Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for determining a control policy for a vehicles or other robot through the performance of a reinforcement learning simulation of the robot.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an actor neural network used to select actions to be performed by an agent interacting with an environment. One of the methods includes obtaining a minibatch of experience tuples; and updating current values of the parameters of the actor neural network, comprising: for each experience tuple in the minibatch: processing the training observation and the training action in the experience tuple using a critic neural network to determine a neural network output for the experience tuple, and determining a target neural network output for the experience tuple; updating current values of the parameters of the critic neural network using errors between the target neural network outputs and the neural network outputs; and updating the current values of the parameters of the actor neural network using the critic neural network.

Abstract: The current application is directed to intelligent controllers that use sensor output and electronically stored information to determine whether or not one or more types of entities are present within an area, volume, or environment monitored by the intelligent controllers. The intelligent controllers select operational modes and/or modify control schedules with respect to the presence and absence of the one or more entities. The intelligent controllers selectively carry out scheduled control operations during periods of time when one or more types of entities are determined not to be in a controlled environment.

Abstract: A system includes one or more computers and one or more storage devices storing instructions that when executed by the one or more computers cause the computers to implement a combined machine learning model for processing an input including multiple features to generate a predicted output for the machine learning input. The combined model includes: a deep machine learning model configured to process the features to generate a deep model output; a wide machine learning model configured to process the features to generate a wide model output; and a combining layer configured to process the deep model output generated by the deep machine learning model and the wide model output generated by the wide machine learning model to generate the predicted output, in which the deep model and the wide model have been trained jointly on training data to generate the deep model output and the wide model output.

Abstract: A system and method for troubleshooting problems in complex systems using multiple knowledgebases comprises a first knowledgebase that has a case-based reasoning engine and knowledge from which a first set of possible solutions can be determined, and a second knowledgebase that has a case-based reasoning engine and knowledge from which a second set of possible solutions can be determined. The first knowledgebase pertains to a first equipment, and the second knowledgebase pertains to a second equipment. The second equipment is a component of the first equipment. A Federation Manager associated with the first knowledgebase transfers a case-based reasoning session between the first and second knowledgebases.

Abstract: A method of determining a set of prescribed actions includes receiving a configuration script identifying a set of influencers, a set of performance indicators, a model type, a target time, and a prescription method. The method further includes deriving a model of the model type based on data associated with the set of influencers or with the set of performance indicators. The method also includes projecting a set of future influencer values associated with the set of influencers and projecting a set of future indicator values of the set of performance indicators at the target time using the model. The method can further include prescribing using the prescription method and based on projecting using the model a set of prescribed actions associated with the subset of actionable influencers. The method also includes displaying the set of prescribed actions.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a model using parameter server shards. One of the methods includes receiving, at a parameter server shard configured to maintain values of a disjoint partition of the parameters of the model, a succession of respective requests for parameter values from each of a plurality of replicas of the model; in response to each request, downloading a current value of each requested parameter to the replica from which the request was received; receiving a succession of uploads, each upload including respective delta values for each of the parameters in the partition maintained by the shard; and updating values of the parameters in the partition maintained by the parameter server shard repeatedly based on the uploads of delta values to generate current parameter values.

Abstract: Detecting propensity profile for a person may comprise receiving artifacts associated with the person; detecting profile characteristics for the person based on the artifacts; receiving a plurality of predefined profiles comprising a plurality of characteristics and relationships between the characteristics over time, each of the plurality of predefined profiles specifying an indication of propensity; matching the profile characteristics for the person with one or more of the plurality of predefined profiles; and outputting one or more propensity indicators based on the matching, the propensity indicators comprising at least an expressed strength of a given propensity in the person at a given time.

Abstract: Testing machine learning sensors by adding obfuscated training data to test data, and performing real time model fit analysis on live network traffic to determine whether to retrain.

Abstract: A method for performing program analysis includes receiving programs of a first platform that have been assigned a first label and programs of the first platform that have been assigned a second label. Each of the programs of the first platform is expressed as platform-independent logical features. A discriminatory model or classifier is trained, using machine learning, based on the expression of the programs of the first platform as platform-independent logical features, to distinguish between programs of the first label and programs of the second label. An unlabeled program of a second platform is received and is expressed as platform-independent logical features. The trained discriminatory model or classifier is used to determine if the unlabeled program warrants the first label or the second label, based on the expression of the unlabeled program as platform-independent logical features.

Abstract: Approaches are provided for answering an inquiry of a cognitive distributed network. An approach includes receiving the inquiry at the cognitive distributed network. The approach further includes determining a classification for the inquiry based on natural language of the inquiry. The approach further includes classifying the inquiry as a single question class. The approach further includes determining, by at least one computing device, a type of introspection to be used by the cognitive distributed network on the inquiry. The approach further includes generating an answer to the inquiry based on the determined type of introspection.

Abstract: Certain relationships representing material insights are identified from among a set of discovered relationships. Cognitive discovery of relationships in a knowledge base, or corpus, are ranked according to one or more metrics indicative of material insights, including recentness and degree of alignment.

Abstract: Systems and methods are provided for a computer-implemented method of providing an interactive avatar that reacts to a communication from a communicating party. Data from an avatar characteristic table is provided to an avatar action model, where the avatar characteristic table is a data structure stored on a computer-readable medium that includes values for a plurality of avatar personality characteristics. A communication with the avatar is received from the communicating party. A next state for the avatar is determined using the avatar action model, where the avatar action model determines the next state based on the data from the avatar characteristic table, a current state for the avatar, and the communication. The next state for the avatar is implemented, and the avatar characteristic table is updated based on the communication from the communicating party, where a subsequent state for the avatar is determined based on the updated avatar characteristic table.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for selecting action slates using reinforcement learning. One of the methods includes receiving an observation characterizing a current state of an environment; selecting an action slate by processing the observation and a plurality of candidate action slates using a deep neural network, wherein each candidate action slate comprises a respective plurality of actions from the set of actions, and wherein the deep neural network is configured to, for each of the candidate action slates, process the observation and the actions in the candidate action slate to generate a slate Q value for the candidate action slate that is an estimate of a long-term reward resulting from the candidate action slate being provided to the action selector in response to the observation; and providing the selected action slate to an action selector in response to the observation.

Abstract: A method for performing program analysis includes receiving programs of a first platform that have been assigned a first label and programs of the first platform that have been assigned a second label. Each of the programs of the first platform is expressed as platform-independent logical features. A discriminatory model or classifier is trained, using machine learning, based on the expression of the programs of the first platform as platform-independent logical features, to distinguish between programs of the first label and programs of the second label. An unlabeled program of a second platform is received and is expressed as platform-independent logical features. The trained discriminatory model or classifier is used to determine if the unlabeled program warrants the first label or the second label, based on the expression of the unlabeled program as platform-independent logical features.

Abstract: A method for optimizing a hyperparameter of an auto-labeling device performing auto-labeling and auto-evaluating of a training image to be used for learning a neural network is provided for computation reduction and achieving high precision. The method includes steps of: an optimizing device, (a) instructing the auto-labeling device to generate an original image with its auto label and a validation image with its true and auto label, to assort the original image with its auto label into an easy-original and a difficult-original images, and to assort the validation image with its own true and auto labels into an easy-validation and a difficult-validation images; and (b) calculating a current reliability of the auto-labeling device, generating a sample hyperparameter set, calculating a sample reliability of the auto-labeling device, and optimizing the preset hyperparameter set. This method can be performed by a reinforcement learning with policy gradient algorithms.

Abstract: The present disclosure provides for generating a spiking neural network. Generating a spiking neural network can include determining that a first input fan-in from a plurality of input neurons to each of a plurality of output neurons is greater than a threshold, generating a plurality of intermediate neurons based on a determination that the first input fan-in is greater than the threshold, and coupling the plurality of intermediate neurons to the plurality of input neurons and the plurality of output neurons, wherein each of the plurality of intermediate neurons has a second input fan-in that is less than the first input fan-in and each of the plurality of output neurons has a third input fan-in that is less than the first input fan-in.

Abstract: A computational method via a hybrid processor comprising an analog processor and a digital processor includes determining a first classical spin configuration via the digital processor, determining preparatory biases toward the first classical spin configuration, programming an Ising problem and the preparatory biases in the analog processor via the digital processor, evolving the analog processor in a first direction, latching the state of the analog processor for a first dwell time, programming the analog processor to remove the preparatory biases via the digital processor, determining a tunneling energy via the digital processor, determining a second dwell time via the digital processor, evolving the analog processor in a second direction until the analog processor reaches the tunneling energy, and evolving the analog processor in the first direction until the analog processor reaches a second classical spin configuration.

Abstract: A method includes maintaining respective episodic memory data for each of multiple actions; receiving a current observation characterizing a current state of an environment being interacted with by an agent; processing the current observation using an embedding neural network in accordance with current values of parameters of the embedding neural network to generate a current key embedding for the current observation; for each action of the plurality of actions: determining the p nearest key embeddings in the episodic memory data for the action to the current key embedding according to a distance measure, and determining a Q value for the action from the return estimates mapped to by the p nearest key embeddings in the episodic memory data for the action; and selecting, using the Q values for the actions, an action from the multiple actions as the action to be performed by the agent.

Abstract: Embodiments of the present invention provide an automated image tagging system that can predict a set of tags, along with relevance scores, that can be used for keyword-based image retrieval, image tag proposal, and image tag auto-completion based on user input. Initially, during training, a clustering technique is utilized to reduce cluster imbalance in the data that is input into a convolutional neural network (CNN) for training feature data. In embodiments, the clustering technique can also be utilized to compute data point similarity that can be utilized for tag propagation (to tag untagged images). During testing, a diversity based voting framework is utilized to overcome user tagging biases. In some embodiments, bigram re-weighting can down-weight a keyword that is likely to be part of a bigram based on a predicted tag set.