Abstract: One or more computer processors discover an anomalous subset through sparsity-based automatic feature selection.

Abstract: Described are techniques for A/B testing including a computer-implemented method of identifying, in an A/B testing database, a set of feature values with a statistically significant difference in A/B testing outcomes above a threshold. The method further includes partitioning records in the A/B testing database into a plurality of population strata according to the set of feature values. The method further includes performing A/B testing, and identifying heterogeneous outcomes of the A/B testing for respective strata of the plurality of population strata.

Abstract: A computer-implemented method for differentiating patterns of care (DPoC) to detect anomalous subsets in any given population with a defined set of outcomes and features. The computer-implemented method includes detecting the anomalous subsets, ranking the anomalous subsets based on a score of each anomalous subset that is reflective of an anomaly thereof, specifying whether each of the anomalous subsets overlaps with another one of the anomalous subsets, whether each of the anomalous subsets is unique and whether each of the anomalous subsets is conditional and specifying as to whether the detecting of each of the anomalous subsets has a higher or lower outcome than expected.

Abstract: A training process a predictive model uses a dataset of features and an outcome. The method generates a table for a dataset comprising multiple features, the table contains values for each pair of features in the dataset, randomly selects features from the dataset, thereby creating a first subset of features, operates a propensity score matching using the randomly selected features to identify cases and controls using the outcome variable, rewards one or more features of a second subset of features in the multiple features that were not selected randomly, each feature of the second subset addresses a statistical significance criteria, updating each entry in the table with a reward distance between each pair of features, calculates a cumulative reward measure, iterating the steps until convergence, selects a final subset of features when a variability criteria of the cumulative reward measure addresses convergence criteria, and trains the predictive model.

Abstract: Context-dependent memory stimulation is provided. A context-dependent task resumption memory cue personalized to a user is sent to a client device of the user via a network during a task interruption lag period corresponding to a primary task prior to the user starting a set of secondary tasks to support a context-dependent memory of the user associated with the primary task. It is determined that the user started and then stopped performing the set of secondary tasks based on monitoring the client device. The context-dependent task resumption memory cue personalized to the user is resent to the client device via the network during a task resumption lag period corresponding to the primary task to stimulate the context-dependent memory of the user associated with the primary task in response to the computer determining that the user stopped performing the set of secondary tasks and resumed the primary task.

Abstract: A plurality of key features that contribute to a level of anomalousness of an anomalous subgroup are identified. One or more minimal perturbations to a set of features of the anomalous subgroup that result in a reduction of the level of anomalousness are identified. The application of one or more minimal perturbations to members of the anomalous subgroup is facilitated.

Abstract: An embodiment associates a virtual-meeting host designation with a first user. During the virtual meeting, the embodiment generates meeting context data that includes user characteristic data and meeting characteristic data. The embodiment uses the meeting context data to generate a predicted end time for the virtual meeting and predicts a meeting run-over condition indicative of the virtual meeting continuing past a scheduled end time by comparing the predicted end time and a scheduled end time. Responsive to predicting the meeting run-over condition, the embodiment uses the user characteristic data to generate a ranked list of other users and renders, as a display to the first user, the ranked list and an indication of the meeting run-over condition. The embodiment receives a selection of a second user as a new host of the virtual meeting and, responsive to receiving the selection, associates the host designation with the second.

Abstract: One or more computer processors discover an anomalous subset through sparsity-based automatic feature selection.

Abstract: Methods, systems, and computer program products for predicting states from patient-caregiver dyadic biomarker data using artificial intelligence are provided herein. A computer-implemented method includes obtaining biomarker data derived from one or more dyads, each dyad comprising at least one patient and at least one caregiver associated with the at least one patient; determining, based on processing the obtained biomarker data, data-based representations of mental distress and/or social rhythm disruption among at least one of the one or more dyads; predicting mental distress and/or social rhythm disruption among a given dyad of at least one patient and at least one caregiver associated with the at least one patient by processing input biomarker data, derived from the given dyad, using artificial intelligence techniques in connection with at least a portion of the data-based representations; and performing automated actions based on the predicting.

Abstract: Described are techniques for A/B testing including a computer-implemented method of identifying, in an A/B testing database, a set of feature values with a statistically significant difference in A/B testing outcomes above a threshold. The method further includes partitioning records in the A/B testing database into a plurality of population strata according to the set of feature values. The method further includes performing A/B testing, and identifying heterogeneous outcomes of the A/B testing for respective strata of the plurality of population strata.

Abstract: A computer system may create a language model corpus including multilingual alignment for training a combined language model and train (or pre-train) the combined language model. The computer system may create an adverse medication reaction corpus to include adverse medication reaction utterances and label an N-gram of an utterance in the adverse medication reaction utterances as a response to query, for multiple N-grams. The computer system may generate a code-mixed utterance model to perform code-mixed utterances in a turn by turn dialogue, by at least adding additional output layer including at least a start vector, language vector, and a query vector including at least the labeled N-gram, which are additional to the combined language model's predicted next words.

Abstract: A comfortability-tension threshold for a user is determined and a current pose of the user and a determined optimal pose are compared. The current pose and the determined optimal pose are mapped to an angle for adjusting on-screen content of a display of the computer system and the on-screen content of the display of the computer system is adjusted based on the angle.

Abstract: A translation-engine suggestion method, system, and computer program product include identifying probes for third-party translation-engines for an input text, segmenting sections of the input text into a plurality of segments according to the identified probes, fragmenting the input text into fragments according to the segments, applying each fragment to the identified probe using the corresponding third-party translation-engine, and outputting a translation by combining each fragment.

Abstract: A computer system may create a language model corpus including multilingual alignment for training a combined language model and train (or pre-train) the combined language model. The computer system may create an adverse medication reaction corpus to include adverse medication reaction utterances and label an N-gram of an utterance in the adverse medication reaction utterances as a response to query, for multiple N-grams. The computer system may generate a code-mixed utterance model to perform code-mixed utterances in a turn by turn dialogue, by at least adding additional output layer including at least a start vector, language vector, and a query vector including at least the labeled N-gram, which are additional to the combined language model's predicted next words.

Abstract: A comfortability-tension threshold for a user is determined and a current pose of the user and a determined optimal pose are compared. The current pose and the determined optimal pose are mapped to an angle for adjusting on-screen content of a display of the computer system and the on-screen content of the display of the computer system is adjusted based on the angle.

Abstract: A bias detection method, system, and computer program product include creating a context of an applicant based on a profile of the applicant and a context of a reviewer based on a profile of the reviewer, predicting a probability of overlapping data points between the applicant and the reviewer, building enriched embeddings for a deep learning model based on the context of the applicant, the context of the reviewer, the overlapping data points, and text from a review and a final decision by the reviewer, and calculating a bias score via a deep learning model run over the enriched embeddings.

Abstract: A translation-engine suggestion method, system, and computer program product include identifying probes for third-party translation-engines for an input text, segmenting sections of the input text into a plurality of segments according to the identified probes, fragmenting the input text into fragments according to the segments, applying each fragment to the identified probe using the corresponding third-party translation-engine, and outputting a translation by combining each fragment.

Abstract: Identify an emergency patient's potential emergency contacts and their social relationships by processing the patient's Internet footprint with a first natural language processing algorithm. Identify the emergency patient's possible health context by processing at least one of the emergency patient's Internet footprint and diagnostic data with a second natural language processing algorithm. Link at least one item of the emergency patient's possible health context to at least one linked emergency contact, who is selected from the potential emergency contacts using a sorting module that is trained to associate types of medical conditions with attributes of emergency contacts. Verify the at least one item of the emergency patient's possible health context by contacting the at least one linked emergency contact. Aggregate a verified health profile for the emergency patient based on at least one response provided by the at least one linked emergency contact.

Abstract: Methods and systems for monitoring an ongoing procedure are described. In an example, a processor can receive a first set of data indicating a current context of a procedure being performed by a first entity on a second entity. The processor can determine a first score based on the first set of data, where the first score can indicate an effectiveness of the procedure. The processor can identify a set of requirements based on the first score. The set of requirements can be associated with a second score indicating a target effectiveness of the procedure. The processor can generate a recommendation for the first entity to perform the set of requirements to achieve the second score.

Abstract: An apparatus obtains participant sentiment data including at least one of: a text conversation between a participant and an investigator in a clinical trial; a video conversation between the participant and the investigator; and the participant's social network information; extracts and normalizes a sentiment score from the participant sentiment data; generates a compliance score for the participant by using a trained regressor on at least the sentiment score; compares the compliance score to a lower threshold and to a higher threshold; selects an action from a decision tree in response to the compliance score; and facilitates the action.