SAS INSTITUTE INC.

A system, method, and computer-program product includes obtaining a data query schema that provides instructions for constructing computer-executable search queries according to a search syntax permitted by a target data query language, extracting, from the data query schema, one or more sets of query components that define the search syntax of the target data query language, synthetically generating a plurality of natural language-to-search query training data samples based at least in part on the one or more sets of query components, and configuring a natural language-to-search query machine learning model based on training a machine learning text-to-text transformer model using the plurality of natural language-to-search query training data samples.

A system and method include receiving a plurality of elements in a data array to be sorted, assigning the elements a sequential index value, computing an amount of additional memory needed to sort the elements, creating a Linked-Array Tournament Tree (LATT), storing the LATT in the allocated amount of additional memory, initializing the LATT, initializing a number of iterations, initializing a next sorted index position in the data array, determining from the LATT, the index value of a champion element from all the winners of the pairwise comparisons, swapping the champion element with an element at the next sorted index position in the data array, updating the LATT responsive to the swapping, updating the next sorted index position, repeating a plurality of times, and outputting the data array having the elements in a sorted order.

An agent deployment request is received for an external computing environment. An agentic module is configured for deployment to the external computing environment based on characteristics of the external computing environment, wherein the agentic module comprises a computational agent operable to process information stored within the external computing environment and a corresponding agentic controller. Responsive to the agent deployment request, a first unit of software instructions is generated that, when executed, instantiates an agentic module instance within the external computing environment. The first unit of software instructions is transmitted to the external computing environment. A performance metric related to processing of the information stored within the external computing environment by a computational agent instance of the agentic module instance is received from an agentic controller instance of the agentic module instance.

A system and method include receiving real-time streaming data, detecting a change in a distribution of the real-time streaming data by defining a reference window, defining a current window, computing a first weighted cumulative distribution function for the reference window based on a first weight value, computing a second weighted cumulative distribution function for the current window based on a second weight value, computing a maximum difference between the first weighted cumulative distribution function and the second weighted cumulative distribution function, computing a threshold value, and determining that the maximum difference is greater than the threshold value.

A system, method, and computer-program product includes receiving a plurality of vector embeddings having an initial dimensionality and projecting the plurality of vector embeddings into lower-dimensional spaces using at least two different dimension reduction algorithms to generate corresponding sets of projected vector embeddings. Each set of projected embeddings may be quantized and nearest neighbors for the original embeddings and for each quantized set of projected embeddings may be calculated. Additionally, a neighbor preservation metric may be evaluated for each quantized set by comparing its nearest neighbors to those of the original embeddings. Based on the neighbor preservation metrics and a predefined error tolerance, an optimal compression configuration may be selected.

A system and method to detect and reduce noise include receiving a request and deploying a linear mixed model in a plurality of iterations to determine, based on one or more parameters of the linear mixed model, that noise is present in an output of the linear mixed model, determine that the noise is not larger than a threshold and that the output is capable of further improvement, adjust at least some of the one or more parameters to obtain one or more adjusted parameters, adjust the output of the linear mixed model based on the one or more adjusted parameters to reduce the noise in the output, determine whether a stopping criterion is reached, and responsive to determining that the stopping criterion has not reached, repeat next iteration with the adjusted output or responsive to determining that the stopping criterion has reached, transmit the output.

A system, method, and computer-program product includes receiving, from a third-party identity management system, an authentication response indicating a set of login attributes; obtaining, from an identity resolution service, a set of identity and authorization attributes using the set of login attributes, the set of identity and authorization attributes including a unique user identifier (UID); granting, by the identity resolution service, a session initiation token when the set of identity and authorization attributes satisfy predefined authorization criteria; in response to the identity resolution service granting the session initiation token: allocating a compute session and a persistent storage resource to the unique UID; executing, via the compute session, an operation that modifies files stored in the persistent storage resource; and transmitting, to a version control system, a version control operation that records the files modified in the persistent storage resource to a code repository using the unique UID or group identifiers.

A computing system generates a feature for a machine learning model by receiving identification of character data. The character data is stored in a database comprising multiple documents. One or more documents of the multiple documents pertain to a respective entity of multiple entities. The computing system generates structured data from the character data by generating multiple structured data terms; and storing the multiple structured data terms in respective data elements of multiple memory blocks in memory of the computing system. The computing system generates the feature for the machine learning model based on one or more of: at least two structured data terms in different memory blocks of the multiple memory blocks; and data manipulation of a structured data term in a memory block of the multiple memory blocks.

A system, method, and computer-program product include receiving, via an API server, an upload request that includes a logical table reference parameter and a structured data file. A table metadata schema corresponding to the logical table reference parameter is retrieved from a metadata repository. A plurality of data entries of the structured data file are converted into schema-conforming data records using column metadata definitions extracted from the schema. The conversion includes identifying source column headers matching source column identifiers, parsing associated cell value sets, validating the cell value sets based on criteria defined in the metadata definitions, and associating validated values with database columns. The schema-conforming data records are persisted to a database table designated in the table metadata schema.

A system, method, and computer-program product includes obtaining a decisioning dataset comprising a plurality of favorable decisioning records and at least one unfavorable decisioning record; detecting, via a machine learning algorithm, a favorable decisioning record of the plurality of favorable decisioning records that has a vector value closest to a vector value of the unfavorable decisioning record; executing a counterfactual assessment between the favorable decisioning record and the unfavorable decisioning record; generating an explainability artifact based on one or more bias intensity metrics to explain a bias in a machine learning-based decisioning model; and in response to generating the explainability artifact, displaying the explainability artifact in a user interface.