Respective initial feature sets are obtained for the nodes of a graph in which the nodes represent instances of entity types and edges represent relationships. Using the initial feature sets and the graph, a graph convolutional model is trained to generate one or more types of predictions. In the model, a representation of a particular node at a particular hidden layer is based on aggregated representations of neighbor nodes, and an embedding produced at a final hidden layer is used as input to a prediction layer. The trained model is stored.

Method, apparatus and computer program product for performing a cross-model predictive inference to generate a cross-model predictive output for a plurality of predictive inputs using a plurality of predictive models. For example, the apparatus includes at least one processor and at least one non-transitory memory including program code. The at least one non-transitory memory and the program code are configured to, with the at least one processor, obtain a model selection probability distribution which defines, for each predictive model, a respective selection probability score; obtain, for each predictive model, respective cross-model normalization data; for each predictive input, determine a cross-model predictive score; and determine, based on each determined cross-model predictive score, the cross-model predictive output.

A system and method for geolocation-aware, cyber-enabled infrastructure inventory and asset management with state prediction capability. The system tracks tangible and intangible assets, including states associated with each asset such as the location, condition, and value of each asset. Physical assets may be cyber-enabled by attaching wireless computing devices to some or all of the physical assets to provide data about the physical assets using sensors of the computing devices, including but not limited to, such data as location, conditions of storage, and hours of operation or use. Data for each item is stored in a multi-dimensional time series database, which keeps a historical record of the states of each item. Unknown or future states can be predicted by applying predictive models to the time series data. Parametric evaluations of current and predicted future states can be used to optimize the assets against an objective.

In various example embodiments, a system and method for computing a query feature score that can be used by a machine learning algorithm to rank search results is described. A query is received. The assigned probabilities for the arbitrary query is retrieved. The assigned probabilities is based on a probability model for a query feature. A score is computed for the query feature based on the assigned probabilities for the arbitrary query.

Social networking system users may create events where a group of other users invited to the event meet at a specified time and location. The social networking system suggests users to invite to an event based on a prediction that the users would attend the event if invited. Various factors may be used to make the prediction, such as an affinity between the inviting user and the other users, the availability of the other users at the time of the event and/or the proximity of the other users to the location of the event. An inviting user receives the suggested users and selects suggested users to invite to the event or invitations may be automatically sent to the suggested users by the social networking system.

A method, apparatus, and computer readable medium are provided for maximizing consumption of broadcasts by a producer. An example method includes receiving selection of a total number of time slots to use for scheduling broadcasts, and receiving information regarding the producer's followers. The example method further 5 includes identifying, by a processor and based on the received information, discount factors associated with the producer's followers, and calculating, by the processor and based on the received information, a predicted number of competitor broadcasts during each time slot of the total number of time slots. Finally, the example method includes determining, by the processor and based on the discount factors and the predicted 10 number of competitor broadcasts during each time slot, a number of broadcasts for the producer to transmit in each time slot of the total number of time slots.

Specialized robot motion planning hardware and methods of making and using same are provided. A robot-specific hardware can be designed using a tool that receives a robot description comprising a collision geometry of a robot, degrees of freedom for each joint of the robot, and joint limits of the robot; receives a scenario description; generates a probabilistic roadmap (PRM) using the robot description and the scenario description; and for each edge of PRM, produces a collision detection unit comprising a circuit indicating all parts of obstacles that collide with that edge. The hardware is implemented as parallel collision detection units that provide collision detection results used to remove edges from the PRM that is searched to find a path to a goal position.

Aspects of the present disclosure relate to systems and methods for automatically constructing a new formula based upon an existing formula. Various methods are employed to analyze an existing formula in order to determine an intent. Once the intent is determined, a new formula may be constructed that satisfies the intent of the existing formula. Further aspects relate to validating the new formula to ensure it produces the same result as the existing formula which it is intended to replace. Various user interface elements are also disclosed which may be employed to provide the newly constructed formulas to users of an application. The various aspects to disclosed herein may be performed by an application, such as a spreadsheet application, performed in real-time, or provided as a remote service.

The present invention relates to methods for evaluating and/or predicting the outcome of a clinical condition, such as cancer, metastasis, AIDS, autism, Alzheimer's, and/or Parkinson's disorder. The methods can also be used to monitor and track changes in a patient's DNA and/or RNA during and following a clinical treatment regime. The methods may also be used to evaluate protein and/or metabolite levels that correlate with such clinical conditions. The methods are also of use to ascertain the probability outcome for a patient's particular prognosis.

Systems and methods for determining location-based bid modifier suggestions include determining a content placement cost based in part on a likelihood of a user that has entered a physical establishment completing a transaction, an average transaction amount for the establishment, and an expected return on investment (ROI). A location-based bid modifier may be determined using the computed cost and a base bid amount. In some implementations, the location-based bid modifier may also be based on a probability model that models the probability of the user visiting the establishment.