Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a convolutional neural network using a regularization scheme. One of the methods includes repeatedly performing the following operations: obtaining a kernel of a particular convolutional layer; applying a Fourier transform to the kernel; generating a decomposition using singular-value decomposition (SVD); generating a regularized diagonal matrix; generating a recomposition; applying an inverse Fourier transform to the recomposition; and training the convolutional neural network on training inputs.

Abstract: A method of clustering spatial data includes receiving a point cloud comprised of a plurality of points defined within three-dimensional (3D) space. The method further includes selecting one or more adaptable clustering parameters and traversing each of the plurality of points in the point cloud and selectively adding each of the points to one or more clusters based on the selected clustering parameters associated with each point.

Abstract: A method for identifying regions of interest (ROIs) includes receiving, by a processor from a video camera, a video image and computing, by the processor, an optical flow image, based on the video image. The method also includes computing, by the processor, a magnitude of optical flow image based on the video image and computing a histogram of optical flow magnitudes (HOFM) image for the video image based on the magnitude of optical flow image. Additionally, the method includes generating, by the processor, a mask indicating ROIs of the video image, based on the HOFM.

Abstract: Aspects of the present invention provide methods, apparatuses, systems, computing devices, computing entities, and/or the like for identifying data processing activities associated with various data assets based on data discovery results. In accordance various aspects, a method is provided comprising: identifying and scanning data assets to detect a subset of the data assets, wherein each asset of the subset is associated with a particular data element used for target data; generating a prediction for each pair of data assets of the subset on the target data flowing between the pair; identifying a data flow for the target data based on the prediction generated for each pair; and identifying a data processing activity associated with handling the target data based on a correlation identified for the particular data element, the subset, and/or the data flow with a known data element, subset, and/or data flow for the data processing activity.

Abstract: Techniques are generally described for automatic scoring of alt-text for image data. In various examples, first image data and first text data describing the first image data may be received. A feature representation of the first image data may be determined using an encoder machine learning model. A hidden state representation may be determined using a decoder machine learning model based on the feature representation and a first word of the first text data. In some examples, a first score may be determined using the hidden state representation. The first score may include an indication of a descriptive capability of the first text data with respect to the first image data.

Abstract: An image processing apparatus includes a main unit configured to convert an input image into information of a first feature amount using a first convolutional neural network having at least two layers, an input unit configured to convert the input image into information of a second feature amount using a second convolutional neural network, and an output unit configured to convert information of a third feature amount generated by adding the information of the first feature amount and the information of the second feature amount to each other, into an output image using a third convolutional neural network.

Abstract: The invention refers to an image extraction method importing into SDL model in the field of information processing, and is characterized in that the target image are labeled artificially for plural times on computer data. The target image that has been labeled to be extracted for plural times will obtain the maximum probability value and scale of each parameter constituting the target image by machine learning. And the target image can be obtained from the sample computer data according to the maximum probability value or its scale range. The implementation effect of this method is to extract the required image arbitrarily from an image, eliminate the interference of background image affecting the result of image recognition, and improve the effect of image processing and the accuracy of image recognition, which is a new image processing algorithm that subverts the traditional binaryzation algorithm.

Abstract: Systems and methods for performing video understanding and analysis. Sets of feature maps for high resolution images and low resolution images in a time sequence of images are combined into combined sets of feature maps each having N feature maps. A time sequence of temporally aggregated sets of feature maps is created for each combined set of feature maps by: selecting a selected combined set of feature maps corresponding to an image at time “t” in the time sequence of images; applying, by channel-wise multiplication, a feature map weighting vector to a number of combined sets of feature maps that are temporally adjacent to the selected combined set of feature maps; and summing elements of the number of combined set of feature maps into a temporally aggregated set of feature maps. The time sequence of temporally aggregated sets of feature maps is processed to perform video understanding processing.

Abstract: An estimation 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 first point cloud data; generate, from the first point cloud data, second point cloud data in which an attention point and at least one observation point are combined, the attention point gaining attention as a target of attribute estimation; and estimate an attribute of the attention point by calculating, for each attribute, a belonging probability of belonging to the attribute by using the second point cloud data.

Abstract: A real-time detection method and apparatus for DGA domain name. An original domain name is translated into a multi-dimensional numeric vector, the multi-dimensional numeric vector is input into a deep learning model pre-trained based on an ImageNet data set, to generate a domain name feature, a domain name classifier is trained based on the generated domain name feature, and a DGA domain name is classified and predicted based on the domain name classifier obtained by training. The method firstly uses a deep learning model pre-trained based on an ImageNet data set, from the field of visual image classification and detection, for real-time detection of a DGA domain name, avoiding the process of high-intensity training and parameter weight adjustment for the deep learning model in DGA domain name detection. The detection rate is higher, and detection speed is faster.

Abstract: A system, the system comprising processing circuitry configured to: obtain an ordinal decision-tree classifier generated by taking into account an ordinality of classes of an ordinal class variable having at least three classes; and classify an input record to a given class of the classes using the ordinal decision-tree classifier.

Abstract: Novel techniques for pooling the available transmit power of a beam across the subcarriers that are or that are scheduled to be in use (and not across all available subcarriers) are disclosed. The scheduled subcarriers may be located in the same or different carriers of a modulation transmitter modulation system, and the pooled transmit power may be allocated or distributed across the scheduled subcarriers of the beam. Modulation symbols or resource elements may be transmitted in accordance with allocated, per-subcarrier power budgets, thereby maximizing the SNIR of signals that are transmitted in the beam via the scheduled subcarriers. Additionally, the allocation of the pooled transmit power to various subcarriers may continuously and/or dynamically vary over time, e.g., based on traffic demands, interference characteristics, etc., as well as based on subsequent scheduling of subcarriers to transmit subsequent modulation symbols or resource elements.

Abstract: A technique making use of a few-shot model to determine graphical features present in an image based on a small set of examples with known graphical features. Where a support set including a number of images that each have a known combination of graphical features, the image recognition can identify unknown combinations of those graphical features in any number of query images. In an embodiment of the present disclosure examples of a filled-out form are used to interpret any number of additional filled out versions of the form.

Abstract: In various examples, a deep neural network (DNN) is trained to accurately predict, in deployment, distances to objects and obstacles using image data alone. The DNN may be trained with ground truth data that is generated and encoded using sensor data from any number of depth predicting sensors, such as, without limitation, RADAR sensors, LIDAR sensors, and/or SONAR sensors. Camera adaptation algorithms may be used in various embodiments to adapt the DNN for use with image data generated by cameras with varying parameters—such as varying fields of view. In some examples, a post-processing safety bounds operation may be executed on the predictions of the DNN to ensure that the predictions fall within a safety-permissible range.

Abstract: According to one embodiment, a file transmission/reception device includes a communication direction managing unit and an application unit. The communication direction managing unit, in near field communication, cuts off a connection with an opposing device in a case where a conflict occurs with the opposing device, and, after being reconnected to the opposing device, switches the file transmission/reception device to any one mode of a master mode and a slave mode. The application unit performs transmission, reception, or transmission/reception of a file between the opposing device and the file transmission/reception device in the master mode or the slave mode in accordance with a mode specified by the communication direction managing unit.

Abstract: A method and a device for transmitting reference signals in a wireless communication system are disclosed. For these, a sequence is acquired to be used for the reference signals, and the reference signals are transmitted through subframes comprising a first type subframe and a second type subframe. Here, the first type subframe includes a first number of OFDM symbols and the second type subframe includes a second number of OFDM symbols. And, a first position of OFDM symbol for transmitting the reference signals at the first type subframe is the same as a second position of OFDM symbol for transmitting the reference signals at the second type subframe.

Abstract: Multi-radio antenna apparatuses and stations for wireless networks including multiple radios coupled to a single transmit/receive antenna, in which the antenna is highly synchronized by an external (e.g., GPS) signal. These multi-radio antenna systems may provide highly resilient links. Synchronization may allow these apparatuses to organically scale the transmission throughput while preventing data loss. The single transmit/receive antenna may have a single dish or a compound (e.g., a single pair of separate transmitting and receiving dishes) and connections for two or more radios.

Abstract: Methods, systems, and computer program products are provided for stroke attribute matrices. User input strokes may be converted into attributes encoded in one or more stroke attribute matrices (SAMs), such as bitmaps, for image or other multidimensional analysis. One or more convolutional neural networks (CNNs) may recognize letters, symbols, shapes and gestures in SAMs. A selector may select output classifications from among multiple CNNs. A sequence analyzer may select a sequence of selected CNN outputs. Stroke information may comprise, for example, velocity (e.g. direction and speed), tilt, pressure, line width, pen up/down events, hover height, etc. Stroke information may be stored, for example, in bitmap color channels (e.g. to facilitate human review). For example, an x, y velocity vector and x, y tilt may be encoded, respectively, as RGBA components of pixel data. Stroke crossings may be encoded, for example, by combining attribute values at pixels where strokes intersect.

Abstract: Multi-dimensional data can be mapped to a projection shape and converted for image analysis. In some examples, the multi-dimensional data may include data captured by a LIDAR system for use in conjunction with a perception system for an autonomous vehicle. Converting operations can include converting three-dimensional LIDAR data to multi-channel two-dimensional data. Data points of the multi-dimensional data can be mapped to a projection shape, such as a sphere. Characteristics of the projection shape may include a shape, a field of view, a resolution, and a projection type. After data is mapped to the projection shape, the projection shape can be converted to a multi-channel, two-dimensional image. Image segmentation and classification may be performed on the two-dimensional data. Further, segmentation information may be used to segment the three-dimensional LIDAR data, while a rendering plane may be positioned relative to the segmented data to perform classification on a per-object basis.

Abstract: A computer system and method for extending parallelized asynchronous reinforcement learning to include agent modeling for training a neural network is described. Coordinated operation of plurality of hardware processors or threads is utilized such that each functions as a worker process that is configured to simultaneously interact with a target computing environment for local gradient computation based on a loss determination mechanism and to update global network parameters. The loss determination mechanism includes at least a policy loss term (actor), a value loss term (critic), and a supervised cross entropy loss. Variations are described further where the neural network is adapted to include a latent space to track agent policy features.