316 results about "Single task" patented technology

A framework for easy and accurate transcription of speech data is provided. Utterances related to a single task are grouped together and processed using combinations of associated sets of recognition results and / or context information in a manner that allows the same transcription for a selected recognition result to be assigned to each of the utterances under consideration.

The present invention discloses a multi-task learning convolutional neural network (CNN)-based face attribute analysis method. According to the method, based on a convolutional neural network, a multi-task learning method is adopted to carry out age estimation, gender identification and race classification on a face image simultaneously. In a traditional processing method, when face multi-attribute analysis is carried out, a plurality of times of calculation are required, and as a result, time can be wasted, and the generalization ability of a model is decreased. According to the method of the invention, three single-task networks are trained separately; the weight of a network with the lowest convergence speed is adopted to initialize the shared part of a multi-task network, and the independent parts of the multi-task network are initialized randomly; and the multi-task network is trained, so that a multi-task convolutional neural network (CNN) model can be obtained; and the trained multi-task convolutional neural network (CNN) model is adopted to carry out age, gender and race analysis on an inputted face image simultaneously, and therefore, time can be saved, and accuracy is high.

Application boundary decomposition and dynamic grouping of application components may provide a user with a task-oriented, user-configurable, dynamic user interface. An operating system may include various individual application components and a user may be able to organize the application components to create custom task groupings for performing specific tasks. The components of a single task grouping may be displayed within a single task frame or border. Alternatively, the components may be displayed upon on the user's desktop without such a border along with other desktop icons. Application components may be dynamically rearranged, either by the user or automatically in response to user context changes or according to a priority relationship among the application components. Additionally, multiple application components may access a single shared copy of data and task groupings may be shared among different users using different computer systems.

The invention relates to a scene and target identification method and device based on multi-task learning. The method comprises the steps that pictures containing different scenes and targets are collected as image sample data; the image sample data is subjected to manual label marking, and target class labels and scene class labels are obtained; a multi-layer convolutional neural network model is built, and network initialization is conducted; the image sample data and the corresponding target class labels are adopted for pre-training the built model till convergence, and a target identification model is obtained; based on a multi-task learning technology, network branches are added into a specific layer of the target identification model, random initialization is conducted, and a multi-task network is obtained; the image sample data and the corresponding scene class labels and target class labels are adopted for e-training the multi-task network till convergence, and a multi-task learning model is obtained; new image data is input to the multi-task learning model, and classification results of scene and target identification of images are obtained. Accordingly, the single task identification precision is improved.

A method of detecting tasks performed by users wherein a single task is a sequence of web URLs invocation. Task patterns are detected in web logs to identify tasks performed by users and analyze task trends over time, across corporate divisions and geographies. A grammar-based framework is used to model and detect tasks from web log patterns. The framework has two components: a declarative unit—to generate a task grammar, and a processing unit—to detect tasks from access logs by generating a state machine for applying the task grammar to the tokens associated with the access records. By analyzing user tasks, rather than just URLs, useful business information can be extracted.

Systems and methods for collaborative electronic communications are presented. One or more messages within a collaborative electronic communication platform, such as emails or chat messages, can be selected for further action. Messages such as chat messages within a chronological, multi-user chat stream can be selected for generation of an associated post content item. Content from the selected messages may be automatically inserted into the generated post. The post may be displayed together in a content group along with discussion or other further content contributions. Multiple posts and their content groups may be displayed in sequence by post date or date of last content contribution. Messages may also be selected for generation of an associated task item within a task management platform. In a chat platform, multiple non-adjacent messages may be selected for generation of a single task item. Generated posts or tasks may incorporate default information extracted from their associated messages.

The invention discloses a method of monitoring image mark quality and an apparatus thereof. The method comprises the following steps of executing an acquisition step and acquiring accuracy of mark data of a single task; issuing an image to be marked to a mark account, wherein the image to be marked includes a sample image of a predetermined proportion; acquiring manual mark data of the image to be marked, wherein the manual mark data includes current mark data of the sample image; acquiring an accurate specific value of the current mark data relative to standard mark data of the sample image; responding to the specific value which is greater than a pre-determined value; inputting the image to be marked into a picture identification model so as to acquire automatic mark data; comparing the manual mark data and the automatic mark data so as to acquire accuracy of the mark data of the single task; calculating a variance of the accuracy of mark data of multiple tasks and preset average mark accuracy; and based on the variance, determining the number of the sample images issued to the mark account and / or determining an image type suitable for marking of the mark account. Efficiency and accuracy of mark result detection are high.

The invention discloses a task processing method and apparatus, a task segmentation method and apparatus, a task processing system, and an electronic device, which are used for a distributed task scheduling system. After the distributed task scheduling system starts a to-be-executed task, the task processing method comprises the steps of allocating the to-be-executed task to a first task execution server in a task execution server cluster, and instructing the first task execution server to perform task segmentation processing on the to-be-executed task; and receiving at least one to-be-executed sub-task generated after segmentation of the to-be-executed task and returned by the first task execution server, and taking the at least one to-be-executed sub-task as a task needed to be executed by the task execution server cluster. By adopting the method provided by the invention, the problem that tasks processed by a single task execution server include excessive to-be-processed objects can be solved, so that the effects of improving the task execution efficiency and increasing the success rate are achieved.

The invention discloses a swarm robot control system and method based on visual positioning, and the system comprises a video collection positioning unit, an upper computer unit, and a swarm robot unit. The video collection positioning unit is used for collecting an image in a movement range of the swarm robot unit, building a coordinate system, recognizing the pose information of each robot in the swarm robot unit, and transmitting the pose information to the upper computer unit. The upper computer unit is used for generating control information for correcting the movement posture of the robots in real time, and transmitting the control information to the robots through a wireless sensing network. The swarm robot unit consists of a plurality of robots, and is used for receiving and analyzing the control information, adjusting a walking strategy, and completing the following control of a path track and a group dispatching task. The system and method can achieve the accurate coordinative control of the movement states of the swarm robots through the technology of visual positioning, and complete the execution of a single task and a group of tasks simply, conveniently and quickly.

The invention discloses a multimodal brain network feature fusion method based on multi-task learning, and the multimodal brain network feature fusion method based on the multi-task learning includes the steps of preprocessing the obtained functional magnetic resonance imaging (fMRI) images and diffusion tensor imaging (DTI) images, registrating the preprocessed fMRI image to the standard AAL template, carrying out a fiber tracking for preprocessed DTI images, calculating fiber anisotropy (FA) value, and constructing structure connection matrix through the AAL template. Clustering coefficient of each brain area in a function connection matrix and the structure connection matrix is calculated to be regarded as function features and structure features. As two different tasks, the function features and the structure features assess an optimal feature set by solving the problem of multi-task learning optimization. The method uses information with multiple modalities complementing each other to learn simultaneously and to classify, improves the classification accuracy, solves the problems that a single task feature does not consider the correlation between features, and the fact that only one modality feature is used for pattern classification can bring to insufficient amount of information.

The invention discloses a face analysis method and a face analysis system integrating a multi-task combined multi-scale convolution neural network, Firstly, the key region search algorithm is used toextract K face regions of interest with different scales from a picture of N *N size, which is used as the input of three channels of multi-scale CNN. Then, CNN is used to extract the features of theK regions of interest to obtain the features of faces with different scales, and the extracted features of faces with different scales are fused in a cascade manner to obtain the fused feature expression; Then, the loss functions of a plurality of tasks are fused to obtain a joint loss function, and the feature expression is used as a learning input to obtain an optimal solution of the joint lossfunction, so as to obtain a processing result of the plurality of tasks. The invention utilizes the correlation between tasks to promote each other, and improves the prediction accuracy rate of a single task.

Gaming rewards are provided as an incentive for performing business goals. A business goal can be comprised of one or more tasks and a threshold for each task. When one or more participants perform tasks which satisfies the corresponding thresholds, a corresponding gaming reward is awarded to the participants. A business goal may require a single participant to satisfy a single task threshold, multiple participants to satisfy a threshold for one or more tasks, or a single participant to satisfy one or more thresholds for several tasks.

A method and system for launching multiple applications simultaneously on a device under the control of application switching framework so that the operating system is only running one task for all the applications is provided. A single task is run under the control of an operating system. An application manager is run within the task. One or more applications are launched within the task under the control of the application manager. One of the applications is made the current application by switching, under user control, among the launched applications. A list of application descriptors is maintained for all the launched applications, and when switching, the application descriptor of one of the applications is used for displaying the application to a user on a screen. Each application descriptor contains forms of the launched applications. Each of the application descriptors contains a tree of forms with one root or parent form. A form represents an image to be displayed to the user. The image consists of text, pictures, bitmaps, or menus.

Gaming rewards are provided as an incentive for performing business goals. A business goal can be comprised of one or more tasks and a threshold for each task. When one or more participants perform tasks which satisfies the corresponding thresholds, a corresponding gaming reward is awarded to the participants. A business goal may require a single participant to satisfy a single task threshold, multiple participants to satisfy a threshold for one or more tasks, or a single participant to satisfy one or more thresholds for several tasks.

The embodiment of the invention provides a human face detection method and a human face detection device. The method comprises the following steps of extracting a plurality of face features of different hierarchical networks from a to-be-detected face image through a pre-trained convolutional neural network model so as to obtain a plurality of face feature vectors corresponding to different hierarchical networks; fusing the plurality of face feature vectors into a face feature vector; conducting dimensionality reduction treatment on the face feature vector after fusion treatment, and obtainingtwo face feature vectors in the same dimension; subjecting one face feature vector of the two face feature vectors to face detection treatment to obtain a face detection result; and subjecting the other face feature vector of the two face feature vectors to gesture estimation treatment to obtain a gesture estimation result. According to the invention, an image can be described more abundantly based on face detection features and gesture estimation features, and the accuracy is higher. The error rate of subsequent human face detection is reduced, and multiple related tasks can be executed at the same time. The performance of a single task is improved and the processing efficiency of the task is improved.

A storage system, a disk controller, a disk drive and a method of operating thereof. The method includes: configuring a disk drive in a manner enabling executing one or more read requests concurrently with executing one or more write requests addressed to the same data track of the disk drive; responsive to a received write request addressed to a certain track of the disk drive, identifying with the help of the control layer one or more read requests concurrent to received write request and addressed to the same track; if the received write request and the identified one or more read requests match a predefined criterion, generating and issuing, with the help of the control layer, a command to the disk drive for executing a single task corresponding to the concurrent read and write requests combined in accordance with a certain mask.

A multi-task machine learning method is performed to generate a multi-task (MT) predictor for a set of tasks including at least two tasks. The machine learning method includes: learning a multi-task decision tree (MT-DT) including learning decision rules for nodes of the MT-DT that optimize an aggregate information gain (IG) that aggregates single-task IG values for tasks of the set of tasks; and constructing the MT predictor based on the learned MT-DT. In some embodiments the aggregate IG is the largest single-task IG value of the single-task IG values. In some embodiments the machine learning method includes repeating the MT-DT learning operation for different subsets of a training set to generate a set of learned MT-DT's, and the constructing comprises constructing the MT predictor as a weighted combination of outputs of the set of MT-DT's.

The invention discloses a method of planning three dimensional route of an unmanned plane by means of an improved artificial fish swarm algorithm. The method of planning three dimensional route of an unmanned plane by means of an improved artificial fish swarm algorithm is used for carrying out static state planning of route and real-time dynamic re-planning of route when an unmanned plane executes a single task. The method of planning three dimensional route of an unmanned plane by means of an improved artificial fish swarm algorithm includes the steps: constructing a digital map through a landform model and a simplified threat model, considering the influence of space division granularity on the complexity of an optimizing control algorithm, and realizing division of space according to a fence self-adaptive algorithm; realizing static state route planning by means of an improved artificial fish swarm algorithm; and considering the time factor, constructing a threat prediction model based on a dynamic Bayesian network, predicting the unexpected threat, combined with flight constraint of the unmanned plane, obtaining the re-planning starting point, and realizing global route dynamic re-planning by means of the improved artificial fish swarm algorithm. The method of planning three dimensional route of an unmanned plane by means of an improved artificial fish swarm algorithm has the advantages of reducing the complexity of a route optimizing control algorithm, improving the optimum route searching capability, and satisfying the practical route planning demand.

The invention relates to a class case recommendation method based on text content. The method is divided into a pre-training part and a fine adjustment part. The pre-training part adopts a transformerencoder as a main structure, a Chinese language model is trained, Chinese language knowledge is learned from other corpora, and a high-quality language model is obtained. A triad model is used as a framework of the fine adjustment part, a preprocessed judicial document is used as training data, more knowledge about judgment is learned from the judicial field, and a better text vector representation is obtained. Compared with a traditional keyword-based class case recommendation method and a single-task neural network-based class case recommendation method, the content-based class case recommendation method provided by the invention is better in effect, and has better robustness based on a semantic training model, which indicates that the method provided by the invention is effective and practical.

The invention discloses a workflow mechanism-based concurrent ETL (Extract, Transform and Load) conversion method. By using a workflow technology and a multi-thread concurrent technology, concurrent execution of a plurality of ETL tasks of an ETL workflow and concurrent execution of a plurality of ETL behaviors in a single task are realized. When a plurality of ETL workflows are executed simultaneously and more parallel branches are available in the ETL workflows and the ETL operation, the execution efficiency can be obviously increased. At the same time, according to the method, cluster distribution processing is constructed, and parallel ETL data extraction engines are constructed by a parallel pipeline technology, so that the extraction efficiency of data can be greatly increased, and the parallel processing problem of multiple data flows and the bottleneck problem of conversion processing are solved.

The invention discloses a D2D communication collaboration based task unloading algorithm. The algorithm includes steps of S1, initiating all tasks to execute locally, and calculating energy consumption and execution time delay Ti(0) of the tasks based on a local execution mode; S2, in an MEC environment, calculating the energy and the execution time delay of a single task in a specific execution mode; S3, in the MEC environment, comparing the energy consumption and the execution time delay of the single task based on the specific execution mode with the energy consumption and the execution time delay of the single task based on the local execution mode and calculating the corresponding energy consumption-execution time delay ratio according to a comparison result; S4, selecting the optimalexecution mode of the single task in the MEC environment according t the energy consumption-execution time delay ratio; S5, executing S2-S4 in circulation until the optimal execution modes of all thetasks in the MEC environment are obtained. According to the invention, through determining the execution mode of each task and through adding cooperative communication and cooperative calculation, multiple selectable execution modes are provided for a device, so that transmission time delay is reduced effectively and device energy consumption is reduced.

The invention discloses a task monitoring method and device for a task scheduling server. An embodiment of the method includes: acquiring an unexecuted task and a self-defined task related to the same from a storer; selecting a task execution server from a task execution server cluster, and distributing the unexecuted task to the task execution server; monitoring an execution state of the unexecuted task through receiving a message; when the execution state meets a preset condition, distributing the self-defined task related to the unexecuted task to the task execution server. By the embodiment, task jamming caused by faults of a single task execution server is avoided, so that task execution reliability is improved; a user is enabled to customize tasks, so that high customizability is realized.

Power equipment widely used in commercial, residential and recreational settings comprises a host implement and a demountable motor. The host implement has a working piece dedicated to a single task. The host implement has a frame, a mounting block secured to the frame, a driven shaft connected to the frame with a working piece connected to one terminus and a self-locating and locking coupler on the other terminus. The demountable motor has a drive shaft with a second self-locating and locking coupler on its terminus. When the demountable motor is positioned on the host implement, the respective couplers mechanically mate so that a power from the demountable motor can be transferred to the working piece. The demountable motor is capable of being used on any one of several different host implements, all designed for performing different work tasks.

The invention provides a method for realizing medical image multi-task auxiliary diagnosis based on deep learning. On the condition that a focus position is known, multi-task auxiliary diagnosis of afocus is realized, such as focus area classification, segmentation, meridian regression, etc. The method comprises main steps of extracting medical image focus area data, constructing a multi-label dataset which corresponds with multi-task auxiliary diagnosis; building a single-model multi-task deep learning network, and realizing multi-task model preliminary training; adjusting a learning strategy, selecting a certain single task for training until convergence; fixing the trained single task and a trunk coding network parameter, and performing one-by-one training of other branch tasks. The method can settle a problem of single function of a medical image algorithm and simultaneously realizes multi-task auxiliary diagnosis of the focus by means of single-model single-input. Furthermore, the coding network in the characteristic extracting process is shared in training different tasks, thereby realizing mutual supplementing, and realizing performance which is not lower than or even exceed a single task in different branch tasks.

The invention provides a data export method and a data export device. The method comprises the following steps of: receiving the export conditions and export configuration information of data to be exported, wherein the export configuration information comprises a maximum thread number of a single task and the single record count of each thread; generating an export task according to the export conditions, the maximum thread number and the single record counts; and exporting the data to be exported according to the export task. Compared with the prior art, the method and the device of the embodiment of the invention solve the problem that the response time of single-thread operation is overlong under the condition of the large amount of data, and improve the export efficiency of a large amount of data.

The invention provides a face beauty prediction method based on multi-task transfer learning, and the method comprises the steps: building a multi-task face database, carrying out the feature learning, carrying out the feature fusion, and building a face beauty prediction model. The method improves the accuracy of face beauty prediction through improving the expression recognition and age recognition. In order to avoid over-fitting of a deep network trained by a small amount of sample data and insufficient computing equipment, a VGG, ResNet and GoogleNet backbone deep convolutional network isused as a shared feature learning network structure, model migration is used, and the trained convolutional network is used to train a migratable shared feature. Network parameters are shared among tasks in the training process, and shared characteristics are learned, so that the accuracy of single task learning of the network is improved. Through using the deep learning network for multi-task learning, the shared representation layer can enable the tasks with universality to be better combined with the correlation information, and the task specific layer can independently model the task specific information.

The invention discloses a single-task and multi-core scheduling method based on critical path and task duplication. A conventional multi-core task scheduling algorithm cannot perform effective scheduling when inner cores of a processor are insufficient. The method comprises the following steps that a directed acyclic graph (DAG) task graph processing module duplicates a fork node in a DAG task graph to a subsequent task node by using a task duplication method, forms a join structure graph, and further converts the join graph into a product processing tree; a task node scheduling allocation module introduces a key path idea in integrated scheduling, searches a key path of the product processing tree, and schedules nodes in the key path preferentially to start executing the nodes in the key path in advance to the greatest extent; and a scheduling sequence adjustment optimization module combines scheduling sequences in a mode of combining the scheduling sequences with the maximum similarity, so as to make the quantity of the scheduling sequences not greater than that of the inner cores of the processor, so the inner cores of the processor can fully process in parallel. The method is applicable to single-task and multi-core scheduling based on homogeneous multi-core processors which are interconnected on chip.

The invention relates to a scientific work flow scheduling method with safe perception under a cloud calculation environment. The method comprises the following steps of: firstly, calculating idle time of a single task in a scientific work flow task according to calculating time of the single task in a scientific work flow and data transmission time among the tasks; randomly setting an allowable safety service of the idle time for the single task in the scientific work flow according to a dependence relation among a safety service model, the idle time and the task; adding an expenditure brought about by the single task safety service in the scientific work flow into predication of task executing time, and adding an improvement on the calculation of the task expenditure into an MCP (Mixed Complementary Problem) algorithm; and finally, mapping a resource for the single task in the scientific work flow according to a resource condition in a data center of the cloud calculation environment and utilizing the resource to establish a virtual machine and execute the task. According to the scientific work flow scheduling method disclosed by the invention, the safety service grade of a whole body is improved under the precondition of not influencing the scheduling performance of the scientific work flow, and the safety risk of deploying under the cloud calculation environment is reduced.