Sas Institute Inc.

David Bruce Elsheimer

Michael James Leonard

Xilong Chen

Gunce Eryuruk Walton

Ming-Chun Chang

Sylvie Tchumtchoua Kabisa

Edward Tilden Blair

Jennifer Leigh Sloan Beeman

Thiago Santos Quirino

Javier Delgado

Yang Zhao

Wanxi Gu

Dillon Frame

Chuan Xu

Douglas Allan Cairns

Joshua David Griffin

Jan Chvosta

Stacey Michelle Christian

Michele Angelo Trovero

Keith Eugene Crowe

Steven Joseph Gardner

Yan Gao

Wilma Stainback Jackson

Connie Stout Dunbar

Michael Thomas Lamm

Peter Dillman

Gregory Scott Dunbar

Yuelei Sui

A system and method include generating a topological order of a DAG by creating residual series vectors, calculating normality statistic and MSE values for the residual series vectors, comparing the normality statistic values with a critical value, for each normality statistic value that is less than or equal to the critical value, adding a variable index to a temporary order list and the MSE value to an MSE list, counting a number of elements in the temporary order list, if the number of elements in the temporary order list is zero, updating an order list based on the normality statistic values or if the number of elements in the temporary order list is not zero, updating the order list based on at least one of the temporary order list or the MSE list, and outputting the order list as the topological order of the DAG.

Topological order determination in causal graphs

A point estimate value for an individual is computed using a Bayesian neural network model (BNN) by training a first BNN model that computes a weight mean value, a weight standard deviation value, a bias mean value, and a bias standard deviation value for each neuron of a plurality of neurons using observations. A plurality of BNN models is instantiated using the first BNN model. Instantiating each BNN model of the plurality of BNN models includes computing, for each neuron, a weight value using the weight mean value, the weight standard deviation value, and a weight random draw and a bias value using the bias mean value, the bias standard deviation value, and a bias random draw. Each instantiated BNN model is executed with the observations to compute a statistical parameter value for each observation vector of the observations. The point estimate value is computed from the statistical parameter value.

Bayesian neural network point estimator

A treatment model trained to compute an estimated treatment variable value for each observation vector of a plurality of observation vectors is executed. Each observation vector includes covariate variable values, a treatment variable value, and an outcome variable value. An outcome model trained to compute an estimated outcome value for each observation vector using the treatment variable value for each observation vector is executed. A standard error value associated with the outcome model is computed using a first variance value computed using the treatment variable value of the plurality of observation vectors, using a second variance value computed using the treatment variable value and the estimated treatment variable value of the plurality of observation vectors, and using a third variance value computed using the estimated outcome value of the plurality of observation vectors. The standard error value is output.

Standard error for deep learning model outcome estimator

A computing device selects new test configurations for testing software. (A) First test configurations are generated using a random seed value. (B) Software under test is executed with the first test configurations to generate a test result for each. (C) Second test configurations are generated from the first test configurations and the test results generated for each. (D) The software under test is executed with the second test configurations to generate the test result for each. (E) When a restart is triggered based on a distance metric value computed between the second test configurations, a next random seed value is selected as the random seed value and (A) through (E) are repeated. (F) When the restart is not triggered, (C) through (F) are repeated until a stop criterion is satisfied. (G) When the stop criterion is satisfied, the test result is output for each test configuration.

Automated machine learning test system

One example described herein involves a system receiving task data and distribution criteria for a state space model from a client device. The task data can indicate a type of sequential Monte Carlo (SMC) task to be implemented. The distribution criteria can include an initial distribution, a transition distribution, and a measurement distribution for the state space model. The system can generate a set of program functions based on the task data and the distribution criteria. The system can then execute an SMC module to generate a distribution and a corresponding summary, where the SMC module is configured to call the set of program functions during execution of an SMC process and apply the results returned from the set of program functions in one or more subsequent steps of the SMC process. The system can then transmit an electronic communication to the client device indicating the distribution and its corresponding summary.

Flexible program functions usable for customizing execution of a sequential Monte Carlo process in relation to a state space model

A treatment model that is a first neural network is trained to optimize a treatment loss function based on a treatment variable t using a plurality of observation vectors by regressing t on x,z. The trained treatment model is executed to compute an estimated treatment variable value {circumflex over (t)}for each observation vector. An outcome model that is a second neural network is trained to optimize an outcome loss function by regressing y on xand an estimated treatment variable t. The trained outcome model is executed to compute an estimated first unknown function value {circumflex over (&#x3b1;)}(x) and an estimated second unknown function value {circumflex over (&#x3b2;)}(x) for each observation vector. An influence function value is computed for a parameter of interest using {circumflex over (&#x3b1;)}(x) and {circumflex over (&#x3b2;)}(x). A value is computed for the predefined parameter of interest using the computed influence function value.

Causal inference and policy optimization system based on deep learning models

Machines can be controlled using advanced control systems that implement an automated version of singular spectrum analysis (SSA). For example, a control system can perform SSA on a time series having one or more time-dependent variables by: generating a trajectory matrix from the time series, performing singular value decomposition on the trajectory matrix to determine elementary matrices; and categorizing the elementary matrices into groups. The elementary matrices can be automatically categorized into the groups by: generating one or more w-correlation matrices based on spectral components associated with the time series, determining w-correlation values based on the one or more w-correlation matrices; categorizing the w-correlation values into a predefined number of w-correlation sets, and forming the groups based on the predefined number of w-correlation sets. The control system can then generate a predictive forecast using the groups and control operation of a machine using the predictive forecast.

Advanced control systems for machines

In some examples, computing devices can partition timestamped data into groups. The computing devices can then distribute the timestamped data based on the groups. The computing devices can also obtain copies of a script configured to process the timestamped data, such that each computing device receives a copy of the script. The computing devices can determine one or more code segments associated with the groups based on content of the script. The one or more code segments can be in one or more programming languages that are different than a programming language of the script. The computing devices can then run the copies of the script to process the timestamped data within the groups. This may involve interacting with one or more job servers configured to run the one or more code segments associated with the groups.

Scalable cloud-based time series analysis

Timestamped data can be read in parallel by multiple grid-computing devices. The timestamped data, which can be partitioned into groups based on time series criteria, can be deterministically distributed across the multiple grid-computing devices based on the time series criteria. Each grid-computing device can sort and accumulate the timestamped data into a time series for each group it receives and then process the resultant time series based on a previously distributed script, which can be compiled at each grid-computing device, to generate output data. The grid-computing devices can write their output data in parallel. As a result, vast amounts of timestamped data can be easily analyzed across an easily expandable number of grid-computing devices with reduced computational expense.

Machines can be controlled using advanced control systems. Such control systems may use an automated version of singular spectrum analysis to control a machine. For example, a control system can perform singular spectrum analysis on a time series by: generating a trajectory matrix from the time series, performing singular value decomposition on the trajectory matrix to determine elementary matrices and corresponding eigenvalues, and automatically categorizing the elementary matrices into groups. The elementary matrices can be automatically categorized into the groups by: generating a matrix of w-correlation values based on the eigenvalues, categorizing the w-correlation values into a predefined number of w-correlation sets, and forming the groups based on the predefined number of w-correlation sets. The control system can then determine component time-series based on the groups, and generate a predictive forecast using the component time-series. The control system can use the predictive forecast to control operation of the machine.

Systems and methods are included for adjusting a set of predicted future data points for a time series data set including a receiver for receiving a time series data set. One or more processors and one or more non-transitory computer readable storage mediums containing instructions may be utilized. A count series forecasting engine, utilizing the one or more processors, generates a set of counts corresponding to discrete values of the time series data set. An optimal discrete probability distribution for the set of counts is selected. A set of parameters are generated for the optimal discrete probability distribution. A statistical model is selected to generate a set of predicted future data points. The set of predicted future data points are adjusted using the generated set of parameters for the optimal discrete probability distribution in order to provide greater accuracy with respect to predictions of future data points.

Systems and methods for time series analysis techniques utilizing count data sets

Systems and methods are provided for analyzing through one-pass of unstructured time stamped data of a physical process. A distribution of time-stamped unstructured data is analyzed to identify a plurality of potential hierarchical structures for the unstructured data. A hierarchical analysis of the potential hierarchical structures is performed to determine an optimal frequency and a data sufficiency metric for the potential hierarchical structures. One of the potential hierarchical structures is selected as a selected hierarchical structure based on the data sufficiency metrics. The unstructured data is structured according to the selected hierarchical structure and the optimal frequency associated with the selected hierarchical structure, where said structuring of the unstructured data is performed via a single pass though the unstructured data. The identified statistical analysis of the physical process is performed using the structured data.

Computer-implemented systems and methods for efficient structuring of time series data

Systems and methods for forecasting ratios in hierarchies are provided. Hierarchies can be formed that have components, including a numerator time series with values from input data, a denominator time series with values from input data, and a ratio time series of the numerator time series over the denominator time series. The components can be modeled to generate forecasted hierarchies. The forecasted hierarchies can be reconciled so that the forecasted hierarchies are statistically consistent throughout nodes of the forecasted hierarchies.

Growing community of inventors

Clayton, NC, United States of America

David Bruce Elsheimer