Abstract: Training of a machine vision model, a segmentation model, is performed by using an acquisition function for a small number of pixels of one or more training images. The acquisition function uses first mutual information and second mutual information to identify unlabelled pixels which are labelled with high uncertainty when predicting possible label values. Training, prediction of labels, identifying pixels with highly uncertain labels, obtaining labels only for those pixels with highly uncertain labels and retraining are performed iteratively to finally provide the machine vision model. The iterative approach uses very few labelled pixels to obtain the final machine vision model. The machine vision model accurately labels areas of a data image.

Abstract: A problem of supervised learning is overcome by using patches to discover objects in unlabeled training images. The discovered objects are embedded in a pattern space. An AI machine replaces manual entry steps of training with a machine-centric process including clustering in a pixel space, clustering in latent space and building the pattern space based on different losses derived from pixel space clustering and latent space clustering. A distance structure in the pattern space captures the co-occurrence of patterns due to frequently appearing objects in training image data. Embodiments provide image representation based on local image patch naturally handles the position and scale invariance property that is important to effective object detection. Embodiments successfully identifies frequent objects such as human faces, human bodies, animals, or vehicles from unorganized data images based on a small quantity of training images.

Abstract: A problem of imbalanced big data is solved by decoupling a classifier into a neural network for generation of representation vectors and into a classification model for operating on the representation vectors. The neural network and the classification model act as a mapper classifier. The neural network is trained with an unsupervised algorithm and the classification model is trained with a supervised active learning loop. An acquisition function is used in the supervised active learning loop to speed arrival at an accurate classification performance, improving data efficiency. The accuracy of the hybrid classifier is similar to or exceeds the accuracy of comparative classifiers in all aspects. In some embodiments, big data includes an imbalance of more than 10:1 in image classes. The hybrid classifier reduces labor and improves efficiency needed to arrive at an accurate classification performance, and improves recognition of previously-unrecognized images.

Abstract: The present invention is a method and system for detecting shopping groups based on the dynamic relation between shoppers' trajectories. First, shopper trajectories are generated using video images or positioning devices. Then, group behavior features are extracted from a candidate shopping group trajectory pair. The group behavior features of a given pair of trajectories are typically the changes in positional differences and the average speed of these trajectories. From a model of shopping group behavior, a given pair of candidate shopping group trajectories is analyzed to determine the group score—the likelihood of the pair of shoppers indeed belonging to the same group. Lastly, the system utilizes graph segmentation framework to find clusters where trajectories belonging to each cluster have tight group scores with each other.

Abstract: The present invention is a method and system to provide correspondences between point-of-sale data registered at the store checkout and shopper behavior data observed at point-of-purchase through video analysis. The point-of-sale data include the list of shoppers and purchase items, and the shopper behavior data include the purchase events along with observed purchase items. The correspondence in the form of checkout shopper IDs matched to purchase event IDs is derived based on the algebraic constraint among the point-of-sale data and the purchase event data. Additional constraint based on shopper tracks and checkout/event times can also be incorporated to the correspondence problem. Uncertainties due to the video measurement can be systematically handled utilizing a Bayesian model.

Abstract: Described is a system for neural decoding of neural activity. Using at least one neural feature extraction method, neural data that is correlated with a set of behavioral data is transformed into sparse neural representations. Semantic features are extracted from a set of semantic data. Using a combination of distinct classification modes, the set of semantic data is mapped to the sparse neural representations, and new input neural data can be interpreted.

Abstract: The present invention is a method and system for measuring viewership of people for a displayed object. The displayed object can be specific in-store marketing elements, such as static signage, POP displays, and other forms of digital media, including retail TV networks and kiosks. In the present invention, the viewership comprises impression level, impression count of the viewers, such as how many people actually viewed said displayed object, average length of impression, distribution of impressions by time of day, and rating of media effectiveness based on audience response. The viewership of people is performed automatically based on the 3-dimensional face pose estimation of the people, using a plurality of means for capturing images and a plurality of computer vision technologies on the captured visual information.

Abstract: Described is a system for identifying predictive keywords and generating a forecast. The system receives time-series of keyword counts (the time-series of keyword counts having a plurality of candidate keywords). The time-series of keyword counts are separated (i.e., marked or designated) into a group of time-series from active periods and a group of time-series from inactive periods. A covariance matrix is generated for each group of time-series. Generalized eigenvectors are generated between the two covariance matrices. Candidate keywords are ranked based on the generalized eigenvectors, such that candidate keywords having a rank exceeding a predetermined threshold are designated as predictive keywords. The predictive keywords are then provided to a machine learning system that generates a forecast based on the predictive keywords. Finally, a device is operated based on the forecast.

Abstract: Described is a system for generating a semantic space based on the lexical relations between words. The system determines synonym and antonym relations between a set of words. A lexical graph is generated based on the synonym and antonym relations. Manifold embedding of the lexical graph is determined, and Laplacian coordinates of the manifold embedding are assigned as semantic features of the set of words. A quantitative representation of the set of words is generated using the semantic features.

Abstract: A method and apparatus for improving a performance of an antenna system. An influence network is generated for an array of elements in the antenna system using a reconfiguration algorithm. The influence network indicates an influence of an undesired event occurring at any element in the array of elements on a remaining portion of the array of elements given a current state of the array of elements. A relative ranking of vulnerability is created for elements in the array of elements based on the influence network. The reconfiguration algorithm is modified to take into account the relative ranking of vulnerability to form a modified reconfiguration algorithm. The antenna system using the modified reconfiguration algorithm to compensate for undesired events improves the performance of the antenna system.

Abstract: The present invention is a method and system for automatically recognizing which products a shopper intends to find or purchase based on the shopper's trajectory in a retail aisle. First, the system detects and tracks the person to generate the trajectory of the shopper. Then some of the dynamic features are extracted from the shopper trajectory. The shopper trajectory features of a given trajectory are typically the positions, the motion orientations, and speeds at each point of the trajectory. A shopper behavior model is designed based on some of the primitive actions of shoppers. The last step of the method is to analyze a given shopper trajectory to estimate the shopper's intention. The step either utilizes decision rules based on the extracted shopper trajectory features, or utilizes a trained Hidden Markov Model, to estimate the progression of the primitive actions from the trajectory.

Abstract: Described is a system for predicting system instability. The system can measure the degree of the network's instability due to critical transitions using the leading eigenvalue of the covariance matrix, where the instability measure is invariant to (1) the changes in network structure in terms of addition/removal of nodes and links, and (2) the feedback of the global system stability to the changes in stability. Based on that, the system is operable for providing an estimation of the network's changing connectivity when the network is near critical transitions.

Abstract: Described is a system for supporting human intelligence analysis. The system detects changes in social relations among users within a dynamic information network and enables understanding of a current social situation in the dynamic information network through multiple integrated modules. An active network mining module identifies incomplete data that is related to at least one change in the social relations and resolves conflicting and missing data in the dynamic information network. A relevant network discovery module constructs a relevant network from hidden relations within the dynamic information network. An information-aware social network module constructs an information-aware social network using the relevant network, then classifies and prioritizes items of interest to provide an assessment of a current social situation to a user.

Abstract: The present invention is a system and method for performing ethnicity classification based on the facial images of people, using multi-category decomposition architecture of classifiers, which include a set of predefined auxiliary classifiers that are specialized to auxiliary features of the facial images. In the multi-category decomposition architecture, which is a hybrid multi-classifier architecture specialized to ethnicity classification, the task of learning the concept of ethnicity against significant within-class variations, is handled by decomposing the set of facial images into auxiliary demographics classes; the ethnicity classification is performed by an array of classifiers where each classifier, called an auxiliary class machine, is specialized to the given auxiliary class. The facial image data is annotated to assign the age and gender labels as well as the ethnicity labels.

Abstract: A method and apparatus for improving a performance of a system. An influence network is generated for an array of elements in the system using a reconfiguration algorithm. The influence network indicates an influence of an undesired event occurring at any element in the array of elements on a remaining portion of the array of elements given a current state of the array of elements. A relative ranking of vulnerability is created for elements in the array of elements based on the influence network. The reconfiguration algorithm is modified to take into account the relative ranking of vulnerability to form a modified reconfiguration algorithm. The system using the modified reconfiguration algorithm to compensate for undesired events improves the performance of the system.

Abstract: Described is a cyber security system for digital artifact genetic modeling and forensic analysis. The system identifies the provenance (origin) of a digital artifact by first receiving a plurality of digital artifacts, each digital artifact possessing features. Raw features are extracted from the digital artifacts. The raw features are classified into descriptive genotype-phonotype structures. Finally, lineage, heredity, and provenance of the digital artifacts are determined based on mapping of the genotype-phenotype structures.

Abstract: The present invention is directed to a data processing apparatus and a computer implemented method for modeling and analyzing relational data represented in a network that includes a plurality of nodes and a plurality of connections between the nodes. The method includes assigning at least one weight to a connection between two nodes in the network. A set of possible dendrograms is then generated for the network, and a likelihood of each dendrogram in the set is determined. The determination of the likelihood is based on at least the one weight of the connection. One of the dendrograms from the set is selected as an optimal dendrogram based on the determined likelihood. The selected dendrogram is then output via an output device. The dendrogram may be used to predict missing links or identify any possible false-positive (noisy) links within a relational dataset.

Abstract: Described is a system for detecting insider threats in a network. In detecting the insider threat, the system receives data from the network relevant to network activity and extracts observable actions from the data relevant to a mission. The observable actions are combined to provide contextual cues and reasoning results. Based on the observable actions and reasoning results, proposed security policy updates are proposed to force insiders into using more observable actions. Finally, the system detects potential insider threats through analyzing the observable actions and reasoning results.

Abstract: Described is a system for catastrophe prediction. The system generates a time series of observables at multiple time steps from data observed from a complex system. A surrogate time series based on the time series of observables is then generated. Inferred network structures for both the time series of observables and the surrogate time series are reconstructed. Next, spatial autocorrelation for each inferred network structure in both the time series of observables and the surrogate time series is computed. A statistical test of a detected trend between the time series of observables and the surrogate time series is computed to determine if the detected trend occurred by chance. Finally, an early warning signal of the detected trend occurring by chance is generated.

Abstract: Described is a system for explosive network attack and mitigation analysis. A network structure is received as input. A network attack method that applies an Achlioptas process is selected. Then, an explosive mitigation strategy is selected. An attack-mitigation competing process is simulated for the network structure. A sequence of network structures under competing processes is generated. The effectiveness of the selected explosive mitigation strategy against the selected network attack method is quantified by analyzing the sequence of network structures under competing processes.