A method for optimizing a neural network is provided, including: (1) capturing, via a first sensor group having a first field of view, a first sample set having a first sensor domain corresponding to the first field of view; (2) capturing, via a second sensor group having a second field of view, a second sample set having a second sensor domain corresponding to the second field of view; (3) generating regions of interest of the second sample set; (4) translating the regions of interest to the first sensor domain; (5) identifying nodes of the neural network which correspond to the translated regions; and (6) optimizing the neural network by at least one of (a) increasing the weight value of the nodes corresponding to the one or more translated regions and (b) decreasing the weight value of the nodes not corresponding to the one or more translated regions.

An encoding device for video analysis and composition includes circuitry configured to receive an input video having a first data volume, determine at least a region of interest of the input video, and encode at least an output video as a function of the input video and the at least a region of interest, wherein the at least an output video has at least a second data volume, and the at least a second data volume is less than the first data volume.

An automatic recognition and classification method for electrocardiogram heartbeat based on artificial intelligence, comprising: processing a received original electrocardiogram digital signal to obtain heartbeat time sequence data and lead heartbeat data; cutting the lead heartbeat data according to the heartbeat time sequence data to generate lead heartbeat analysis data; performing data combination on the lead heartbeat analysis data to obtain a one-dimensional heartbeat analysis array; performing data dimension amplification and conversion according to the one-dimensional heartbeat analysis array to obtain four-dimensional tensor data; and inputting the four-dimensional tensor data to a trained LepuEcgCatNet heartbeat classification model, to obtain heartbeat classification information. The method overcomes the defect that the conventional method only depends on single lead independent analysis for result summary statistics and thus classification errors are more easily obtained, and the accuracy of the electrocardiogram heartbeat classification is greatly improved.

Techniques are described herein for generating a dynamical systems control policy. A non-parametric family of smooth maps is defined on which vector-field learning problems can be formulated and solved using convex optimization. In some implementations, techniques described herein address the problem of generating contracting vector fields for certifying stability of the dynamical systems arising in robotics applications, e.g., designing stable movement primitives. These learning problems may utilize a set of demonstration trajectories, one or more desired equilibria (e.g., a target point), and once or more statistics including at least an average velocity and average duration of the set of demonstration trajectories. The learned contracting vector fields may induce a contraction tube around a targeted trajectory for an end effector of the robot. In some implementations, the disclosed framework may use curl-free vector-valued Reproducing Kernel Hilbert Spaces.

A method of secure classification of input data by a convolutional neural network (CNN), including (a) determination, by application of the CNN to the input data, of a first classification vector associating with each of a plurality of potential classes a representative integer score of the probability of the input data belonging to the potential class, the first vector corresponding to one possible vector, each possible vector of the first set associating with each of the plurality of potential classes an integer score; (b) construction, from the first vector, of a second classification vector of the input data, such that the second vector also belongs to the first space of possible vectors and has a distance with the first vector according to a given distance function equal to a non-zero reference distance; and return of the second vector as result of the secure classification.

Various embodiments are provided for automating decision making for a neural architecture search by one or more processors in a computing system. One or more specifications may be automatically selected for a dataset, tasks, and one or more constraints for a neural architecture search. The neural architecture search may be performed based on the one or more specifications. A deep learning model may be suggested, predicted, and/or configured for the dataset, the tasks, and the one or more constraints based on the neural architecture search.

Methods, apparatus, systems, and articles of manufacture are disclosed to improve data training of a machine learning model using a field-programmable gate array (FPGA). An example system includes one or more computation modules, each of the one or more computation modules associated with a corresponding user, the one or more computation modules training first neural networks using data associated with the corresponding users, and FPGA to obtain a first set of parameters from each of the one or more computation modules, the first set of parameters associated with the first neural networks, configure a second neural network based on the first set of parameters, execute the second neural network to generate a second set of parameters, and transmit the second set of parameters to the first neural networks to update the first neural networks.

The present disclosure relates to systems, non-transitory computer-readable media, and methods for classifying an input image utilizing a classification model conditioned by a generative model and/or self-supervision. For example, the disclosed systems can utilize a generative model to generate a reconstructed image from an input image to be classified. In turn, the disclosed systems can combine the reconstructed image with the input image itself. Using the combination of the input image and the reconstructed image, the disclosed systems utilize a classification model to determine a classification for the input image. Furthermore, the disclosed systems can employ self-supervised learning to cause the classification model to learn discriminative features for better classifying images of both known classes and open-set categories.

Methods and systems for dynamic network link prediction include generating a dynamic graph embedding model for capturing temporal patterns of dynamic graphs, each of the graphs being an evolved representation of the dynamic network over time. The dynamic graph embedding model is configured as a neural network including nonlinear layers that learn structural patterns in the dynamic network. A dynamic graph embedding learning by the embedding model is achieved by optimizing a loss function that includes a weighting matrix for weighting reconstruction of observed edges higher than unobserved links. Graph edges representing network links at a future time step are predicted based on parameters of the neural network tuned by optimizing the loss function.

An example apparatus for semantic image segmentation includes a receiver to receive an image to be segmented. The apparatus also includes a gated dense pyramid network including a plurality of gated dense pyramid (GDP) blocks to be trained to generate semantic labels for respective pixels in the received image. The apparatus further includes a generator to generate a segmented image based on the generated semantic labels.