Methods, processing units and media for multi-bandwidth separated feature extraction convolution in a neural network are described. A convolution block splits input channels of an activation map into multiple branches, each branch undergoing convolution at a different bandwidth by using down-sampling of the inputs. The outputs are concatenated by up-sampling the outputs of the low-bandwidth branches using pixel shuffling. The concatenation operation may be a shuffled concatenation operation that preserves separated multi-bandwidth feature information for use by subsequent layers of the neural network. Embodiments are described which apply frequency-based and magnitude-based attention to the weights of the convolution kernels based on the frequency band locations of the weights.

The invention relates to a tiered storage system comprising tiers of data storage. The tiered storage system further comprises a processor; and a memory coupled to the processor. The memory comprises instructions which, when executed by the processor, cause the processor to: receive usage data descriptive of usage of memory extents stored by the tiered storage system; identify periodic usage patterns of the memory extents at least partially by calculating a correlation coefficient between the usage data and a predetermined list of conditions; determine a projected data usage for each of the memory extents using the periodic usage patterns, wherein the projected data usage is temporally dependent; sort the memory extents into usage bins according to the projected data usage; and control the tiers of data storage to migrate the at memory extents between the tiers of data storage using temporal changes of the sorting into the usage bins.

An image matching device includes a weight determination unit and a matching unit. Based on a synthesized frequency characteristic obtained by synthesizing a frequency characteristic of a first image with a frequency characteristic of a second image and an ideal synthesized frequency characteristic that is an ideal one obtained by synthesizing the frequency characteristic of the first image with the frequency characteristic of the second image, the weight determination unit determines a weight relating to frequency in performing matching of the first image and a third image. The matching unit performs matching of the first image and the third image based on the determined weight.

A system and a method are disclosed for forming an output feature map (OFM). Activation values in an input feature map (IFM) are selected and transformed on-the-fly into the Winograd domain. Elements in a Winograd filter is selected that respectively correspond to the transformed activation values. A transformed activation value is multiplied by a corresponding element of the Winograd filter to form a corresponding product value in the Winograd domain. Activation values are repeatedly selected, transformed and multiplied by a corresponding element in the Winograd filter to form corresponding product values in the Winograd domain until all activation values in the IFM have been transformed and multiplied by the corresponding element. The product values are summed in the Winograd domain to form elements of a feature map in the Winograd domain. The elements of the feature map in the Winograd domain are inverse-Winograd transformed on-the-fly to form the OFM.

A System-on-Chip (SoC) includes a first memory configured to store first data, a second memory, and a data processing circuit configured to divide the first data obtained from the first memory into a plurality of pieces of division data, assign a plurality of tags to the plurality of pieces of division data, each of the plurality of tags including a coordinate value for a corresponding piece of division data, obtain second data based on at least one of the first data and the plurality of tags for the plurality of pieces of division data, and provide an address and the second data to the second memory. The address and the second data are obtained based on the plurality of tags.

A method for determining an event periodic value includes: acquiring a time series of a target event, where the time series includes a preset number of sequential values; calculating an autocorrelation sequence of the time series, and based on peak values and trough values of the autocorrelation sequence, determining a first candidate periodic value set; calculating a Fourier transform result of the time series, and based on amplitude values of frequency points of the Fourier transform result, determining a second candidate periodic value set; acquiring an union of the first candidate periodic value set and the second candidate periodic value set, determining a total value of confidences corresponding to each candidate periodic value in the union, and based on the total value of confidences, determining the periodic value of the target event.

A usage prediction method executed by a computer, the usage prediction method includes classifying a plurality of records corresponding to a plurality of times included in first time-series data indicating a history of usages of a resource into a plurality of groups respectively corresponding to attributes of the plurality of times; generating second time-series data for each attribute by combining the records belonging to the group corresponding to the same attribute for the plurality of classified groups in order of the times; generating, for each attribute, an expression for calculating a predicted value to be used for calculating a predicted value of the usage based on the generated second time-series data; and calculating the predicted value of the usage based on the expression for calculating the predicted value for each attribute.

Methods, apparatuses, and systems for improving data mapping are provided. An example method may include retrieving a first plurality of data objects associated with a first database schema from a database, determining a first data classifier corresponding to the first database schema, generating a mapping specification based at least in part on the first data classifier and the first plurality of data objects, and generating a second plurality of data objects based at least in part on the first plurality of data objects and the mapping specification.

Devices and techniques are generally described for determining a recalibration frequency of a state space model. In various examples, a first hyperparameter for a first dataset may be determined. A residual value between a first data point of the first dataset and a machine learning model fitted to the first dataset may be determined. A plurality of second datasets may be generated based on the residual value. Second hyperparameters may be determined for the plurality of second datasets. A variability of the second hyperparameters may be determined. A third hyperparameter may be determined for a subset of the first dataset. A recalibration frequency may be determined for the machine learning model by comparing the third hyperparameter to the variability of the second hyperparameters.

The systems and methods discussed herein implement a volumetric approach to point cloud representation, compression, decompression, communication, or any suitable combination thereof. The volumetric approach can be used for both geometry and attribute compression and decompression, and both geometry and attributes can be represented by volumetric functions. To create a compressed representation of the geometry or attributes of a point cloud, a suitable set of volumetric functions are transformed, quantized, and entropy-coded. When decoded, the volumetric functions are sufficient to reconstruct the corresponding geometry or attributes of the point cloud.