Abstract: A system may include, as part of a doubly robust large language model (LLM) architecture for data summarization, a pre-trained LLM in addition to first and second fine-tuned LLMs derived from the pre-trained LLM and fine-tuned with training datasets collected over different time periods. In response to receiving a request to summarize a dataset, the dataset is provided as input to each of the LLMs to obtain, as output, a first summary, a second summary, and a third summary generated by the pre-trained LLM, first fine-tuned LLM, and second fine-tuned LLM, respectively. Each of the second and third summary may be compared to the first summary relative to a similarity threshold to determine whether a hallucination associated with the first or second fine-tuned LLM is indicated and, based on the determination, select one of the second or third summary as a response to the request or halt the data summarization.

Abstract: Methods and systems for reducing network traffic associated with authenticating user requests via interactionless authentication are disclosed. For example, in connection with a user device transmitting a user request to access a resource, the system determines a first address associated with the user device. The system then transmits the first address to a third-party authentication platform that determines a Uniform Resource Locator (URL) that is specific to a mobile carrier. The system then transmits a verification token request associated with the mobile carrier. The system receives a verification token from the URL, and transmits an authentication request to the third-party authentication platform comprising the verification token. The system then receives, from the third-party authentication platform, an authentication response comprising a user device identifier. In response to receiving the authentication response, the system performs an authentication action associated with the user request.

Abstract: Various embodiments are generally directed to authenticating a user for non-payment purposes utilizing a payment protocol, a computer device and a contactless card. The payment protocol may be consistent with an EMV standard. An application may determine that authorization or verification of a user may be required to access non-payment features of another application associated with the user and the computer device. The application may then receive and/or facilitate transmission of encrypted data from a communications interface of a contactless card associated with an account and utilizing either an offline or online technique to do so. The offline or online technique may involve one or more operations that can verify the identity of the user and/or otherwise authorize the user to have access to various aspects of the other application.

Abstract: Methods and systems described herein for validating machine learning models in federated machine learning model environments. More specifically, the methods and systems relate to unloading training and validation techniques to client devices using newly collected data to improve accuracy of federated machine learning models.

Abstract: In some implementations, a device may obtain first location data associated with one or more exchanges, where the first location data is associated with a first level of granularity. The device may determine second location data associated with the one or more exchanges based on records associated with the one or more exchanges, where the second location data is associated with a second level of granularity that is higher than the first level of granularity. The device may detect an authentication event associated with an access attempt to the account. The device may provide, based on detecting the authentication event, one or more authentication queries that are based on the second location data. The device may authenticate the access attempt to the account based on one or more responses to the one or more authentication queries.

Abstract: Systems and methods for determining the likelihood of a supplier accepting a certain transaction method such as a payment card. Exemplary systems can generate a machine learning predictive model to determine the likelihood of a supplier accepting a card payment.

Abstract: In some embodiments, a network operation may be authenticated via a dynamic selection of an access token. In some embodiments, use of a software application, configured to store a set of access tokens, in connection with processing a network operation may be detected. A risk value associated with each access token of the set of access tokens may be received in response to providing the network operation to a set of entities associated with the set of access tokens. Based on a determination, for each access token of the set of access tokens, whether the risk value associated with the access token satisfies a threshold risk value, the network operation may be authenticated using a selected access token associated with a first risk value that satisfies the threshold risk value.

Abstract: Described herein are systems, methods, and programming for enforcing attribute-level entitlements within a super-graph. Entitlements can be enforced prior to a graph API routing a request to a sub-graph, corresponding to pre-enforcement entitlement policies. Post-enforcement entitlements can also be applied to a response payload from the sub-graph prior to a response being provided to a requesting device. The entitlements can prevent certain data from being accessed. The entitlements can protect data down to the attribute-level (e.g., a graph endpoint). Furthermore, the entitlements can redact or otherwise obfuscate certain data.

Abstract: Disclosed embodiments may include a system for recognizing new devices. The system may receive data indicative of a user being associated with a new device, and may receive, via the new device, a request to perform an action. The system may generate, via a machine learning model (MLM), a trust score associated with the user and the new device. The system may determine whether the trust score exceeds a predetermined threshold. Responsive to determining the trust score does not exceed the predetermined threshold, the system may conduct fraud prevention action(s) with respect to the user and the new device. Responsive to determining the trust score exceeds the predetermined threshold, the system may cause the new device to display a notification. Responsive to receiving a response to the notification, the system may identify user setting(s) associated with a previous device, and enable the user setting(s) via the new device.

Abstract: According to certain aspects of the disclosure, a computer-implemented method may be used for regulating vehicle stock. The method may include receiving one or more queries indicative of one or more characteristics of a vehicle for purchase by a user and determining based on the one or more queries indicative of the one or more characteristics of the vehicle, at least one vehicle available for purchase at a location of a merchant. The method may also include determining a quantity of the at least one vehicle purchased and assigning a value to the at least one vehicle based on the quantity of the at least one vehicle purchased and a quantity of received queries about the vehicle. The method may also include transmitting the value to the user, with a recommendation regarding the at least one vehicle available for purchase based on the value.

Abstract: Systems and methods for an artificial intelligence model architecture that involves a first artificial intelligence model trained to map a plurality of entities to ranked documentation from a documentation source, and a second artificial intelligence model that comprises a language model trained to generate an additional query to run on the plurality of documents from the documentation source. By training the second model to generate additional queries as entities and/or links are discovered, the system may quickly and efficiently determine links and/or potential resolutions as well as received feedback thereon.

Abstract: Disclosed embodiments may include a method for systems and methods for providing mobility-impaired transactional or identify documents processing. The method may include receiving login credentials associated with an account and a request to deposit a check into the account, and identifying, using an image recognition model, a presence of a check in a visual field of the image capture device. Then the method may include obtaining, via the image capture device, a plurality of digital images of a front side of the check, and generating, based on the plurality of digital images of the front side of the check, a composite image of the front side of the check. The method may further include creating a recording of an endorsement gesture and transmitting the composite image of the front side of the check and the recording of the endorsement gesture to a back-end server.

Abstract: Systems, apparatuses, and methods are described for data labeling for training artificial intelligence systems. A candidate dataset comprising data samples and corresponding labels may be used to update an incumbent dataset comprise data samples and corresponding labels. The integrity of a data sample-label pair in the candidate dataset may be determined before the data sample-label pair is added to the incumbent dataset. For determining labeling integrity, a plurality of machine classifiers may be trained based on the incumbent dataset and portions of the candidate dataset. The plurality of machine classifiers as trained may be used to generate predicted labels for data samples in the candidate dataset. The integrity of the data sample-label pair in the candidate dataset may be measured based on the predicted labels for the data sample.

Abstract: A system determines, based on a request to assess access risk associated with an application, a plurality of layer access risk scores that are associated with the application. The system determines, based on access risk mitigation information and the plurality of layer access risk scores, a plurality of layer residual access risk scores. The system determines respective risk contribution weights associated with the plurality of layer residual access risk scores. The system determines, based on the respective risk contribution weights and on the plurality of layer residual access risk scores, an aggregated application access risk score associated with the application. The system causes, based on the aggregated application access risk score, one or more actions to be performed. For example, the system may initiate an access risk analysis process for the application and/or may generate an access control policy associated with the application.

Abstract: Systems and methods for a shared common microservice. In some aspects, the system may receive a domain model. The domain model may include a resource module and a service module. The system may receive, from a user device, a first request. The first request may include a first set of resources and a first task to be completed. The system may retrieve, from the domain model, a required set of resources corresponding to the first set of resources that a first API requires access to complete the first task. Based on the required set of resources from the domain model, the system may generate a template for the first API. Based on the template, the system may generate a first endpoint for the first API. The system may deploy the first endpoint. The first API accesses one or more microservice functions at the execution of the first task.

Abstract: Aspects described herein may facilitate an automated trade-in of a vehicle with limited human interaction. A server may receive a request to begin a value determination of a vehicle associated with the user. The server may receive first data comprising: vehicle-specific identifying information, and multimedia content showing a first aspect of the vehicle. The user may be directed to place the vehicle within a predetermined staging area. The server may receive, from one or more image sensors associated with the staging area, second data comprising multimedia content showing a second aspect of the vehicle. The server may create a feature vector comprising the first data and the second data. The feature vector may be inputted into a machine learning algorithm corresponding to the vehicle-specific identifying information of the vehicle. Based on the machine learning algorithm, the server may determine a value of the vehicle.

Abstract: Systems and methods for improving network security across computer networks when granting access entitlements to a plurality of applications. For example, the system may receive, from a first account, a first access request for a first application entitlement. The system may retrieve a first access credential profile for the first account. The system may determine a first embedding for the first access credential profile. The system may retrieve a first required embedding for the first application entitlement. The system may determine a first dimensionality conflict between the first embedding and the first required embedding. The system may determine a dependency between the first dimensionality conflict and a credential attribute. The system may query the first account for the credential attribute.

Abstract: A computer-implemented method includes: receiving a request to scan a resource for detection of a vulnerability, the resource comprising one or more of: a component of an application, a code-base of the application, or a third-party library associated with the application; determining one or more matching vulnerability scanning operators among a plurality of vulnerability scanning operators based at least in part on one or more properties associated with the request; transmitting the request to the one or more matching vulnerability scanning operators, the one or more matching vulnerability scanning operators having access to the resource; receiving an initial result associated with the request from the one or more matching vulnerability scanning operators; obtaining a predetermined format for the initial result; generating a translated result by modifying the initial result based on the predetermined format; and transmitting the translated result to a source of the request.

Abstract: Embodiments disclosed are directed to a computing system that performs steps to automatically identify risk control features and entities in a risk control document. The computing system regenerates, by a semantic prediction machine learning (ML) model, phrases in a risk control document. The computing system then classifies, by the semantic prediction ML model, risk control features associated with the regenerated phrases. Subsequently, the computing system corrects, by a discriminative natural language processing (NLP) model, the classified risk control features based on the phrases and the regenerated phrases.

Abstract: In some aspects, a computing system may generate uninformative features that may be added to a dataset of real features to use as a baseline for determining the quality of an explanation of model output. The uninformative features may be features that do not correlate with what a model is tasked with predicting (e.g., the uninformative features may be random values), and the real features may be informative and correlate with what the model is tasked with predicting (e.g., variables of a dataset sample). A machine learning model may be trained on a dataset that includes both the real features and the uninformative features. The computing system may generate feature attributions for model output, which may include feature attributions for the uninformative features and the real features in the dataset.