Abstract: Remote distribution of multiple neural network models to various client devices over a network can be implemented by identifying a native neural network and remotely converting the native neural network to a target neural network based on a given client device operating environment. The native neural network can be configured for execution using efficient parameters, and the target neural network can use less efficient but more precise parameters.

Abstract: Remote distribution of multiple neural network models to various client devices over a network can be implemented by identifying a native neural network and remotely converting the native neural network to a target neural network based on a given client device operating environment. The native neural network can be configured for execution using efficient parameters, and the target neural network can use less efficient but more precise parameters.

Abstract: Remote distribution of multiple neural network models to various client devices over a network can be implemented by identifying a native neural network and remotely converting the native neural network to a target neural network based on a given client device operating environment. The native neural network can be configured for execution using efficient parameters, and the target neural network can use less efficient but more precise parameters.

Abstract: The present invention relates to artificial neural networks, for example, deep neural networks. In particular, the present invention relates to a compression method for deep neural networks with proper use of mask and the device thereof. More specifically, the present invention relates to how to compress dense neural networks into sparse neural networks while maintaining or even improving the accuracy of the neural networks after compression.

Abstract: A neural network recognition method includes obtaining a first neural network that includes layers and a second neural network that includes a layer connected to the first neural network, actuating a processor to compute a first feature map from input data based on a layer of the first neural network, compute a second feature map from the input data based on the layer connected to the first neural network in the second neural network, and generate a recognition result based on the first neural network from an intermediate feature map computed by applying an element-wise operation to the first feature map and the second feature map.

Abstract: A neural network system is configured to receive an input image and to generate a classification output for the input image. The neural network system includes: a separable convolution subnetwork comprising a plurality of separable convolutional neural network layers arranged in a stack one after the other, in which each separable convolutional neural network layer is configured to: separately apply both a depthwise convolution and a pointwise convolution during processing of an input to the separable convolutional neural network layer to generate a layer output.

Abstract: A neural network is used to recognize characters from a character set. Based upon the character recognized, a smaller neural network is used for verification of the character recognized. The smaller neural network is trained to recognize only a single character of the set and provides a "yes" or "no" type verification of correct identification of the character.

Abstract: A camera system has a neural network for calibration of image distortion. The neural network learns the conversion from image coordinates with distortion to image coordinates with substantially reduced distortion, whereby the neural network provides image coordinates having substantially reduced distortion. In a learning process of the neural network, a relatively simple camera model is used to provide an instruction signal to the neural network according to sample data provided from the real camera.

Abstract: Hopfield and BAM neural network training or learning rules allowing memorization of a greater number of patterns. Successive over-relaxation is used in the learning rules based on the training patterns and the output vectors. Neural networks trained in this manner can better serve as the neural networks in a variety of pattern recognition and element correlation systems.

Abstract: Use is made of a neural network in order to restore a binary image to an original multi-level image, by way of example. Using the neural network makes it possible to raise the accuracy of restoration and the speed of processing.

Abstract: Systems and methods for sequence transcription with neural networks are provided. More particularly, a neural network can be implemented to map a plurality of training images received by the neural network into a probabilistic model of sequences comprising P(S|X) by maximizing log P(S|X) on the plurality of training images. X represents an input image and S represents an output sequence of characters for the input image. The trained neural network can process a received image containing characters associated with building numbers. The trained neural network can generate a predicted sequence of characters by processing the received image.

Abstract: Efficiently simulating an Amari dynamics of a neural field (a), the Amari dynamics being specified by the equation (1) where a(x,t) is the state of the neural field (a), represented in a spatial domain (SR) using coordinates x,t, i(x,i) is a function stating the input to the neural field at time t, f[.] is a bounded monotonic transfer function having values between 0 and 1, F(x) is an interaction kernel, s specifies the time scale on which the neural field (a) changes and h is a constant specifying the global excitation or inhibition of the neural field (a). A method for simulating an Amari dynamics of a neural field (a), comprising the step of simulating an application of the transfer function (f) to the neural field (a). According to the invention, the step of simulating an application of the transfer function (f) comprises smoothing the neural field (a) by applying a smoothing operator (S).

Abstract: Systems and methods for sequence transcription with neural networks are provided. More particularly, a neural network can be implemented to map a plurality of training images received by the neural network into a probabilistic model of sequences comprising P(S|X) by maximizing log P(S|X) on the plurality of training images. X represents an input image and S represents an output sequence of characters for the input image. The trained neural network can process a received image containing characters associated with building numbers. The trained neural network can generate a predicted sequence of characters by processing the received image.

Abstract: A neural network is trained to focus on a domain of interest. For example, in a pre-training phase, the neural network in trained using synthetic training data, which is configured to omit or limit content less relevant to the domain of interest, by updating parameters of the neural network to improve the accuracy of predictions. In a subsequent training phase, the pre-trained neural network is trained using real-world training data by updating only a first subset of the parameters associated with feature extraction, while a second subset of the parameters more associated with policies remains fixed.

Abstract: A system comprising an encoder neural network, a scene structure decoder neural network, and a motion decoder neural network. The encoder neural network is configured to: receive a first image and a second image; and process the first image and the second image to generate an encoded representation of the first image and the second image. The scene structure decoder neural network is configured to process the encoded representation to generate a structure output characterizing a structure of a scene depicted in the first image. The motion decoder neural network configured to process the encoded representation to generate a motion output characterizing motion between the first image and the second image.

Abstract: In a radiation image processing method utilizing a neural network, an image signal representing a radiation image is fed into a neural network, image processing is carried out on the image signal by the neural network, and an output representing the results of the image processing is obtained from the neural network. Image processing, with respect to the whole region of the radiation image, is carried out on the image signal by a first group of neurons of an intermediate layer of the neural network. Image processing, with respect to parts of the region of the radiation image, is carried out on the image signal by a second group of neurons of the intermediate layer of the neural network. Regardless of set values of initial conditions, the output of the neural network becomes converged to a global minimum corresponding to the stored information, and the results of operation obtained from the neural network are not trapped at a local minimum.

Abstract: A neural network is trained to focus on a domain of interest. For example, in a pre-training phase, the neural network in trained using synthetic training data, which is configured to omit or limit content less relevant to the domain of interest, by updating parameters of the neural network to improve the accuracy of predictions. In a subsequent training phase, the pre-trained neural network is trained using real-world training data by updating only a first subset of the parameters associated with feature extraction, while a second subset of the parameters more associated with policies remains fixed.

Abstract: The present invention relates to a convolutional-neural-network-based classifier, a classifying method by using a convolutional-neural-network-based classifier and a method for training the convolutional-neural-network-based classifier. The convolutional-neural-network-based classifier comprises: a plurality of feature map layers, at least one feature map in at least one of the plurality of feature map layers being divided into a plurality of regions; and a plurality of convolutional templates corresponding to the plurality of regions respectively, each of the convolutional templates being used for obtaining a response value of a neuron in the corresponding region.

Abstract: Pattern recognition, for instance optical character recognition, is achieved by training a neural network, scanning an image, segmenting the image to detect a pattern, preprocessing the detected pattern, and applying the preprocessed detected pattern to the trained neural network. The preprocessing includes determining a centroid of the pattern and centrally positioning the centroid in a frame containing the pattern. The training of the neural network includes randomly displacing template patterns within frames before applying the template patterns to the neural network.

Abstract: A system for automatically classification of human fingerprints. An unidentified fingerprint is processed to produce a direction map. The direction map is processed to generate a course direction map. The coarse direction map is input to a locally connected, highly constrained feed-forward neural network. The neural network has a highly structured architecture well-suited to exploit the rotational symmetries and asymmetries of human fingerprints. The neural network classifies the unidentified fingerprint into one of five classifications: Whorl, Double Loop, Left Loop, Right Arch and Arch.