction” The Accountant mentally or with pencil and paper considers the additional data and determines that the unknown financial data is actually anomalous!
“Classify the first financial transaction with the typical class based at least in part on the first output data or the second output data…” The Accountant mentally or with pencil and paper determines that the first financial transaction is normal based on the data he is examining.
“Cause … to process the first financial transaction based at least in part on the typical class” The Accountant mentally or with pencil and paper processes the typical data because it is normal financial work.
“classify the second financial transaction with the anomalous class based at least in part on the first output data or the second output data obtained from the LLM” The Accountant mentally or with pencil and paper classifies the second financial transaction as anomalous based on his examination of the data both before and after the new information he considered.
“Cause the … to refrain from processing the second financial transaction based at least in part on the anomalous class” The Accountant mentally or with pencil and paper decides not to process the financial transaction due to the anomalous classification.
“Determine … one or more labels for the first data instance and the second data instance based at least in part on the first output or the second output data…” The Accountant mentally or with pencil and paper labels the financial transactions based on his determinations of their class.
“Execute … one or more operations based at least in part on classifying the first financial transaction and the second financial transaction, wherein the one or more operations comprise” The Accountant mentally or with pencil and paper takes the appropriate actions to implement the financial transactions as needed.
“adjust or initiate … a monitoring process for monitoring subsequent data instances that are associated with the user that initiated the second financial transaction” The Accountant mentally or with pencil and paper makes a note to further examining subsequent transactions from the user who made the anomalous transaction
2A Prong 2: This judicial exception is not integrated into a practical application.
Additional elements:
A non-transitory computer readable storage medium, one or more processors, A computing device, a receiver computer (mere instructions to apply the exception using a generic computer component);
“an LLM”, “the LLM”, “the LLM being previously trained to classify instances of input data based at least in part on the anomalous class, the typical class, and the indeterminate class” (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: Claims require a generic LLM with no additional limitations or details that make it anything more than a generic, off the shelf LLM model.
“collect, by a computing device, data instances that meet a set of criteria based at least in part on historical data and a prioritized performance, wherein the set of criteria comprises at least one of: an amount that exceeds a value for an account associated with a user or a previous data instance that is anomalous that is associated with the user”, “displaying, by a graphical user interface, the second data instance and the one or more labels” Adding insignificant extra-solution activity to the judicial exception - see MPEP 2106.05(g)).
2B: The claim does not include additional elements individually or in combination that are sufficient to amount to significantly more than the judicial exception.
Additional elements:
A non-transitory computer readable storage medium, one or more processors, A computing device, a receiver computer (mere instructions to apply the exception using a generic computer component)
“an LLM”, “the LLM”, “the LLM being previously trained to classify instances of input data based at least in part on the anomalous class, the typical class, and the indeterminate class” (Adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea - see MPEP 2106.05(f) – Examiner’s note: Claims require a generic LLM with no additional limitations or details that make it anything more than a generic, off the shelf LLM model.
“collect, by a computing device, data instances that meet a set of criteria based at least in part on historical data and a prioritized performance, wherein the set of criteria comprises at least one of: an amount that exceeds a value for an account associated with a user or a previous data instance that is anomalous that is associated with the user”, “displaying, by a graphical user interface, the second data instance and the one or more labels” (MPEP 2106.05(d)(II) indicate that merely “receiving or transmitting data” is a well‐understood, routine, conventional function when it is claimed in a merely generic manner (as it is in the present claim). Thereby, a conclusion that the claimed collecting and displaying steps are well-understood, routine, conventional activity is supported under Berkheimer).
As per claims 16-17, 19 and 23, these claims describe similar mental steps to claim 15, and are rejected for similar reasons to claim 15.
As per claim 20, this claim describes similar mental steps and similar generic machine learning models similar to claim 15, and is rejected for similar reasons to claim 15.
Response to Arguments
In pg.17, the Applicant argues in regards to the rejection under U.S.C. 101,
Regarding (e), step (e) integrates the abstract idea into a practical application by directing a receiver computer to process a financial transaction based at least in part on the typical class provided by the LLM. The receiver computer evaluates the classification output and selectively advances only those transactions identified as typical, which operationalizes the system's ability to mitigate risk and enhance throughput. This implementation, supported by paragraphs [0033] and [0069] of the specification, demonstrates an improvement over conventional systems by shifting the determination and processing decision closer to the beginning of the workflow. This allocation of processing resources allows the system to avoid unnecessary downstream analysis or reversal procedures, reflecting a practical application of anomaly detection in real-world financial operations.
In response, the Examiner maintains the rejection as shown above. The use of generic computer equipment and machine learning models such as a “receiver computer” and “the LLM” with the abstract idea of examining and classifying financial data does not cause the claim to be significantly more than the abstract idea. Making decisions about the financial records in a different order constitutes an improvement to the abstract idea, and not to the LLM or computer system, and therefore the rejection is maintained as shown above.
In pg.17-18, the Applicant further argues in regards to the rejection under U.S.C. 101,
Step (f) integrates the abstract idea into a practical application by classifying a financial transaction with the anomalous class, which triggers a specific operational response. The computing device, relying on the LLM's output, identifies transactions likely to be unauthorized or outlier events and marks them accordingly. This classification process, discussed in paragraphs [0033] and [0069], is used to drive subsequent actions within the computing environment, such as flagging the transaction for review or initiating procedures that prevent further processing. By embedding the classification process directly within the transaction workflow, the system implements anomaly detection in a manner that actively reduces risk and improves the responsiveness of transaction management.
In response, the Examiner maintains the rejection as shown above. Once again, the acts of examining and flagging/labeling/classifying financial records is an abstract idea that can be performed entirely within the human mind. Labeling financial records as anomalies or as typical documents is an action that an accountant has performed for thousands of years, and merely using generic computer equipment and machine learning models does not cause this to be significantly more than the abstract idea. The claimed improvements are to the abstract idea, and not the computer hardware or machine learning models, and therefore the rejection is maintained as shown above.
In pg.18, the Applicant argues in regards to the rejection under U.S.C. 101,
Step (g) integrates the abstract idea into a practical application by causing the receiver computer to refrain from processing a financial transaction identified as anomalous. The operational decision to halt processing is based on the LLM's output and is executed to stop the financial transaction, as outlined in paragraphs [0033], [0069], and [0071]. This feature demonstrates a practical application by preventing the completion of potentially fraudulent or unauthorized transactions, reducing the likelihood of subsequent remediation or reversal efforts. The integration of this decision point into the transaction workflow exemplifies how the claimed subject matter is applied to improve the efficiency, security, and reliability of financial processing systems.
In response, the Examiner maintains the rejection as shown above. Once again, Applicant is claiming that actions like flagging financial records as anomalous or fraudulent and making decisions about those documents is an improvement to financial processing systems. These are not improvements to the computer hardware or machine learning models, they are improvements to the abstract idea of examining and responding to potentially fraudulent financial transactions, and therefore are not significantly more than the abstract idea. Therefore the rejection is maintained as shown above.
In pg.18, the Applicant further argues in regards to the rejection under U.S.C. 101,
Step (h) reflects integration into a practical application by executing one or more operations that include adjusting or initiating a monitoring process for subsequent data instances associated with the user and displaying, through a graphical user interface, the second data instance and its corresponding labels. As described in specification paragraphs [0033] and [0071], these operations go beyond classification and extend the use of the LLM's output to real-time system management and user interaction. Adjusting or initiating monitoring based on detected anomalies allows the system to adaptively focus resources on users or transactions exhibiting unusual patterns, which improves ongoing risk management and system responsiveness. Displaying the second data instance and relevant labels using a graphical user interface enables direct user or operator review, facilitating intervention and oversight. By linking the anomaly detection and classification directly to operations in the transaction processing environment, step (h) grounds the process in a technical context and demonstrates the claim's integration into a practical application with meaningful limits.
In response, the Examiner maintains the rejection as shown above. Choosing to take further actions of examining other documents is, as described above, an improvement to the abstract idea, and not to a technology. Displaying the data is extra-solutionary activity that does not cause the claim to be significantly more than the abstract idea. Therefore the rejection is maintained as shown above.
In pg.18-19, the Applicant further argues for the 101 rejection,
Step (b) demonstrates practical application through grouping data instances with common features into a single prompt, taking into account character limits of the LLM. This grouping, as described in the specification at paragraphs [0037] and [0038], enables the system to optimize prompt management, reduce unnecessary processing, and conserve memory. By partitioning data based on shared characteristics such as amounts or data instance types, the system efficiently formats prompts for analysis by the LLM, directly addressing technical constraints and improving throughput and performance.
The nuances of using a generic machine learning model such as an LLM, such as there being a limit to what the algorithm can respond to, is not enough to make the algorithm anything more than a generic, off the shelf LLM model. All LLM models will have some limit to what data they can process. The manipulation of data in the prompts for the LLM merely discloses manipulating the data that is chosen to be used with the generic LLM, and including these limitations in the claim does not cause the claim to be significantly more than the abstract idea, and therefore the rejection is maintained as shown above.
In pg.19, the Applicant further argues in regards to the rejection under U.S.C. 101,
Step (c) integrates the abstract idea into a practical application by providing instructions to the LLM to classify data instances in one or more classes, including anomalous, typical, and indeterminate classes, where the LLM has been trained to distinguish among these based on input data. The specification, particularly at paragraphs [0028] and [0039], describes how these classifications enable the system to differentiate between outlier transactions, typical transactions, and those where the LLM cannot make a determination. This classification step not only supports automated decision-making but also informs subsequent actions such as prompt generation and transaction processing. The use of LLMs for real-time, context-aware classification addresses technical challenges in anomaly detection and provides a concrete improvement over conventional systems.
In response, the Examiner maintains the rejection as shown above. Manipulation of the inputs and outputs of the generic machine learning algorithm denotes an improvement to the abstract idea of examining and classifying financial records, and does not denote an improvement to the machine learning model. Improvements to the abstract idea does not cause the claim to be significantly more than the abstract idea, and therefore the rejection is maintained as shown above.
In pg.19, the Applicant further argues in regards to the rejection under U.S.C. 101,
Still further, step (d) further integrates the abstract idea into a practical application by generating a subsequent prompt that includes both the relevant data instance and additional user- specific account history not present in the original prompt. According to specification paragraphs [0036] and [0072], this step enables the system to augment prompts only when the LLM identifies an indeterminate class, thus avoiding unnecessary augmentation for all data instances. By selectively adding context and resubmitting for classification, the method reduces memory and processing demands, improves classification accuracy, and streamlines anomaly detection. This is a specific technical solution that optimizes performance and addresses limitations in LLM-based classification, demonstrating significantly more than generic application of an abstract idea.
In response, the Examiner maintains the rejection as shown above. Adding additional data when the classification of a financial record is unclear does not constitute and improvement to the LLM model, it constitutes an improvement to the abstract idea of examining financial records. Improvements to the abstract idea does not cause the claim to be significantly more than the abstract idea, and therefore the rejection is maintained as shown above.
In pg.21, the Applicant argues in regards to the rejection under U.S.C. 101,
The specification, in paragraph [0001], identifies that prior systems use statistical methods such as isolation forest statistics for anomaly detection in complex datasets, which may not function effectively with large or complicated data and do not perform further classification as part of the anomaly detection process. Paragraph [0023] explains that conventional approaches require extensive processing power and may lack the capability to make timely or accurate determinations, particularly in a context of real-time or large-scale data. Paragraph [0024] further describes that conventional systems do not perform multi-level classification (e.g., by category, subcategory, or code/identifier) in the same process as anomaly detection, which results in additional resource usage and increased latency.
In response, the Examiner maintains the rejection as shown above. Applicant argues that they do things differently than other systems that examine financial records. However, the question is not whether or not the improvement to the abstract idea of examining financial records is novel or non-obvious, it is whether or not it is an abstract idea. Here the claims denote the abstract idea of examining financial records for anomalous material, and adding in improvement to that process along with generic computer hardware or generic machine learning models does not cause the claims to be significantly more than the abstract idea, and therefore the rejection is maintained as shown above.
In pg.22, the Applicant further argues in regards to the rejection under U.S.C. 101,
The specification emphasizes that conventional systems in industrial and financial contexts encounter significant challenges when attempting to detect anomalies within vast and heterogeneous datasets, particularly where instances have not been previously associated with categories, subcategories, or code identifiers (see, e.g., paragraphs [0001], [0023], and [0024]). The instant application specifically addresses these technical challenges by providing computer- implemented solutions that aggregate, format, and analyze incoming transactional data using large language models trained for classification and anomaly detection, and perform real-world operations based on the anomaly. This technical advancement is situated within the technology field of device-based data analytics and machine learning models, where an objective is to automate the identification of irregularities and to assign meaningful classifications to new data instances for purposes such as fraud detection, and take appropriate action such as push or pull transactions, refrain from transacting, and changing monitoring processes associated with anomalous transactions.
In response, the Examiner maintains the rejection as shown above. The question in determining whether or not something is an abstract ides is not done based upon how long the processing might take or what actions previous examples of the abstract idea have performed. The question is not whether a generic computer or generic machine learning model can make processing or looking at things easier, that is a known fact for any generic processor or generic machine learning model. There is no improvement to a technology here, only improvements to the abstract idea of examining financial documents for anomalies, all of which is an abstract idea as described above. Therefore the rejection is maintained as shown above.
In pg.22-23, the Applicant argues in regards to the rejection under U.S.C. 101,
For example, step (a) recites that a computing device collects data instances meeting a set of criteria based at least in part on historical data and prioritized performance. Specification paragraph [0063] describes the detection and classification module 514 applying an algorithm or rule set that prioritizes performance, SO that only particular data instances meeting the set criteria are analyzed for anomalies or for classification into categories, subcategories, and national code identifiers. By focusing computational resources on data instances that are more likely to be relevant according to objective factors like historical norms and performance needs, the limitation allocates memory and processing cycles efficiently. This processing improves computer efficiency and reduces system resource consumption compared to conventional systems that evaluate all available data without regard to relevance, which results in slower response times and higher computational resource demands. Selecting data instances based on prioritized performance enables the system to scale more effectively and maintain responsive anomaly detection and classification operations, especially with larger or more complex datasets.
The allocation of memory or the use of processing cycles in an efficient manner to run an abstract idea, such as the examining of financial records for anomalies, is not an improvement to a technology, it is an improvement to the abstract idea. The processor and memory are not improved or changed in any way by what types of data run through their systems. Improvement to the amount of data going into or being processed by a system is an improvement to the abstract idea being processed, not the processor. Therefore the rejection is maintained as shown above.
In pg.23, the Applicant further argues in regards to the rejection under U.S.C. 101,
Step (b) recites grouping data instances with shared features into a single prompt, based on LLM character limits. This addresses practical constraints of LLM-based processing, as detailed in paragraphs [0037]-[0038] ("if the input data includes over a threshold number of data instances... partition the data instances into prompt groups based on similar prompt formats in order to, without limitation, reduce memory requirements, reduce processing power requirements, reduce the amount of data the LLM needs to process to make a determination, reduce the time taken the process the input data, or any suitable combination thereof."). This technique further optimizes memory and processing requirements, reflecting a real-world technical solution to LLM prompt size constraints.
As stated above, improvements to the data going into or coming out of a generic machine learning model are improvements to the abstract idea, not the machine learning model itself. Here adjusting the prompt going into a generic LLM is manipulating the abstract idea, not the machine learning model. Improvements to processing using a generic processor or machine learning model is an improvement to the abstract idea, and this is not enough to cause the claims to be significantly more than the abstract idea. Therefore the rejection is maintained as shown above.
In pg.23, the Applicant further argues in regards to the 101 rejection,
Step (d) provides further technical improvements. For example, step (d) recites generating a subsequent prompt including additional user-specific account history for instances classified as "indeterminate." This limitation anticipates cases where the LLM may be unable to classify a data instance, and augments the prompt only for those instances, rather than for all input data, as supported in paragraphs [0039] and [0066] ("the ADE 102 may be configured to provide a subsequent prompt in response to receiving an indeterminate class identifier as output from the LLM 104., the subsequent prompt may include data that differs from the initially provided prompt (e.g., user specific account history that was not provided in the initial input data)"). This limitation reduces resource consumption because not all data is augmented, only those that require it, which is a practical technical improvement over approaches that would require repeated attempts or blanket augmentation.
IN response, the Examiner maintains the rejection as shown above. As stated previously, having the system look at additional data when a classification is not able to be determined is a mental process. If a document is not easily classified, the person can read more of the document, or look at metadata about the document, in order to classify it. Adding additional data is not an improvement to a technology, it is an improvement to the abstract idea. Therefore the rejection is maintained as shown above.
In Pg.24, the Applicant further argues in regards to the rejection under U.S.C. 101,
The combination of features as a whole in amended claim 1 provides significantly more than any alleged abstract idea. The claim recites a series of specific steps at any one of steps (a)- (h), or as a whole, among others, that address technical challenges encountered in anomaly detection and classification of complex transactional data using large language models. For example, the limitations include targeted selection of data instances based on historical norms and prioritized performance, grouping data for prompt management in view of LLM input constraints, and executing operations such as adjusting monitoring and controlling transaction processing based on real-time classifications.
In response, the Examiner maintains the rejection as shown above. There is no improvement to a technology here. The examination of financial transactions to detect fraud or other anomalies has been done as long as there has been commerce between human beings. Merely including a generic LLM and generic computer equipment to apply the abstract idea does not make the claim significantly more than the abstract idea, and therefore the rejection is maintained as shown above.
In pg.24, the Applicant further argues in regards to the rejection under U.S.C. 101,
These limitations contribute technical improvements by optimizing computational resources, reducing unnecessary processing, and improving responsiveness in anomaly detection systems. The specification, including paragraphs [0063], [0033], [0069], [0071], and [0074], describes how each step is implemented to solve particular technical problems, such as efficient memory allocation and early intervention in transaction processing. In accordance with MPEP 2106.04(d)(1) and 2106.05(a), the claim provides a solution that significantly enhances the functioning of computer systems involved in anomaly detection and classification, demonstrating that the claim includes features well beyond any purported abstract idea, and thus amount to significantly more (STEP 2B: YES).
In response, the Examiner maintains the rejection as shown above. As discussed previously, there is no technology here. The claims deal with the abstract idea and mental process of examining financial documents for anomalies. Efficient use of generic hardware and generic machine learning algorithms for the implementation of an abstract idea is not significantly more than the abstract idea itself, and therefore the rejection is maintained as shown above.
Applicant's remaining arguments with respect to claims 1-3, 5-10, 12-17, 19-20, and 23-25 have been considered but are either conclusory or moot in view of the new ground(s) of rejection given above.
Allowable Subject Matter
Claims 1-3, 5-10, 12-17, 19-20, and 23-25 would be considered allowable since when reading the claims in light of the specification should the 101 rejection be overcome. As per MPEP $2111.01 none of the references of record alone or in combination disclose or suggest the combination of limitations found within the independent claims. While there is no individual limitation that is allowable, the combination of details found in the independent claims would be a non-obvious combination for one of ordinary skill in the art. Bakumenko-2
The closest prior art, Bakumenko, shows the use of LLMs and prompts in order to detect fraud (i.e. anomalies) in financial documents. Bakumenko-2 denotes the details of fraud including looking at anomalous values in financial documents (Bakumenko-2, pg.9, Table 5). The Brown reference denotes finding ambiguous documents and using additional data to determine the class of the data in the prompt. The Collins reference describes changing monitoring of users based upon discoveries by LLMs.
However, the particulars of the user details including consent for financial transactions, particular labels inside particular prompts (one being typical, one being anomalous) and particular actions taken upon that particular classification has not been found spelled out in the prior art, and therefore the combination of limitations are allowable over the prior art, should the 101 rejection be overcome.
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 BEN M RIFKIN whose telephone number is (571)272-9768. The examiner can normally be reached Monday-Friday 9 am - 5 pm.
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, Alexey Shmatov can be reached at (571) 270-3428. 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.
/BEN M RIFKIN/Primary Examiner, Art Unit 2123
Prosecution Insights