Systems and methods for predicting yearly performance improvement rates for nearly all definable technologies for the first time are provided. In one embodiment, a correspondence of all patents within the U.S. patent system to a set of technology domains is created. From the identified patent sets, the invention may calculate average centrality of the patents in each domain to predict improvement rates, following a patent network-based methodology. Also disclosed is a system to intake a user technology search query and match user intent with the technology domain as well as the corresponding improvement rate.

A method for estimating position of a mobile device which includes receiving, from a network server, observed time difference of arrival (OTDOA) assistance data for a first plurality of cells from a base station almanac (BSA) accessible to the network server. The OTDOA assistance data is stored, within a memory of the mobile device, as a first micro-BSA. A position estimate for the mobile device is determined based upon time difference of arrival (TDOA) measurements associated with an initial subset of the first plurality of cells and initial OTDOA assistance data corresponding to the initial subset of the first plurality of cells. The initial OTDOA assistance data may be generated by the micro-BSA based upon an initial seed estimate.

A method of detecting clouds in an acquired aerial image includes determining a region of a reference aerial image corresponding to a region of an acquired aerial image. For each of a plurality of locations over the region of the acquired aerial image and corresponding to a plurality of locations over the region of the reference aerial image, the mutual information of one or more variables associated with the location in the acquired aerial image and one or more variables associated with the corresponding location in the reference aerial image is calculated. Using the mutual information calculated for each of the plurality of locations over the region of the acquired aerial image, it is determined when the acquired aerial image displays a cloud at the location in the region of the acquired aerial image.

A system for generating valuation data of a private company. The system includes a data merger, a model trainer, a user input receiver, and a model predictor. The data merger is for receiving company data. At least one company metric of the plurality of company metrics corresponds to a company other than the private company. The model trainer is for generating a machine learning model, based on the company data. The machine learning model includes a plurality of variables. Each variable of the plurality of variables corresponds to at least one company metric of the plurality of company metrics. The user input receiver is for receiving a request to generate the valuation data. The model predictor is for generating the valuation data based on the machine learning model and the request to generate the valuation data.

A learning apparatus (10) includes: a generation unit (11) configured having a mathematical model for generating data through an input of a random number used for deep learning to a nonlinear function; and a prior learning unit (13) configured to cause the generation unit (11) to execute prior learning of a variance and an average using unscented transform (UT). The prior learning unit (13) estimates, using UT, the variance and the average of data generated by the generation unit (11) and updates a parameter of the generation unit (11) to minimize an evaluation function for evaluating a similarity between the estimated variance and average and a variance and an average of true data calculated in advance.

A method for training a robust deep neural network model in collaboration with a standard model in a minimax game in a closed learning loop. The method encourages the robust and standard models to align their feature spaces by utilizing the task-specific decision boundaries and explore the input space more broadly. The supervision from the standard model acts as a noise-free reference for regularizing the robust model. This effectively adds a prior on the learned representations which encourages the model to learn semantically relevant features which are less susceptible to off-manifold perturbations introduced by adversarial attacks. The adversarial examples are generated by identifying regions in the input space where the discrepancy between the robust and standard model is maximum within the perturbation bound. In the subsequent step, the discrepancy between the robust and standard models is minimized in addition to optimizing them on their respective tasks.

Disclosed herein are systems and methods of artificial intelligence learning systems. In some embodiments the artificial intelligence system presents options to users based on their life stage and personality profile. Family or group structures may be created within an application. Options may be created and presented based on the family structure such as chores may be assigned to children, money may be transferred between family members, and scores may be assigned to different users.

Disclosed is an approach for detecting malicious network activity (e.g. based on a data hoarding activity identifies using a graph mixture density neural network (GraphMDN)). Generally, the approach includes generating embeddings using a graph convolution process and then processing the embeddings using a mixture density neural network. The approach may include collecting network activity data, generating a graph representing the network activity, or an aggregation thereof that maintains the inherent graphical nature and characteristics of the data, and training a GraphMDN in order to generate pluralities of distributions characterizing one or more aspects of the graph representing the network activity. The approach may also include capturing new network activity data, and evaluating that data using the distributions generated by the trained GraphMDN, and generation corresponding detection results.

Methods and systems are provided for selectively denoising medical images. In an exemplary method, one or more deep learning networks are trained to map corrupted images onto a first type and a second type of artifacts present in corresponding corrupted images. Then the one or more trained learning networks are used to single out the first and second types of artifacts from a particular medical image. The first type of artifacts is removed to a first extent and the second type of artifacts is removed to a second extent. The first and second extents may be different. For example, one type of artifacts can be fully suppressed while the other can be partially removed form the medical image.

A target speech signal extraction method for robust speech recognition includes: initializing a steering vector for a target speech source and an adaptive vector, setting a real output channel of the target speech source as an output by the adaptive vector, initializing adaptive vectors for a noise and setting a dummy channel as an output by the adaptive vectors for the noise; setting a cost function for minimizing dependency between a real output for the target speech source and a dummy output for the noise; setting an auxiliary function to the cost function, and updating the adaptive vector for the target speech source and the adaptive vectors for the noise by using the auxiliary function and the steering vector; estimating the target speech signal by using the adaptive vector thereby extracting the target speech signal from the input signals; and updating the steering vector for the target speech source.