Abstract: Described is an automatic anomaly detector that receives a time-series of normal and abnormal activities that include features related to entities within a computing system. A feature coherence graph for the features is constructed, with the graph then clustered such that feature spaces of entities are expanded to include features that live within a same cluster but belong to separate entities. The feature spaces are unified by mapping representations of the features spaces into a Euclidean space of feature vectors. The feature vectors related to each feature are then aligned. Sets of clusters of related abnormal activities are then generated by regressing each feature vector over only those features that it possesses. The sets of clusters are used to detect anomalous behavior. The system then identifies a node within the computer system generating the anomalous behavior and initiates an action to minimize a threat posed by the node.

Abstract: A vehicle system prognosis apparatus including sensor(s) for detecting a characteristic of a vehicle system and generating at least one time series of condition indicator values, and a processor that receives the at least one time series and generates an analysis model, for the characteristic, that is trained with one or more of the at least one time series, that are obtained from the one or more sensors with the vehicle system operating under normal conditions, extracts from the at least one time series one or more features embodying an indication of a health of the vehicle system, generates a quantified health assessment of the vehicle system by quantifying the one or more features based on a normal distribution of the one or more features from the analysis model, and communicates the quantified health assessment of the vehicle system to an operator or crew member of the vehicle.

Abstract: Described is a system for discriminant localization of objects. During operation, the system causes one or more processors to perform an operation of identifying an object in an image using a multi-layer network. Features of the object are derived from the activations of two or more layers of the multi-layer network. The image is then classified to contain one or more object classes, and the desired object class is localized. A device can then be controlled based on localization of the object in the image. For example, a robotic arm can be controlled to reach for the object.

Abstract: A fault detection system including one or more sensors onboard a vehicle to detect a characteristic of the vehicle and generate sensor signals corresponding to the characteristic, a processor onboard the vehicle to receive the sensor signals, generate one or more fast Fourier transform vectors based on the sensor signals so that the one or more fast Fourier transform vectors are representative of the characteristic, generate an analysis model from a time history of the fast Fourier transform vectors, and determine, using the analysis model, a degree to which the one or more fast Fourier transform vectors could have been generated by the analysis model, and an indicator to communicate an operational status of the vehicle to an operator or crew member of the vehicle based on the degree to which the one or more fast Fourier transform vectors could have been generated by the analysis model.

Abstract: An autonomous or semi-autonomous system includes a temporal prediction network configured to process a first set of samples from an environment of the system during performance of a first task, a controller configured to process the first set of samples from the environment and a hidden state output by the temporal prediction network, a preserved copy of the temporal prediction network, and a preserved copy of the controller. The preserved copy of the temporal prediction network and the preserved copy of the controller are configured to generate simulated rollouts, and the system is configured to interleave the simulated rollouts with a second set of samples from the environment during performance of a second task to preserve knowledge of the temporal prediction network for performing the first task.

Abstract: A method for retraining an artificial neural network trained on data from an old task includes training the artificial neural network on data from a new task different than the old task, calculating a drift, utilizing Sliced Wasserstein Distance, in activation distributions of a series of hidden layer nodes during the training of the artificial neural network with the new task, calculating a number of additional nodes to add to at least one hidden layer based on the drift in the activation distributions, resetting connection weights between input layer nodes, hidden layer nodes, and output layer nodes to values before the training of the artificial neural network on the data from the new task, adding the additional nodes to the at least one hidden layer, and training the artificial neural network on data from the new task.

Abstract: An apparatus for detecting a fault state of an aircraft is provided. The apparatus accesses a training set of flight data for the aircraft. The training set includes observations of the flight data, each observation of the flight data includes measurements of properties selected and transformed into a set of features. The apparatus builds a generative adversarial network including a generative model and a discriminative model using the training set and the set of features, and builds an anomaly detection model to predict the fault state of the aircraft. The anomaly detection model is trained using the training set of flight data, simulated flight data generated by the generative model, and a subset of features from the set of features. The apparatus deploys the anomaly detection model to predict the fault state of the aircraft using additional observations of the flight data.

Abstract: Described is a system for classifying novel objects in imagery. In operation, the system extracts salient patches from a plurality of unannotated images using a multi-layer network. Activations of the multi-layer network are clustered into key attribute, with the key attributes being displayed to a user on a display, thereby prompting the user to annotate the key attributes with class label. An attribute database is then generated based on user prompted annotations of the key attributes. A test image can then be passed through the system, allowing the system to classify at least one object in the test image by identifying an object class in the attribute database. Finally, a device can be caused to operate or maneuver based on the classification of the at least one object in the test image.

Abstract: A fault detection system including one or more sensors onboard a vehicle, the one or more sensors being configured to detect a predetermined characteristic of the vehicle and generate a plurality of sensor signals corresponding to the predetermined characteristic, and a processor onboard the vehicle and in communication with the one or more sensors, the processor being configured to generate an analysis model for the predetermined characteristic, the analysis model being trained by the processor with a training data set of fast Fourier transform vectors that are generated from the plurality of sensor signals obtained under normal operating conditions of the predetermined characteristic, and determine a health of a vehicle component corresponding to the predetermined characteristic with the analysis model.

Abstract: Described is a plastic hyper-dimensional memory system having neuronal layers. The system includes an input layer for receiving an input and an address matrix for generating a working pre-image vector from the input. A hidden layer is included for transforming the working pre-image vector into a working vector. A data matrix transforms the working vector into a data pre-image vector. Further, the hidden layer performs neurogenesis when a novel input is detected based on the working pre-image vector, where the neurogenesis comprises adding or deleting address units. Novelty detection includes using a set of reinforcement units. Finally, an output layer generates a data vector based on the data pre-image vector.

Abstract: In an example, a method for identifying associated events in an aircraft is described. The method includes obtaining sensor data, obtaining fault code data, generating a set of events, where each event occurs over a time interval over which either (i) the sensor data indicates an anomalous measurement or (ii) a fault code associated with a particular aircraft subsystem of the aircraft was signaled, calculating a value of statistical dependence between the set, based on the value exceeding a threshold, constructing a network representing the set as a sequence of related events and further representing a temporal order in which the sequence occurred, indexing, in a summary table stored in memory and separate from the sensor data and the fault code data, the sequence and the value, and controlling a display device to display the summary table and a visual representation of the network.

Abstract: A method for determining a vehicle system prognosis includes detecting a predetermined characteristic of a vehicle with one or more sensors, receiving a plurality of sensor signals from the one or more sensors and determining an input time series of data based on the sensor signals, clustering a matrix of time series data, generated from the input time series of data, into a predetermined number of hyperplanes, extracting extracted features that are indicative of an operation of a vehicle system from a sparse temporal matrix based on data point behavior with respect to two or more hyperplanes within the sparse temporal matrix and determining an operational status of the vehicle system based on the extracted features, the sparse temporal matrix being based on the predetermined number of hyperplanes; and communicating the operational status of the vehicle system to an operator or crew member of the vehicle.

Abstract: Described is a system for detecting change of context in a video stream on an autonomous platform. The system extracts salient patches from image frames in the video stream. Each salient patch is translated to a concept vector. A recurrent neural network is enervated with the concept vector, resulting in activations of the recurrent neural network. The activations are classified, and the classified activations are mapped onto context classes. A change in context class is detected in the image frames, and the system causes the autonomous platform to perform an automatic operation to adapt to the change of context class.

Abstract: Described is a system for adapting neural networks. The system receives inputs to be learned by a multi-layered spiking neural network. A first mechanism is implemented to adapt weights on the connections via competition among neurons using Hebbian learning. Activity levels of the neurons are stabilized to allow the multi-layered spiking neural network to learn the inputs. A second mechanism is implemented to increase a learning rate of a neuron over time using Hebbian learning. A third mechanism is implemented, wherein newly created neurons, representing new inputs, copy at least one synaptic structure of older neurons in the multi-layered spiking neural network. The mechanisms are used for continuous, online learning of the inputs to the multi-layered spiking neural network. An autonomous system, such as an autonomous vehicle, can use the learned inputs to learn from its environment and perform tasks, such as classification and prediction.

Abstract: Described is a system for adaptive blind source separation. A time-series of data points from one or more mixtures of source signals is continuously passed through adaptable filters, where each filter has a corresponding output signal. An error of each output signal is determined, and a filter state of each filter is determined using the error signals. A set of filter center frequencies are adapted using the set of error signals and the filter states, resulting in new filter center frequencies. The set of filter center frequencies are updated with the new filter center frequencies. Finally, separated source signals are extracted from the mixture of signals.

Abstract: In general, one aspect of the subject matter described can be embodied in a method that includes, obtaining a plurality of search results responsive to an initial search query, the search results including a first search result that identifies a first resource; determining, using a document-to-query-to-document model, that the first resource is relevant to a first suggested query different from the initial search query; generating a presentation of the search results responsive to the initial search query; and providing the presentation of the search results in response to the initial search query. Each search result in the presentation includes a link to a respective resource, wherein the first search result in the presentation includes a link that, upon a selection by a user, can cause the first suggested query to be submitted to a search engine.

Abstract: A method for generating pulse descriptor words (PDWs) including frequency and/or bandwidth data from time-varying signals received by a sensor includes filtering, at a plurality of blind source separation (BSS) modules, signals derived from the time-varying signals, each BSS module including a filtering subsystem having a plurality of filter modules. Each filter module has a frequency filter coefficient (?) and is parameterized by a center frequency (f). The method also includes transmitting at least one blind source separated signal from the BSS modules to a PDW generation module communicatively coupled to the filtering subsystem. The method further includes generating, using the PDW generation module and based on the blind source separated signal, at least one PDW parameter vector signal containing the frequency data. The method also includes updating, upon generating and based on the PDW parameter vector signal, values of ? and/or f for each filter module.

Abstract: Described is a system for classifying novel objects in imagery. In operation, the system extracts salient patches from a plurality of unannotated images using a multi-layer network. Activations of the multi-layer network are clustered into key attribute, with the key attributes being displayed to a user on a display, thereby prompting the user to annotate the key attributes with class label. An attribute database is then generated based on user prompted annotations of the key attributes. A test image can then be passed through the system, allowing the system to classify at least one object in the test image by identifying an object class in the attribute database. Finally, a device can be caused to operate or maneuver based on the classification of the at least one object in the test image.

Abstract: A method for determining a vehicle system prognosis includes detecting a predetermined characteristic of a vehicle with one or more sensors, obtaining a plurality of sensor signals corresponding to the predetermined characteristic, receiving the plurality of sensor signals from the one or more sensors and determining an input time series of data based on the sensor signals, generating, a matrix of time series data based on the input time series of data, clustering the matrix of time series data based on predetermined clustering criteria into a predetermined number of clusters, generating a sparse temporal matrix based on the predetermined number of clusters, extracting extracted features that are indicative of an operation of a vehicle system from the sparse temporal matrix and determining an operational status of the vehicle system based on the extracted features, and communicating the operational status of the vehicle system to an operator or crew member of the vehicle.

Abstract: A vehicle system prognosis apparatus including sensor(s) for detecting a characteristic of a vehicle system and generating at least one time series of condition indicator values, and a processor that receives the at least one time series and generates an analysis model, for the characteristic, that is trained with one or more of the at least one time series, that are obtained from the one or more sensors with the vehicle system operating under normal conditions, extracts from the at least one time series one or more features embodying an indication of a health of the vehicle system, generates a quantified health assessment of the vehicle system by quantifying the one or more features based on a normal distribution of the one or more features from the analysis model, and communicates the quantified health assessment of the vehicle system to an operator or crew member of the vehicle.