Abstract: Methods and systems to provide a form of probabilistic labeling to associate an outage with a disturbance, which could itself be either known based on the available data or unknown. In the latter case, labeling is especially challenging, as it necessitates the discovery of the disturbance. One approach incorporates a statistical change-point analysis to time-series events that correspond to service tickets in the relevant geographic sub-regions. The method is calibrated to separate the regular periods from the environmental disturbance periods, under the assumption that disturbances significantly increase the rate of loss-causing events. To obtain the probability that a given loss-causing event is related to an environmental disturbance, the method leverages the difference between the rate of events expected in the absence of any disturbances (baseline) and the rate of actually observed events. In the analysis, the local disturbances are identified and estimators of their duration and magnitude are provided.

Abstract: An example operation may include one or more of executing a first processing element among a sequence of processing elements within a data processing pipeline on input data to generate a first output, transferring the first output and context associated with the first processing element from the first processing element to a second processing element among the sequence of processing elements, within the data processing pipeline, executing the second processing element on the first output and the context to generate a second output, and storing the second output in memory.

Abstract: Disclosed embodiments provide techniques for estimating imputation algorithm performance. Multiple imputer algorithms are selected, and an evaluation of how well each of the imputer algorithms can estimate the missing data is performed. Disclosed embodiments obtain an imputer candidate dataset (ICD). The imputer candidate dataset is compared to the incomplete data range, and a similarity metric is determined between the data range and the ICD. When the similarity metric exceeds a predetermined threshold, an imputer evaluation dataset (IED) is created from the ICD by removing one or more data points from the ICD. Each imputer algorithm is evaluated by applying the IED to it, and computing an imputer evaluation metric based on its performance. The multiple imputer algorithms are ranked based on the imputer evaluation metric. The best ranked imputer algorithm can then be selected for use on the incomplete data range within the measurement dataset.

Abstract: An approach for generating an artificial intelligence system configurable for use with assets. In this approach, a model recipe is selected for generating the artificial intelligence system for use with assets. Recipe parameters specified in the model recipe are identified. A training dataset is created using the model recipe and input data. A set of artificial intelligence models is trained using the training dataset, the recipe parameters, and the model recipe. The training creates artifact models. The artifact models resulting from training are evaluated. The evaluation is used to select a set of the artifact models in the artifacts that form the artificial intelligence system that is configurable for use in assets.

Abstract: A computer implemented federated learning method of predicting failure of assets includes generating a local model at a local site for each of the cohorts and training the local model on local data for each of the cohorts for each failure type. The local model is shared with a central database. A global model is created based on an aggregation of a plurality of the local models from a plurality of the local sites. At each of the plurality of local sites, one of the global model and the local model is chosen for each of the cohorts. The chosen model operates on local data to predict failure of the assets. The utilized features include partitioning features of the assets into static features, semi-static features, and dynamic features, and forming cohorts of the assets based on the static features and the semi-static features.

Abstract: Machine learning (ML) pipeline selection includes performing cross-validation runs for dataset-pipeline combinations and building a matrix of first accuracy scores, factoring the matrix of accuracy scores into pipeline latent factors and dataset latent factors, augmenting the matrix of accuracy scores by selecting a subset of ML pipelines of a plurality of ML pipelines, then, for a new dataset, running the subset of ML pipelines with the new dataset to build and test respective ML models, obtain second accuracy scores, and augment the matrix of accuracy scores with the second accuracy scores to produce an augmented matrix of accuracy scores, factoring the augmented matrix of accuracy scores into refined pipeline latent factors and refined dataset latent factors, and identifying, based on the refined pipeline latent factors and the refined dataset latent factors, ML pipeline(s), of the plurality of ML pipelines, as most optimal for model building based on the new dataset.

Abstract: A method, system, and computer program product that is configured to: receive an input time series from an external device in a first system, divide the input time series to a set of univariate time subseries, transform the set of univariate time subseries into a univariate prediction result series using a transformer model, concatenate the univariate prediction result series to a multivariate predictive result, and output the multivariate predictive result for providing time series forecasting to a second system.

Abstract: Using exported data of a machine learning model and a model training environment specification, a resource usage specification and a code module usage specification of the model are identified. A code module installation specification is determined from a code module requirements specification and a target execution environment specification. The code modules specified by the code module installation specification are caused to be installed in the target execution environment. Using data of the updated target execution environment, the updated target execution environment is validated for execution of the model. Execution of the model in the updated target execution environment is simulated. The model is deployed in the updated target execution environment responsive to the simulating being successful.

Abstract: A computer-implemented method for developing a hierarchical machine-learning pipeline can include receiving a hierarchy specification, a set of estimators for the root node, and one or more transformer options for each of the transformer nodes. The hierarchy specification provides a configuration of the root node, transformer nodes, and edges interconnecting the root and transformer nodes. A rank can be obtained for each estimator in the root node. A hierarchy pipeline traverser can then traverse a first child layer of the transformer nodes connected to the root node via one of the edges. A first ranked list of pathways can be determined with respect to the one or more transformer options selected for the first child layer and at least one selected estimator of the root node.

Abstract: A computer implemented method of preparing process data for use in an artificial intelligence (AI) model includes collecting and storing raw data as episodic data for each episode of a process. An episode data generator assigns an episode identifier each set of episodic data. The raw data per episode is transformed into a standardized episodic data format that is usable by the AI model. Metrics are assigned to the episodic data and the episodic data is aggregated in an episode store. The data in the episode store is used by a feature extraction and learning module to extract and rank features.

Abstract: Techniques for generating model ensembles are provided. A plurality of models trained to generate predictions at each of a plurality of intervals is received. A respective prediction accuracy of each respective model of the plurality of models is determined for a first interval of the plurality of intervals by processing labeled evaluation data using the respective model. Additionally, a model ensemble specifying one or more of the plurality of models for each of the plurality of intervals is generated, comprising selecting, for the first interval, a first model of the plurality of models based on (i) the respective prediction accuracies and (ii) at least one non-error metric.

Abstract: A method includes obtaining, by a processor set, labeled training data associated with a system; identifying, by the processor set, a first region and a second region in the labeled training data, wherein the first region is associated with a failure of the system and the second region is exclusive of the first region; and creating, by the processor set, re-labeled training data by altering one or more labels of the labeled training data in the first region based on data in the second region.

Abstract: A computing device, method, and system are provided of improving data quality to conserve computational resources. The computing device receives a raw dataset. One or more data quality metric goals corresponding to the received raw dataset are received. A schema of the dataset is determined. An initial set of validation nodes is identified based on the schema of the dataset. The initial set of validation nodes are executed. A next set of validation nodes are iteratively expanded and executed based on the schema of the dataset until a termination criterion is reached. A corrected dataset of the raw dataset is provided based on the iterative execution of the initial and next set of validation nodes.

Abstract: An embodiment for identifying influential disturbances is provided. The embodiment may automatically receive a set of service records including disturbance-related probability values corresponding to the disturbance-revealing events, and wherein one or more service records are mislabeled or have no label relating to an associated disturbance. The embodiment may generate baselines for a series of relevant sub-regions associated with the service records, and normalize daily summaries of disturbance probabilities for each of the relevant sub-regions. The embodiment may automatically identify subsets of service records corresponding to a series of newly-discovered disturbances by using the disturbance-related probability values and a series of associated features to identify deviations from normal non-disturbance event distributions.

Abstract: Systems and methods for optimizing and training machine learning (ML) models are provided. In embodiments, a computer implemented method includes: performing, by a processor set, a group execution of ML pipelines using a first subset of a training data set as input data for the ML pipelines, thereby generating a trained ML model for each of the ML pipelines, wherein data transformations that are common between the ML pipelines are implemented only once to generate an output, and the output is shared between the ML pipelines during the group execution of the ML pipelines; generating, by the processor set, performance metrics for each of the trained ML models based on validation data; ranking, by the processor set, the trained ML models based on the performance metrics, thereby generating a list of ranked ML models; and outputting, by the processor set, the list of ranked ML models to a user.

Abstract: Disclosed embodiments provide techniques for estimating imputation algorithm performance. Multiple imputer algorithms are selected, and an evaluation of how well each of the imputer algorithms can estimate the missing data is performed. Disclosed embodiments obtain an imputer candidate dataset (ICD). The imputer candidate dataset is compared to the incomplete data range, and a similarity metric is determined between the data range and the ICD. When the similarity metric exceeds a predetermined threshold, an imputer evaluation dataset (IED) is created from the ICD by removing one or more data points from the ICD. Each imputer algorithm is evaluated by applying the IED to it, and computing an imputer evaluation metric based on its performance. The multiple imputer algorithms are ranked based on the imputer evaluation metric. The best ranked imputer algorithm can then be selected for use on the incomplete data range within the measurement dataset.

Abstract: To automate time series forecasting machine learning pipeline generation, a data allocation size of time series data may be determined based on one or more characteristics of a time series data set. The time series data may be allocated for use by candidate machine learning pipelines based on the data allocation size. Features for the time series data may be determined and cached by the candidate machine learning pipelines. Predictions of each of the candidate machine learning pipelines using at least the one or more features may be evaluated. A ranked list of machine learning pipelines may be automatically generated from the candidate machine learning pipelines for time series forecasting based upon evaluating predictions of each of the one or more candidate machine learning pipelines.

Abstract: Using exported data of a machine learning model and a model training environment specification, a resource usage specification and a code module usage specification of the model are identified. A code module installation specification is determined from a code module requirements specification and a target execution environment specification. The code modules specified by the code module installation specification are caused to be installed in the target execution environment. Using data of the updated target execution environment, the updated target execution environment is validated for execution of the model. Execution of the model in the updated target execution environment is simulated. The model is deployed in the updated target execution environment responsive to the simulating being successful.

Abstract: Selecting a time-series forecasting pipeline by receiving target variable time-series data and exogenous variable time-series data, generating a regular forecasting pipeline comprising a model according to the target variable time-series data, generating an exogenous forecasting pipeline comprising a model according to the target variable time-series data and the exogenous variable time-series data, evaluating the regular forecasting pipeline and the exogenous forecasting pipeline, selecting a pipeline according to the evaluation, and providing the selected pipeline.

Abstract: A system and method for learning a predictive function that can automatically learn different operating modes for a multi-modal system and predict the number of operating states for a multi-modal system and additionally the detailed structure for each state. Once learned, the predictive function (model) can be used to determine a mode of a new sample (an asset). Based on the determined components that maximize a log likelihood function, a mode of the new sample is detected into the model via dependency graphs. One aspect includes enforcing a lower bound for the number of sample points to form an operational mode for an asset. While a mode relates to sample points which maximizes like log-likelihood, an ability is provided to remove artifact modes due to noisy data by considering a sufficient sample data condition and maximizing log-likelihood. Domain knowledge can be incorporated into the model via dependency graphs.