Abstract: According to an embodiment, a learning method of optimizing a neural network, includes updating and specifying. In the updating, each of a plurality of weight coefficients included in the neural network is updated so that an objective function obtained by adding a basic loss function and an L2 regularization term multiplied by a regularization strength is minimized. In the specifying, an inactive node and an inactive channel are specified among a plurality of nodes and a plurality of channels included in the neural network.

Abstract: A learning device includes a structure search unit that searches for a first learned model structure obtained by selecting search space information in accordance with a target constraint condition of target hardware for each of a plurality of convolution processing blocks included in a base model structure in a neural network model; a parameter search unit that searches for a learning parameter of the neural network model in accordance with the target constraint condition; and a pruning unit that deletes a unit of at least one of the plurality of convolution processing blocks in the first learned model structure based on the target constraint condition and generates a second learned model structure.

Abstract: A learning device according to an embodiment includes one or more hardware processors configured to function as a generation unit, an inference unit, and a training unit. The generation unit generates input data with which an error between a value output from each of one or more target nodes and a preset aimed value is equal to or less than a preset value, the target nodes being in a target layer of a plurality of layers included in a first neural network. The inference unit causes the input data to propagate in a forward direction of the first neural network to generate output data. The training unit trains a second neural network differing from the first neural network by using training data including a set of the input data and the output data.

Abstract: According to an embodiment, an information processing apparatus includes a computing unit and a compressing unit. The computing unit is configured to execute computation of an input layer, a hidden layer, and an output layer of a neural network. The compressing unit is configured to irreversibly compress output data of at least a part of the input layer, the hidden layer, and the output layer and output the compressed data.

Abstract: According to an embodiment, a learning device includes one or more hardware processors configured to function as a structure search unit. The structure search unit searches for a first learned model structure. The first learned model structure is obtained by selecting search space information in accordance with a target constraint condition of target hardware for each of a plurality of convolution processing blocks included in a base model structure in a neural network model.

Abstract: A learning device according to an embodiment includes one or more hardware processors configured to function as a generation unit, an inference unit, and a training unit. The generation unit generates input data with which an error between a value output from each of one or more target nodes and a preset aimed value is equal to or less than a preset value, the target nodes being in a target layer of a plurality of layers included in a first neural network. The inference unit causes the input data to propagate in a forward direction of the first neural network to generate output data. The training unit trains a second neural network differing from the first neural network by using training data including a set of the input data and the output data.

Abstract: An inference apparatus according to an embodiment of the present disclosure includes a memory and a hardware processor coupled to the memory. The hardware processor is configured to: acquire at least one control parameter of second machine learning model, the second machine learning model having a size smaller than a size of a first machine learning model input to the inference apparatus; change the first machine learning model to the second machine learning model based on the at least one control parameter; and perform inference in response to input data by using the second machine learning model.

Abstract: According to an embodiment, a learning method of optimizing a neural network, includes updating and specifying. In the updating, each of a plurality of weight coefficients included in the neural network is updated so that an objective function obtained by adding a basic loss function and an L2 regularization term multiplied by a regularization strength is minimized. In the specifying, an inactive node and an inactive channel are specified among a plurality of nodes and a plurality of channels included in the neural network.

Abstract: According to an embodiment, a video compression apparatus includes a first compressor, a second compressor, a partitioner and a communicator. The first compressor compresses a first video to generate a first bitstream. The second compressor sets regions in a second video and compresses the regions so as to enable each region to be independently decoded, to generate a second bitstream. The partitioner partitions the second bitstream according to the set regions to obtain a partitioned second bitstream. The communicator receives region information indicating a specific region that corresponds to one or more regions and selects and transmits a bitstream corresponding to the specific region from the partitioned second bitstream.

Abstract: According to an embodiment, an information processing apparatus includes a computing unit and a compressing unit. The computing unit is configured to execute computation of an input layer, a hidden layer, and an output layer of a neural network. The compressing unit is configured to irreversibly compress output data of at least a part of the input layer, the hidden layer, and the output layer and output the compressed data.

Abstract: According to an embodiment, a network coefficient compression method includes: outputting, with respect to input data input into an input layer of a learned neural network, an output value in a hidden layer or an output layer of the neural network; and generating a compressed network coefficient by learning a network coefficient of the neural network, with the input data and the output value as training data, while performing lossy compression of the network coefficient.

Abstract: A data compressing device according to an embodiment includes a data cutting unit configured to divide continuously inputted data into W-bit data blocks and to output the data blocks in segments such that each of the segments is composed of N data blocks, and a compression-method determining unit configured to select, as a compression portion for each of the segments, a run length system, a flag system, or no compression, according to a ratio of data blocks of specific data in any of the segments. The data compressing device further includes an RL compression unit configured to execute, on any of the segments, a run length system of storing a consecutive amount of the specific data into compressed data, and a flag compression unit configured to execute, on any of the segments, a flag system of storing positional information of the specific data into compressed data.

Abstract: According to one embodiment, a route generation apparatus includes a memory and a circuit coupled with the memory. The circuit acquires a depth image regarding a capturing object including a first object, generates three-dimensional data by using the depth image receives first region information that specifies a first region including at least part of the first object based on the three-dimensional data, and generates route data by using the first region information and the three-dimensional data.

Abstract: A data compressing device according to an embodiment includes a data cutting unit configured to divide continuously inputted data into W-bit data blocks and to output the data blocks in segments such that each of the segments is composed of N data blocks, and a compression-method determining unit configured to select, as a compression portion for each of the segments, a run length system, a flag system, or no compression, according to a ratio of data blocks of specific data in any of the segments. The data compressing device further includes an RL compression unit configured to execute, on any of the segments, a run length system of storing a consecutive amount of the specific data into compressed data, and a flag compression unit configured to execute, on any of the segments, a flag system of storing positional information of the specific data into compressed data.

Abstract: A medical image processing apparatus according to an embodiment includes compressing circuitry. The compressing circuitry is configured to compress, for each of a first data group and a second data group both of which being included in data pertaining to a medical image, each of the first data group and the second data group separately, starting from data corresponding to a boundary between the first data group and the second data group and shifting sequentially to a direction away from the boundary.

Abstract: A medical image processing apparatus according to an embodiment includes compressing circuitry. The compressing circuitry is configured to compress, for each of a first data group and a second data group both of which being included in data pertaining to a medical image, each of the first data group and the second data group separately, starting from data corresponding to a boundary between the first data group and the second data group and shifting sequentially to a direction away from the boundary.

Abstract: According to an embodiment, a video compression apparatus includes a first compressor, a second compressor, a partitioner and a communicator. The first compressor compresses a first video to generate a first bitstream. The second compressor sets regions in a second video and compresses the regions so as to enable each region to be independently decoded, to generate a second bitstream. The partitioner partitions the second bitstream according to the set regions to obtain a partitioned second bitstream. The communicator receives region information indicating a specific region that corresponds to one or more regions and selects and transmits a bitstream corresponding to the specific region from the partitioned second bitstream.

Abstract: A medical image processing apparatus according to an embodiment includes compressing circuitry. The compressing circuitry is configured to compress, for each of a first data group and a second data group both of which being included in data pertaining to a medical image, each of the first data group and the second data group separately, starting from data corresponding to a boundary between the first data group and the second data group and shifting sequentially to a direction away from the boundary.

Abstract: According to an embodiment, an encoding apparatus includes a processor and a memory. The memory stores processor-executable instructions that, when executed by the processor, cause the processor to: divide an image included in an image group into a plurality of regions; calculate a priority for each of the regions on the basis of levels of importance of the regions; determine an order of processing for the regions on the basis of the corresponding priority; and encode the regions according to the determined order of processing.

Abstract: According to an embodiment, a multi-view image encoding device encodes a multi-view image including a plurality of viewpoint images. The device includes an assignor, a predictor, a subtractor, and an encoder. The assignor assigns reference image numbers to the reference images according to a number of reference images used in predicting already-encoded blocks obtained by dividing the viewpoint images. The predictor generates a prediction image with respect to an encoding target block obtained by dividing the viewpoint images by referring to the reference images. The subtractor calculates a residual error between an encoding target image and the prediction image. The encoder encodes: a coefficient of transformation which is obtained by performing orthogonal transformation and quantization with respect to the residual error; and the reference image numbers of the reference images used in generating the prediction image.