Abstract: A road sign interpretation system includes a front-facing camera mounted on or in a vehicle collecting image data of multiple road signs. A first convolutional neural network (CNN) receives the image data from the front-facing camera and yields a set of sign predictions including one or more sign text instances. A second CNN defining a text extractor receives the image data from the front-facing camera and extracts text candidates including the multiple sign text instances. Sign and sign data localization is provided in the second CNN to compute a text order from the multiple sign text instances. A sign text synthesizer module receives individual sign text instances from the first CNN and individual ones of the sign text instances in digitized forms from an optical character recognizer (OCR). A semantic encoding and interpretation module receives the sign text instances and identifies semantics of the multiple road signs.

Abstract: A system for determining a relevance of a traffic sign for a vehicle includes at least one vehicle camera configured to provide a view of an environment surrounding the vehicle and a vehicle controller in electrical communication with the at least one vehicle camera. The vehicle controller is programmed to capture an image using the at least one vehicle camera. The vehicle controller is further programmed to identify the traffic sign in the image. The vehicle controller is further programmed to determine a pan angle and a tilt angle of the traffic sign based at least in part on the image. The vehicle controller is further programmed to determine the relevance of the traffic sign based at least in part on the pan angle and the tilt angle of the traffic sign.

Abstract: A method for language processing for a vehicle includes receiving an input text. The input text includes a plurality of words. The method also includes determining a rule-based action representation of the input text, parsing the input text to produce a parsed text. The method also includes determining a model-based action representation of the parsed text. The method also includes determining a final action representation of the input text based at least in part on the rule-based action representation and the model-based action representation.

Abstract: A road sign interpretation system includes a front-facing camera mounted on or in a vehicle collecting image data of multiple road signs. A first convolutional neural network (CNN) receives the image data from the front-facing camera and yields a set of sign predictions including one or more sign text instances. A second CNN defining a text extractor receives the image data from the front-facing camera and extracts text candidates including the multiple sign text instances. Sign and sign data localization is provided in the second CNN to compute a text order from the multiple sign text instances. A sign text synthesizer module receives individual sign text instances from the first CNN and individual ones of the sign text instances in digitized forms from an optical character recognizer (OCR). A semantic encoding and interpretation module receives the sign text instances and identifies semantics of the multiple road signs.

Abstract: Described is a system for detecting adversarial activities based on detection of activity patterns in a multiplex network. The system detects one or more subnetworks that are matches to a template network of template nodes. The subnetworks are detected by filtering multiplex network nodes according to a filtering criteria that utilizes monotone function properties in the multiplex network. Nodes that do not meet the filtering criteria are eliminated, resulting in a list of candidate nodes in the multiplex network. The one or more subnetworks are formed from the list of candidate nodes. An activity pattern corresponding to a pattern of adversarial activity is identified in the one or more subnetworks. Based on the identified activity pattern, an alert of adversarial activity is generated and transmitted.

Abstract: Described is a system for producing indicators and warnings of adversarial activities. The system receives multiple networks of transactional data from different sources. Each node of a network of transactional data represents an entity, and each edge represents a relation between entities. A worldview graph is generated by merging the multiple networks of transactional data. Suspicious subgraph regions related to an adversarial activity are identified in the worldview graph through activity detection. The suspicious subgraph regions are used to generate and transmit an alert of the adversarial activity.

Abstract: Described is a system for continuously predicting and adapting optimal strategies for attacker elicitation. The system includes a global bot controlling processor unit and one or more local bot controlling processor units. The global bot controlling processor unit includes a multi-layer network software unit for extracting attacker features from diverse, out-of-band (OOB) media sources. The global controlling processing unit further includes an adaptive behavioral game theory (GT) software unit for determining a best strategy for eliciting identifying information from an attacker. Each local bot controlling processor unit includes a cognitive model (CM) software unit for estimating a cognitive state of the attacker and predicting attacker behavior. A generative adversarial network (GAN) software unit predicts the attacker's strategies.

Abstract: Described is a system for large-scale event prediction and a corresponding response. The system, using an agent-based model, predicts how many users (agent accounts) on a social media platform will become activists related to a large-scale event. This process is accomplished using both Before and During models. Before the large-scale event, the system operates to generate agent attributes and a posting network based on posts on the social media platform. During the large-scale event and based on the agent attributes and posting network, the system determines if a social media user (agent account) will become an activist of the large-scale event and a corresponding magnitude of the large-scale event. Depending on the magnitude, the system can implement a responsive measure and control a device based on the prediction of the activists.

Abstract: Described is a system for identification of correlations in customer observables (COs). The system extracts key phrases representing COs from textual inputs from multiple data sources, wherein the COs are related to a consumer product. A unified hypergraph is constructed that models co-occurrences of COs. The unified hypergraph includes nodes and types of hyperedges connecting the nodes, where COs are represented by nodes and data sources are represented by different types of hyperedges. Each node of the unified hypergraph is embedded into a latent feature space. The unified hypergraph is partitioned into clusters within the latent feature space, where each cluster contains correlated CO data. The correlated CO data from a cluster are used to generate and provide targeted messages specific to the consumer product to a display device.

Abstract: Described is a system for identifying and communicating with polarized groups in social media platforms. The system generates a tripartite graph from online social network data. The tripartite graph incorporates user data, post data, and tag data obtained from the online social network data. Nonnegative matrix factorization is performed on a decomposed tripartite graph to obtain an optimization function. The optimization function is solved to identify polarized groups in the online social network. Based on the identified polarized groups, the system sends pre-determined communications to members of each group aimed at targeted escalation or de-escalation of polarization in an online social media platform.

Abstract: Described is a system for predicting an occurrence of large-scale events using social media data. A collection of time series is acquired from social media data related to an event of interest. The collection of time series is partitioned into time intervals and semantic features are extracted from the time intervals as a set of semantic intervals. The semantic features are encoded into a multilayer network. Subgraphs of the multilayer network are transformed into a state transition network. A prediction of a future event of interest is generated by analyzing the encoded network using the state transition network. Using the analyzed encoded network, a device is controlled based on the prediction of the future event of interest.

Abstract: Described is a system for characterizing communication devices by device type. The system obtains device information for a variety of communication device types, each device type associated with a user account of a bidirectional network. The communication device types are analyzed to perform regional and temporal device characterization, behavioral and feature device characterization, and device homophily analysis on the bidirectional network. The analysis is then used for targeted regional marketing.

Abstract: This disclosure provides a system for pairwise network alignment. In operation, the system receives datasets from two networks, each network having a plurality of nodes. The two networks are embedded based on multi-layer graph convolution to generate network embeddings. An inner product similarity score is generated between the two networks based on an inner product of the network embeddings. Next, a node correspondence is estimated between the two networks using a SoftMax function on the inner product similarity score. Finally, the two networks are aligned on the node correspondence.

Abstract: Described is a system for estimating long-term causal effects of interventions on various entities. A synthetic control is generated from selected donor series such that parameters involved in the generation of the synthetic control can be tuned automatically, or manually by a user. Post-intervention synthetic control values are generated from the synthetic control, and estimates of long-term causal effects of an intervention onto aggregate units similar to the donor series are determined. A device is controlled based on the determined estimates.

Abstract: Described is a system for automated event summarization. A multi-layer network representing a multimodal data set is generated, where nodes within a given layer represent information tokens in a given modality. A topically diverse set of nodes is ranked and selected from each layer to represent temporal event highlights. Temporal event highlights are linked into storylines. Using the storylines, the system monitors a progression of an event or opinions regarding a topic. A temporal summary of the progression of the event or the opinions regarding the topic is generated.

Abstract: Described is a system for inferring network dynamics and their sources within the network. During operation, a vector representation is generated based on states of agents in a network. The vector representation including attribute vectors that correspond to the states of the agents in the network. A matrix representation is then generated based on the changing states of agents by packing the attribute vectors at each time step into an attribute matrix. Time-evolving states of the agents are learned using dictionary learning. Influential source agents in the network are then identified by performing dimensionality reduction on the attribute matrix. Finally, in some aspects, an action is executed based on the identity of the influential source agents. For example, marketing material may be directed to a source agent's online account, or the source agent's online account can be deactivated or terminated or some other desired action can be taken.

Abstract: Described is a system for extracting multi-scale hierarchical clustering on customer observables (COs) data in a vehicle. The system selects a parameter for a set of incident data of COs data. Simplicial complexes are generated from the COs data based on the selected parameter. Face networks are generated from the simplicial complexes. For each face network, a set of connected components is extracted. Each connected component is transformed to a cluster of related COs, resulting in a first extracted relation between COs. The first extracted relation is used to automatically generate an alert at a client device when a second extracted relation different from the first extracted relation results from the transformation.

Abstract: Described is a system for identification of correlations in customer observables (COs). The system extracts key phrases representing COs from textual inputs from multiple data sources, wherein the COs are related to a consumer product. A unified hypergraph is constructed that models co-occurrences of COs. The unified hypergraph includes nodes and types of hyperedges connecting the nodes, where COs are represented by nodes and data sources are represented by different types of hyperedges. Each node of the unified hypergraph is embedded into a latent feature space. The unified hypergraph is partitioned into clusters within the latent feature space, where each cluster contains correlated CO data. The correlated CO data from a cluster are used to generate and provide targeted messages specific to the consumer product to a display device.

Abstract: Described is a system for trajectory estimation of a mobile platform, such as a UAV. In operation, the system generates an initial trajectory estimate for the mobile platform which is stored in a trajectory buffer as a buffered trajectory. Images captured at a location are compared with a location recognition database to generate a location label for a current location to designate the current location as a new location or a revisited location. If the location is a revisited location, the system determines if trajectory correction is required. If so, the buffered trajectory is corrected to generate a corrected trajectory as the drift-free trajectory. Finally, the drift-free trajectory can be used in a variety of applications. For example, the drift-free trajectory can be used to cause the mobile platform to traverse a path that coincides with the drift-free trajectory.

Abstract: Described is a system for location recognition for mobile platforms, such as autonomous robotic exploration. In operation, an image in front of the platform is converted into a high-dimensional feature vector. The image reflects a scene proximate the mobile platform. A candidate location identification of the scene is then determined. The candidate location identification is then stored in a history buffer. Upon receiving a cue, the system then determines if the candidate location identification is a known location or a new location.