Abstract: Various systems and methods for processing activity data with a knowledge engine to generate actionable insights for a human subject are described. These actionable insights may include identifying a most likely action given a particular state of the human subject, identifying a most likely state in which the human subject performs a particular activity, or identifying anomalies in human activity patterns. In an example, an electronic processing system operates the knowledge engine with operations that: identify patterns of activity using clustering of events, identify meaningful patterns of activity from the patterns of activity based on co-occurrence of characteristics for respective events, rank the identified meaningful patterns of activity based on confidence and support of respective patterns to occur for a human subject, and generate a personalization action (such as an action for a software application) based on the ranked, identified meaningful patterns of activity.

Abstract: Methods, apparatus, and system to parse an unstructured, natural language input of a user, infer a semantic meaning of an intent of the user, and determine a present or future user state, including a location and a time, in which the intent can be fulfilled.

Abstract: System and techniques for search dimensionality expansion are described herein. A history of intelligent agent activity may be received. A search result generated by an external entity may be obtained that includes a set of geographic points of interest (POI). A geographic segment may be retrieved from a geographic segment library when the geographic segment contains a member of the set of POI. Here, the geographic segment defines a geographic area and a dimension set. The search result may be modified to create a modified search result that includes a member of the dimension set. The modified search result may then be transmitted to a user device.

Abstract: Methods, apparatus, and system to parse an unstructured, natural language input of a user, infer a semantic meaning of an intent of the user, and determine a present or future user state, including a location and a time, in which the intent can be fulfilled.

Abstract: Various techniques for performing contextual event scheduling with an event scheduling service are disclosed herein. In an example, data is processed at an event scheduling service, based on the use of a trained machine learning model that is specific to a user. This trained machine learning model is operated by the event scheduling service determine a proposed time and proposed scheduling parameters based on the contextual information, to identify a proposed event time and event scheduling parameters based on the model, the data indicating a user state, or external data. Further examples to evaluate user activity and identify schedule characteristics based on data inputs from a user's mobile computing device, wearable sensors, and external weather, traffic, or event data sources, are also disclosed.

Abstract: Methods, apparatus, and system to parse an unstructured, natural language input of a user, infer a semantic meaning of an intent of the user, and determine a present or future user state, including a location and a time, in which the intent can be fulfilled.

Abstract: In implementations, a divergence from a pattern of detected instances of an event is identified, where the event is of a routine of a user. The identified divergence corresponds to an out of routine event. Contextual information is generated corresponding to the identified out of routine event. Furthermore, one or more recommended actions for presenting content to the user are selected based on the contextual information. An indication of the identified out of routine event, the contextual information, and the one or more recommended actions are provided to a service, which may be an application on a user device. The service can present content to the user based on the identified out of routine event, the generated contextual information corresponding to the identified out of routine event, and the selected one or more recommended actions. The service may follow or disregard any of the selected one or more recommended actions.

Abstract: A method for inferring venue visits using semantic information includes receiving sensor data from sensors. An indication of a location is received that is associated with a user and determined based on the sensor data. A set of candidate venues associated with the location is determined based on the indication of the location. Sets of semantic information associated with the set of candidate venues are determined based on the sensor data. Candidate venues of the set are ranked by confidence that a given candidate venue corresponds to a visited venue of a venue visit based on the set of semantic information associated with the given candidate venue and additional semantic information associated with the user. A highest ranked candidate venue is selected as the visited venue and an indication is provided to a service causing content to be presented to the user based on the selected visited venue.

Abstract: Various systems and methods for processing activity data with a knowledge engine to generate actionable insights for a human subject are described. These actionable insights may include identifying a most likely action given a particular state of the human subject, identifying a most likely state in which the human subject performs a particular activity, or identifying anomalies in human activity patterns. In an example, an electronic processing system operates the knowledge engine with operations that: identify patterns of activity using clustering of events, identify meaningful patterns of activity from the patterns of activity based on co-occurrence of characteristics for respective events, rank the identified meaningful patterns of activity based on confidence and support of respective patterns to occur for a human subject, and generate a personalization action (such as an action for a software application) based on the ranked, identified meaningful patterns of activity.

Abstract: System and techniques for search dimensionality expansion are described herein. A history of intelligent agent activity may be received. A search result generated by an external entity may be obtained that includes a set of geographic points of interest (POI). A geographic segment may be retrieved from a geographic segment library when the geographic segment contains a member of the set of POI. Here, the geographic segment defines a geographic area and a dimension set. The search result may be modified to create a modified search result that includes a member of the dimension set. The modified search result may then be transmitted to a user device.

Abstract: An exercise schedule optimizer system includes a user interface to receive data input from an exerciser and provide data output to the exerciser. The system also includes an exercise goal generator to generate an exercise goal based on at least data regarding the exerciser, an exercise schedule proposer to propose one or more exercise schedule proposals based on at least a historical exercise pattern of the exerciser, and an optimal exercise scheduler to determine an optimal exercise schedule based on at least the one or more exercise schedule proposals.

Abstract: An event schedule optimizer system includes a user interface, an invitees extractor, a time optimizer, and a place optimizer. The user interface is to receive data input from an organizer and provide data output to the organizer. The invitees extractor is to generate an intended invitees list including intended invitees for a target event based on the data input from the organizer. The time optimizer is to determine an optimal time period for the target event based on a list of historical events of the organizer. The place optimizer is to determine an optimal place for the target event based on a set of proposed times and the list of historical events of the organizer.

Abstract: Various techniques for performing contextual event rescheduling with an event scheduling service are disclosed herein. In an example, data is processed at an event scheduling service, based on the use of a trained machine learning model that is specific to a user. This model is operated by the event scheduling service determine a contextual action option for rescheduling an electronic communication event at a proposed time with proposed scheduling parameters. The model may identify the proposed time and event scheduling parameters, from data indicating a user state, or external data, in addition to a semantic text option (such as “Call Back After Meeting”) corresponding to the proposed time and event scheduling parameters. Further examples to evaluate user activity and identify reschedule options based on data inputs from a user's mobile computing device, wearable sensors, and external weather, traffic, or event data sources, are also disclosed.

Abstract: Various techniques for performing contextual event scheduling with an event scheduling service are disclosed herein. In an example, data is processed at an event scheduling service, based on the use of a trained machine learning model that is specific to a user. This trained machine learning model is operated by the event scheduling service determine a proposed time and proposed scheduling parameters based on the contextual information, to identify a proposed event time and event scheduling parameters based on the model, the data indicating a user state, or external data. Further examples to evaluate user activity and identify schedule characteristics based on data inputs from a user's mobile computing device, wearable sensors, and external weather, traffic, or event data sources, are also disclosed.

Abstract: One embodiment relates to an apparatus, comprising logic, at least partially incorporated into hardware, to receive a plurality of notifications from each of a plurality of notification sources, determine a ranking score for each notification relative to any other notification received from the same notification source, and determine a notification class for each ranked notification relative to the ranked notifications from all notification sources based upon at least one blending attribute and at least one user contextual state.

Abstract: Apparatus, systems, and/or methods may provide control assistance. For example, a mobile device on a user may provide sensor data for the user. In addition, a routine may be determined based on the sensor data from the mobile device and/or sensor data from one or more indoor devices on premises. Moreover, an action of an indoor device may be defined based on the sensor data from the mobile device and/or the sensor data from the one or more indoor devices. The action may include a predicted action, which may be suggested to the user and/or automatically executed, via control data, by the indoor devices. Additionally, control assistance may be provided to a plurality of users.

Abstract: Systems, apparatuses and methods may provide for a mouth proximity detection system to be used with a device, such as a wearable device, to determine when to activate (and/or deactivate) a voice-activated circuit on the device. Embodiments may utilize three layers of analysis to make the determination, including a layer to analyze motion, a layer to detect a mouth of a user, and a layer to fuse the layers in a power saving arrangement to determine proximity of the device to the mouth of the user and to determine whether to activate a voice-activated circuit on the device.

Abstract: In implementations, a divergence from a pattern of detected instances of an event is identified, where the event is of a routine of a user. The identified divergence corresponds to an out of routine event. Contextual information is generated corresponding to the identified out of routine event. Furthermore, one or more recommended actions for presenting content to the user are selected based on the contextual information. An indication of the identified out of routine event, the contextual information, and the one or more recommended actions are provided to a service, which may be an application on a user device. The service can present content to the user based on the identified out of routine event, the generated contextual information corresponding to the identified out of routine event, and the selected one or more recommended actions. The service may follow or disregard any of the selected one or more recommended actions.

Abstract: One or more techniques and/or systems are provided for providing and/or applying privacy preferences for an entity. A multimedia device, such as a mobile phone, may capture multimedia content associated with an entity (e.g., a photo of a person at a restaurant). The multimedia device may identify a privacy preference for the entity. In an example, the person may be wearing a privacy object that may be recognized as corresponding to the privacy preference. In another example, a device associated with the person may emit a signal that may specify a privacy preference for the person and/or may be used by the multimedia device to identify the person (e.g., the identity may be used to query a service to obtain a privacy preference specified for the person). The privacy preference may be applied to the multimedia content (e.g., a no photography privacy preference, a no tagging privacy preference, etc.).

Abstract: Neutropenia is the dose-limiting toxicity of the tri-weekly docetaxel (Taxotere®) schedule. Here, we evaluate in Metastatic Breast Cancer (MBC) patients (N=38) a computerized method for predicting docetaxel-induced neutropenia, and use the model to identify improved docetaxel and Granulocyte Colony Stimulating Factor (G-CSF) regimens. Pharmacokinetics/pharmacodynamics (PK/PD) models were created and simulated concomitantly with a mathematical granulopoiesis model. Individual baseline neutrophil counts and docetaxel schedules served as inputs. Our trial validated the model accuracy in predicting nadir timings (r=0.99), grade 3/4 neutropenia (86% success) and neutrophil profiles (r=0.62). Model was robust to CYP3A-induced variability, except for slightly less accurate grade 3/4 neutropenia predictions.