Abstract: Embodiments of the subject matter facilitate prediction on a directed graph comprising nodes and edges. In one embodiment of the subject matter, prediction is over the entire graph. In another embodiment of the subject matter, prediction is for each node in the graph. The prediction can be categorical or a vector of real values (i.e., multivariate). Embodiments of the subject matter are not sensitive to the order in which the graph's nodes and edges are presented during training or prediction: they are invariant to rotations, translations, reflections, and permutations of the graph. Embodiments of the subject matter facilitate prediction on both directed and undirected graphs.

Abstract: A computer-implemented method includes receiving a group of sets. Each set has values for immutable attributes that match values for at least one mutable attribute in a prediction request. The method includes generating a deviation model based on the group of sets. The method includes generating multiple sets of likely mutable attribute values using the deviation model. The method includes automatically selecting a neural network based on at least one of the likely mutable attribute values. The neural network includes a set of layers. Each layer includes a set of nodes. A first layer receives inputs at the set of nodes of the input layer. A last layer outputs, from the neural network, outputs modified by a preceding layer. The method includes generating a likelihood corresponding to the at least one mutable attribute in the prediction request using the neural network.

Abstract: A method includes, in response to receiving a forecasting request from a user device via a web portal, determining a measure of uncertainty associated with an input vector specified by the forecasting request. The measure of uncertainty includes a compact representation of uncertainty associated with future time intervals specified by the forecasting request. The method includes obtaining a set of historical data from a time-series data store. The method includes generating a forecast model using the set of historical data, the input vector, and the measure of uncertainty to predict outcomes incrementally for the future time intervals. The method includes determining a predicted outcome using the forecast model at an end of the future time intervals. The method includes, in response to the predicted outcome exceeding a threshold, generating a graphical user interface. The graphical user interface illustrates the mean vector and a magnitude of the predicted uncertainty measure.

Abstract: A confidence interval determination system includes a processor and memory. The memory stores a historical data store and instructions. The instructions include, in response to receiving a prediction request, determining a size of each element of the prediction request and obtaining a previous prediction request. The instructions include, in response to the prediction request being different from the previous prediction request, for each element of the prediction request, generating a classifier based on the training data and, for each bin of a set of bins of the element, determining a set of predicted likelihoods using the classifier. The instructions include generating a mean vector prediction and a covariance matrix using the sets of bins generated, determining a predicted outcome and a confidence interval using the mean vector prediction and the covariance matrix, and, in response to the confidence interval exceeding a threshold, outputting the predicted outcome and the confidence interval.

Abstract: A computer-implemented method includes receiving a group of sets. Each set has values for immutable attributes that match values for at least one mutable attribute in a prediction request. The method includes generating a deviation model based on the group of sets. The method includes generating multiple sets of likely mutable attribute values using the deviation model. The method includes automatically selecting a neural network based on at least one of the likely mutable attribute values. The neural network includes a set of layers. Each layer includes a set of nodes. A first layer receives inputs at the set of nodes of the input layer. A last layer outputs, from the neural network, outputs modified by a preceding layer. The method includes generating a likelihood corresponding to the at least one mutable attribute in the prediction request using the neural network.

Abstract: A computer-implemented method includes selecting a group of sets. Each set has values for immutable attributes that match values for at least one mutable attribute in a prediction request. The method includes determining a conditional covariance matrix for the group of sets. The method includes generating a deviation model based on the conditional covariance matrix. The method includes sampling the deviation model to generate multiple sets of likely mutable attribute values. The method includes automatically selecting a neural network from a set of outcome models based on the likely mutable attribute values. Each neural network includes a set of layers. Each layer includes a set of nodes. A first layer receives inputs at the set of nodes of the input layer. Each layer other than the first layer receives outputs from a preceding layer and creates modified outputs. A last layer outputs the modified outputs from the neural network.

Abstract: A method includes, in response to receiving a forecasting request from a user device via a web portal, determining a measure of uncertainty associated with an input vector specified by the forecasting request. The measure of uncertainty includes a compact representation of uncertainty associated with future time intervals specified by the forecasting request. The method includes obtaining a set of historical data from a time-series data store. The method includes generating a forecast model using the set of historical data, the input vector, and the measure of uncertainty to predict outcomes incrementally for the future time intervals. The method includes determining a predicted outcome using the forecast model at an end of the future time intervals. The method includes, in response to the predicted outcome exceeding a threshold, generating a graphical user interface. The graphical user interface illustrates the mean vector and a magnitude of the predicted uncertainty measure.

Abstract: A forecasting system includes a processor and memory. The memory stores a time-series data store and instructions. The instructions include, in response to receiving a forecasting request from a user device for an entity including input data, determining a measure of uncertainty associated with the input data. The input data includes a proposed action and a forecast period. The instructions include obtaining a set of historical data from the time-series data store associated with the proposed action and generating a forecast model using the set of historical data, the input data, and the measure of uncertainty to predict outcomes incrementally during the forecast period. The instructions include determining a predicted outcome using the forecast model for the entity at an end of the forecast period and, in response to the predicted outcome exceeding a threshold, generating a communication indicating the proposed action and transmitting the communication to the user device.

Abstract: During operation, embodiments of the subject matter can perform graph classification. One embodiment of the subject matter can facilitate graph classification by maintaining locality like a Hidden Markov Model (HMM), can handle confluences unlike an HMM, and can improve accuracy by including the class at every phase unlike an MPNN and in Deep Learning.

Abstract: A computer readable medium includes a data set with data stored in rows and N columns. Each of the rows is associated with one individual patient. Each of the N columns is associated with one type of data for patients. One or more processors is configured to: initialize missing values in M ones of the N columns with M values for the M ones of the N columns, respectively; generate M mathematical models for the M ones of the N columns having one or more missing values; for each of the rows having one or more missing values, update ones of the M values for the M ones of the N columns; and fill missing values in the M ones of the N columns with the M values, respectively.

Abstract: A computer readable medium includes a data set with data stored in rows and N columns. Each of the rows is associated with one individual patient. Each of the N columns is associated with one type of data for patients. One or more processors is configured to: initialize missing values in M ones of the N columns with M values for the M ones of the N columns, respectively; generate M mathematical models for the M ones of the N columns having one or more missing values; for each of the rows having one or more missing values, update ones of the M values for the M ones of the N columns; and fill missing values in the M ones of the N columns with the M values, respectively.

Abstract: A computer-implemented method includes selecting a group of sets. Each set has values for immutable attributes that match values for at least one mutable attribute in a prediction request. The method includes determining a conditional covariance matrix for the group of sets. The method includes generating a deviation model based on the conditional covariance matrix. The method includes sampling the deviation model to generate multiple sets of likely mutable attribute values. The method includes automatically selecting a neural network from a set of outcome models based on the likely mutable attribute values. Each neural network includes a set of layers. Each layer includes a set of nodes. A first layer receives inputs at the set of nodes of the input layer. Each layer other than the first layer receives outputs from a preceding layer and creates modified outputs. A last layer outputs the modified outputs from the neural network.

Abstract: A method includes maintaining sets of values for mutable and immutable attributes. Each outcome model of a set of outcome models generates a predicted likelihood of an outcome in response to at least one immutable attribute value and at least one mutable attribute value. A prediction request specifies a first outcome, values for at least one immutable attribute, and values for at least one mutable attribute. The method includes, in response, selecting a group of the sets, where each has values for the immutable attributes that match those of the prediction request. The method includes determining a conditional covariance matrix for the group of sets and then generating a deviation model. The method includes sampling the deviation model to generate sets of mutable attribute values. The method includes, for each of the sets of mutable attribute values, generating, using a selected outcome model, a likelihood of the first outcome occurring.