29results about "Data sorting" patented technology

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for finding splits in gradient boosting systems. One example method includes identifying, based on multi-party computation (MPC), a first split among possible first splits in a first layer of a decision tree. Identifying the first split comprises determining a first derivative and a second derivative for each sample in a dataset of n samples and m features; generating a sorting array indicating a sorting order of at least one feature; identifying the first split based on first derivatives, second derivatives and the sorting array; and generating an index vector that indicates whether each sample belongs to a first or a second sub-dataset. The method further includes identifying a second split among possible second splits in a second layer of the decision tree based on the first derivatives, the second derivatives, the sorting array, and the index vector.

An example includes receiving a read request specifying a target range and a target checkpoint, and executing a first query to retrieve a first entry of a metadata table. The first entry is a most recent entry of those entries that are older than the target checkpoint and record write ranges overlapping the target range. The example also includes, if the write range in the first entry does not cover all of the target range, determining a remaining portion of the target range that excludes the write range in the first entry. The example also includes executing a second query to retrieve a second entry of the metadata table, where the second entry is a most recent record in a second subset of entries that are older that the target checkpoint and record write ranges overlapping the remaining portion of the target location range.

The present technology includes outlier detection circuitry, which can be employed to filter out outlier values so that they do not negatively impact statistical analysis or processing tasks. The outlier detection circuitry can receive data as a stream of multiple sets with multiple values contained within one set. The outlier detection circuitry comprises dedicated hardware and / or software resources that determines the largest and smallest values that occur within each set in the received stream.

The present application provides an alarm method and device. The method includes: in response to a received alarm request, acquiring target data from a target database according to the alarm request, and determining a target processing condition and target alarm conditions according to the alarm request, where the target database includes vehicle data and road section data of at least one transportation subsystem; performing analysis and processing on the target data according to the target processing condition to obtain traffic flow information corresponding to a target road section, where the target road section is at least one road section indicated by the road section data; in response to determining that the traffic flow information meets a set of target alarm conditions, giving an alarm.

Processes and devices for finding signals within a frequency range include converting an analog input signal to a digital domain using an analog-to-digital converter to create a digital input signal. A first fast Fourier transform (FFT) converts the digital input signal to a frequency input signal, which is binned based on frequency and a threshold to create a binary filter. The digital input signal (from the analog-to-digital converter) is sorted into containers (e.g., using the Chinese remainder theorem) and then run through a second FFT to convert the digital input signal to a frequency intermediate signal. The frequency intermediate signal is sorted into bins and multiplied using the binary filter to create an output frequency signal.

An operation accelerator for processing floating-point operations using an integer operator, according to one aspect, comprises: a first exponent decoder configured to determine whether an input data encoded with an exponent is an outlier based on a bias bit of the input data, and to perform pre-alignment processing that shifts a mantissa of the input data based on the bias bit; a second exponent decoder configured to determine whether weight data encoded with an exponent is an outlier based on a bias bit of the weight data, and to perform pre-alignment processing that shifts a mantissa of the weight data based on the bias bit; and a computation unit configured to perform multiplication and accumulation operations in integer arithmetic on the mantissa of the input data output from the first exponent decoder and the mantissa of the weight data output from the second exponent decoder.

The present technology includes outlier detection switching logic that can be used to filter out outlier values ​​so that they do not negatively affect statistical analysis or processing tasks. The outlier detection switching logic can receive data as a stream of multiple sets, with each set containing several values. The outlier detection switching logic includes dedicated hardware and / or software resources that determine the largest and smallest values ​​occurring within each set in the received stream.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for finding splits in gradient boosting systems. One example method includes identifying, based on multi-party computation (MPC), a first split among possible first splits in a first layer of a decision tree. Identifying the first split comprises determining a first derivative and a second derivative for each sample in a dataset of n samples and m features; generating a sorting array indicating a sorting order of at least one feature; identifying the first split based on first derivatives, second derivatives and the sorting array; and generating an index vector that indicates whether each sample belongs to a first or a second sub-dataset. The method further includes identifying a second split among possible second splits in a second layer of the decision tree based on the first derivatives, the second derivatives, the sorting array, and the index vector.

A computing system is disclosed in which a tensor-network structure mutates autonomously during execution. A structural mutation unit modifies tensor-network topology concur- rently with tensor contraction execution without halting computation or requiring host processor intervention.

Methods, network traffic management apparatuses, non-transitory computer readable media, and systems that optimize transport of unknown and changing unstructured log data in a network environment. The method includes retrieving one or more entries associated with target unstructured log data; structuring the one or more entries into sorted structured batches comprising structured data and sorted unstructured batches without structured data, based on a current configuration and producing current batch metadata; converting the sorted structured batches into columnar-oriented representation; compressing the columnar-oriented batches; evaluating the compressed columnar-oriented batches by comparing a compression ratio of compressed sizes of the sorted structured batches and the sorted unstructured batches using the current batch metadata thereby producing a compression rate; receiving the compression rate and the current batch metadata; and determining an optimized configuration for the retrieved one or more entries; and repeating the above steps using the optimized configuration.

The application discloses a deterministic unary sequence template construction method. A threshold set is used as template content to write into a memory array to form a first ordered threshold template with static threshold rows, so that accurate conversion from binary to unary sequence is realized through the first ordered threshold template, and the ordered threshold template is a deterministic unary sequence template. A fixed permutation is performed on the threshold positions of the first ordered threshold template to optimize the threshold position distribution, and a second ordered threshold template is obtained, so that the effective bits of the unary sequence after conversion from binary to unary sequence through the ordered threshold template are more uniformly dispersed, and the second ordered threshold template is a permuted deterministic unary sequence template.

The present disclosure discloses a method, apparatus and device for generating a model and a storage medium. A method can include: acquiring sample resource features and sample labels; determining a first screening factor according to the sample resource features and the sample labels, and determining first resource features from the sample resource features according to the first screening factor; determining a second screening factor, and determining second resource features from the first resource features based on the second screening factor, and obtaining features of a target model based on the second resource features; and training a machine learning model, by taking the features of the target model as an input of the target model, and taking the sample labels corresponding to the features of the target model as an output of the target model, to obtain the trained target model.

This invention provides a method and system for optimizing the movement and sorting of graphic elements, relating to the field of graphics processing technology. The method includes: acquiring target graphic elements and sorting instructions; using a list of all graphic elements in the current canvas as a reference list; establishing a mapping table between graphic element identifiers and position indices based on the reference list; determining the original position indices of the target graphic element and related sub-graphic elements in the reference list according to the mapping table; determining the target position of the target graphic element in the reference list according to the sorting instructions, the type of the target graphic element, and the original position index; updating the arrangement order of the target graphic element and related sub-graphic elements in the reference list according to the target position; and outputting a reference graphic element list after optimization of the target graphic element sorting based on the updated reference list.

Methods, systems, and devices for data management are described. A query response system may receive a user query and query, via data stores storing a corpus of documents, for candidate documents associated with the user query. The query response system may sort the candidate documents based on respective semantic similarities between each candidate document and the user query. The query response system may select a threshold semantic similarity within a range of semantic similarities associated with the sorted candidate documents, where the threshold semantic similarity is selected based on gradient values between adjacent similarities within the range of semantic similarities. The threshold semantic similarity may define a subset of documents to be input to a retrieval augmented response model. The query response system may provide, to the retrieval augmented response model, the user query and the subset of the documents and receive a response to the user query.

The application discloses a data sorting device and method based on a fixed structure sorting network, wherein the method comprises a cache unit, a multiple selection logic unit, a data exchange network and multiple sorting networks; the multiple selection logic unit is used for receiving multiple selection control information, selecting a data matrix output by the sorting network or a data matrix stored in the cache unit based on the multiple selection control information, and transmitting the selected data matrix to the data exchange network; the data exchange network is used for performing data exchange and data rearrangement on the data matrix based on exchange network control information; and the sorting network is used for calling a cut sorting function to alternately perform row sorting or column sorting on the data matrix to obtain a sorted data vector. The application can realize efficient sorting processing on different sizes of input data, and improves the universality, scalability and hardware resource utilization efficiency of the sorting network.

A method and system for vector retrieval acceleration includes at least one calculation module and at least one sorting acceleration module. Each calculation module includes a first receiving end and a second receiving end for receiving query vector blocks and center vector blocks respectively, and performs operations on the received query vector blocks and center vector blocks, so as to obtain distance calculation results between each query vector and all center vectors, sorting acceleration module. The at least one calculation module corresponds one-to-one with the at least one sorting acceleration module. Each sorting acceleration module receives the distance calculation results output by the corresponding calculation module, sorts the distance calculation results, so as to obtain a sorting intermediate state, and then stores sorting intermediate states into a system memory.

The present application relates to a water turbine unit equipment rotation method based on group representation theory and rotation voting algorithm, comprising: establishing a permutation group having an isomorphic relationship with the unit equipment rotation logic; constructing a linear space and a base vector corresponding to the permutation group; determining the state and level of the unit equipment; determining the conjugate class of the permutation group and the corresponding matrix group element; determining the trigger condition of the unit equipment rotation; and calculating the new order of the main and standby equipment in a descending order k-rotation manner. The equipment rotation voting algorithm of the present application realizes the scientific and reasonable rotation and standby of the water turbine unit equipment, and the number of equipment used for rotation is not limited, improving the reliability of the water turbine unit operation, and solving the problems of algorithm complexity, program logic redundancy, poor readability, and difficult to find and troubleshoot logical errors in the design of the equipment rotation logic algorithm relying on experience and enumeration method.

Described are a system, method, and computer program product for efficiently joining time-series data tables. The method includes loading a first table and a second table into a memory and generating a set of first key-value pairs based on a set of first time-series records and a set of second key-value pairs based on a set of second time-series records. The method also includes sorting the set of first key-value pairs and the set of second key-value pairs. The method further includes interleaving the set of first key-value pairs with the set of second key-value pairs and sequentially matching the sets of time-series records to form a joined table. The method further includes, in response to matching each respective second time-series record with the respective first time-series record, removing the respective second time-series record from the at least one memory.

This application discloses an optimization method, device, medium, and product for a sparse matrix-vector multiplication algorithm, relating to the fields of high-performance computing and sparse computing. The method includes: setting multiple preset candidate block shapes; sequentially calculating the score of each candidate block shape; selecting the candidate block shape with the highest score as the optimal block shape; rearranging multiple data rows based on the number of non-zero elements in all data rows; performing block processing on the rearranged target sparse matrix based on the optimal block shape; allocating the computation task of the processed block-shaped target sparse matrix to multiple thread bundles for parallel execution; for each data row: if the number of corresponding non-zero elements exceeds a preset threshold, then splitting the data row into multiple subtasks according to the preset threshold and allocating them to different thread bundles for processing; avoiding performance bottlenecks caused by some thread bundles processing excessive amounts of data, thereby ensuring that GPU parallel computing resources are utilized efficiently and in a balanced manner.

This invention discloses a dual BFS hardware accelerator for sparse matrix RCM reordering, belonging to the field of sparse matrix computation and hardware acceleration technology. The accelerator includes: a top-level control module, a BFS engine module employing a dual BFS positioning architecture, a degree table module, a neighbor sorter module with a degree-first comparison network, a single-cycle access marker bit vector, and a hardware stack manager module. The BFS engine module achieves automatic pseudo-peripheral node positioning through a three-stage pipeline: the first stage starts BFS from the node with the lowest degree and finds the first farthest node through a hierarchy tracing circuit; the second stage starts BFS again from that node to determine the globally optimal starting point; and the third stage executes degree-first RCM traversal. An extended FIFO queue is used to transmit hierarchy information, and an external memory is interacted with through a decoupled interface. The hardware stack manager achieves single-cycle reverse output. This invention achieves a processing latency of less than 1.2ms on a 1000-node graph, reducing the average matrix bandwidth by 73.85%, and improving upon existing software by 31.85%.

Methods and apparatus for finding signals in a frequency range include converting an analog input signal to a digital domain using an analog-to-digital converter to create a digital input signal. A first fast Fourier transform (FFT) converts the digital input signal to a frequency input signal, which is binned based on frequency and a threshold to create a binary filter. The digital input signal (from the analog-to-digital converter) is classified into bins (e.g., using the Chinese remainder theorem), and then converted to a frequency intermediate signal by a second FFT. The frequency intermediate signal is classified into bins and multiplied using the binary filter to create an output frequency signal.

The techniques may include accessing a plurality of subsets of instances. In addition, the techniques may include identifying a net capacity change for each of the subsets of instances, the net capacity change of a respective subset of instances based at least in part on a capacity change of each event instance in the respective subset of instances. The techniques may include generating feature vectors for the subsets of instances. Moreover, the techniques may include adding the feature vectors to a set of training vectors in a data store. Also, the techniques may include training a machine learning model to output a predicted net capacity change for a particular entity by: accessing a subset of the set of training vectors that are associated with the particular entity; and training the machine learning model to output the predicted net capacity change for the particular entity at an input time period.

Systems and methods for determining priority levels to process vulnerabilities associated with a networked computer system can include a data collection engine receiving a plurality of specification profiles, each defining one or more specification variables of the networked computer system or a respective asset. The data collection engine can receive, from a vulnerability scanner, vulnerability data indicative of a vulnerability associated with the networked computer system. A profiling engine can interrogate a computing device of the networked computer system, and receive one or more respective profiling parameters from that computing device. A ranking engine can compute a priority ranking value of the computing device based on the profile specification variables, the vulnerability data and the profiling parameters. The priority ranking value associated with the computing device can be indicative of a priority level, compared to other computing devices of the computer network, for patching a vulnerability affecting that computing device.

A computer-implemented method for storing and querying application performance monitoring (APM) data in a database, includes receiving data records of APM data, each data record associated to a key; storing the data records in storage containers in the database, each record stored in one storage container and each storage container identified by a unique storage identifier; assigning a unique index value to each unique key; defining a distribution function mapping the keys to the index values; generating a secondary function to approximate the distribution function; defining a mapping function to map the index values to lists of storage identifiers; and in response to a request to retrieve data records for a specified key or keys, querying the database to identify a list of storage identifiers based on the secondary function and the mapping function and retrieving the data records for the specified key or keys.

In some aspects, the disclosure is directed to methods and systems for sorting, correlation, and / or matching of structured and unstructured data. In some implementations, metadata associated with structured data may be applied to unstructured data based on the results of the sort or correlation. Such sorting or matching may be performed through an efficient iterative process with progressive confidence scores, and use a priori knowledge from correlations between structured data values to identify potentially associated unstructured data values.