A registration processor (74) is configured to obtain articulated brain substructures using acquired brain image data and template brain image data. The registration processor (74) annotates the brain image data; registers the brain image data with template image data using global brain registration; and registers at least one brain structure of the brain image data a corresponding brain structure of the template image data using a local brain substructure registration. The registration processor (74) articulates articulated substructures of the registered brain structures to improve registration using articulated substructure registration.

A system trains a facial recognition modeling system using an extremely large data set of facial images, by distributing a plurality of facial recognition models across a plurality of nodes within the facial recognition modeling system. The system optimizes a facial matching accuracy of the facial recognition modeling system by increasing a facial image set variance among the plurality of facial recognition models. The system selectively matches each facial image within the extremely large data set of facial images with at least one of the plurality of facial recognition models. The system reduces the time associated with training the facial recognition modeling system by load balancing the extremely large data set of facial images across the plurality of facial recognition models while improving the facial matching accuracy associated with each of the plurality of facial recognition models.

Before dimension reduction is performed while local data distribution is stored as neighborhood data, a distance between data to be subjected to the dimension reduction is calculated, and a parameter (a neighborhood number of the k-nearest neighbor algorithm or a size of a hypersphere) which determines the neighborhood data is determined for each data to be subjected to the dimension reduction. Thereafter, the dimension reduction is performed on the target data based on the determined parameter.

A method for extracting hierarchical features from data defined on a geometric domain is provided. The method includes applying on said data at least an intrinsic convolution layer, including the steps of applying a patch operator to extract a local representation of the input data around a point on the geometric domain and outputting the correlation of a patch resulting from the extraction with a plurality of templates. A system to implement the method is also described.

Decoding of a first message is disclosed, wherein first and second messages are encoded by a code (represented by a state machine) to produce first and second code words, which are received over a communication channel. A plurality of differences (each corresponding to a hypothesized value of a part of the first message) between the first and second messages are hypothesized. An initial code word segment is selected having, as associated previous states, a plurality of initial states (each associated with a hypothesized difference and uniquely defined by the hypothesized value of the part of the first message). The first message is decoded by (for each code word segment, starting with the initial code word segment): determining first and second metrics associated with respective probabilities that the code word segment of the first and second code word (respectively) corresponds to a first message segment content, the probability of the second metric being conditional on the hypothesized difference of the initial state associated with the previous state of the state transition corresponding to the first message segment content, determining a decision metric by combining the first and second metrics, and selecting (for the first message) the first message segment content or a second message segment content based on the decision metric. If the first message segment content is selected, the subsequent state of the state transition corresponding to the first message segment content is associated with the initial state associated with the previous state of the state transition.

An abnormality detection device of an embodiment includes an encoder, a first identifier, a decoder, and a second identifier. The encoder is configured to compress input data using a compression parameter to generate a compressed data. The first identifier is configured to determine whether a distribution of the compressed data input by the encoder is a distribution of the compressed data or a prior distribution prepared in advance, and inputs a first identification result to the encoder. The decoder is configured to decode the compressed data using a compressing parameter to generate reconstructed data. The second identifier is configured to determine whether the reconstructed data input by the decoder is the reconstruction data or the input data and outputs a second identification result to the encoder and the decoder.

A system trains a facial recognition modeling system using an extremely large data set of facial images, by distributing a plurality of facial recognition models across a plurality of nodes within the facial recognition modeling system. The system optimizes a facial matching accuracy of the facial recognition modeling system by increasing a facial image set variance among the plurality of facial recognition models. The system selectively matches each facial image within the extremely large data set of facial images with at least one of the plurality of facial recognition models. The system reduces the time associated with training the facial recognition modeling system by load balancing the extremely large data set of facial images across the plurality of facial recognition models while improving the facial matching accuracy associated with each of the plurality of facial recognition models.

A system trains a facial recognition modeling system using an extremely large data set of facial images, by distributing a plurality of facial recognition models across a plurality of nodes within the facial recognition modeling system. The system optimizes a facial matching accuracy of the facial recognition modeling system by increasing a facial image set variance among the plurality of facial recognition models. The system selectively matches each facial image within the extremely large data set of facial images with at least one of the plurality of facial recognition models. The system reduces the time associated with training the facial recognition modeling system by load balancing the extremely large data set of facial images across the plurality of facial recognition models while improving the facial matching accuracy associated with each of the plurality of facial recognition models.

Described are methods, systems, and computer-program product embodiments for selecting a face image based on a name. In some embodiments, a method includes receiving the name. Based on the name, a name vector is selected from a plurality of name vectors in a dataset that maps a plurality of names to a plurality of corresponding name vectors in a vector space, where each name vector includes representations associated with a plurality of words associated with each name. A plurality of face vectors corresponding to a plurality of face images is received. A face vector is selected from the plurality of face vectors based on a plurality of similarity scores calculated for the plurality of corresponding face vectors, where for each name vector, a similarity score is calculated based on the name vector and each face vector. The face image is output based on the selected face vector.

Generally, the present disclosure is directed to user interface understanding. More particularly, the present disclosure relates to training and utilization of machine-learned models for user interface prediction and/or generation. A machine-learned interface prediction model can be pre-trained using a variety of pre-training tasks for eventual downstream task training and utilization (e.g., interface prediction, interface generation, etc.).