62 results about "Physical computation" patented technology

A method and an apparatus that allocate one or more physical compute devices such as CPUs or GPUs attached to a host processing unit running an application for executing one or more threads of the application are described. The allocation may be based on data representing a processing capability requirement from the application for executing an executable in the one or more threads. A compute device identifier may be associated with the allocated physical compute devices to schedule and execute the executable in the one or more threads concurrently in one or more of the allocated physical compute devices concurrently.

A three-dimensional-image display system generates a first physical-calculation model generator that expresses a real object, based on both position/posture information expressing a position and posture of the real object, and attribute information expressing attribute of the real object. The three-dimensional-image display system displays a three-dimensional image within a display space, based on a calculation result of the interaction between the first physical-calculation model and a second physical-calculation model expressing a virtual external environment of the real object within the display space.

A computer system implemented a method for elastic resource management for executing a machine learning (ML) program. The system is configured to create a set of logical executors, assign them across a set of networked physical computation units of a distributed computing system, partition and distribute input data and Work Tasks across the set of logical executors, assign them across the set of networked physical computation units, where the Work Tasks are partitioned into short units of computation (micro-tasks), each calculates a partial update to the ML program's model parameters and each last for less than one second; create a set of logical servers (LSes); partition and distribute globally shared model parameters of the ML program across the set of logical servers; execute partitioned Work Tasks according to a bounded asynchronous parallel standard, where a current Work Task is allowed to execute with stale model parameters without having all the current calculation updates from Work Tasks it depend on, provided the staleness of the model parameters is within a predefined limit.

The invention discloses a method for physically deforming the skin of virtual human based on a hierarchical model, which comprises the following steps: (1) constructing the hierarchical model, namely, firstly processing an original model to generate a skeleton layer model used for controlling motions, a physical layer model used for physical deformation and a surface layer model used for final drawing, and establishing mapping relationships between a skeleton layer and a physical layer and between the physical layer and a surface layer; (2) producing the physical deformation, namely, after finishing constructing the hierarchical model, causing the skeleton layer to generate a motion trend and drive the physical layer to produce the physical deformation according to a mode of motion set by a user; and (3) driving the skin, namely, in the physical deformation process, driving the surface layer to motion and change along with the physical layer to generate the dynamic effect of the deformation of the skin according to the mapping relationship between the physical layer and the surface layer. In the method, on the basis of keeping the advantage of the sense of reality of the physical deformation and in combination with the characteristic of the data organization flexibility of the hierarchical model, an elaborate network used for drawing the skin of the virtual human and a simplified network used for deformation control and physical calculation are constructed respectively, which reduces calculation overhead on the premise of ensuring the precise deformation process of the virtual human and drawing reality effect.

An application executing on a rendering computer invokes a physics function request, e.g., to model the movement and interaction of objects to be rendered. The physics function request specifies a physics function to be performed on input data. Physics function request data is formatted for transmission over a network. The physics computer receives the physics function request data and performs an associated physics function using a physics GPU to generate physics computation result data. The physics computation result data is transmitted to the rendering computer over the network. A rendering GPU renders an image using the physics computation result data.

The invention provides a method for executing tasks parallelly on a heterogeneous multiprocessor, specifically a method for executing a plurality of executable programs parallelly on one or a plurality of physical computing device CPUs or GPUs. The dependency among the executable programs decides which one of the executable programs is chosen by the numerous threads running in a plurality of physical computing devices simultaneously to call and execute. If one thread is initialized by certain GPU in the physical computing device, but the GPU is busy in processing other image processing threads, then the thread can be initialized by another thread in the physical computing device. In order to execute a plurality of executable programs parallelly on the a plurality of physical computing devices, source codes and existing executable programs as API functions are all stored in an API library. The executed executable programs can be the existing executable programs as well as executable programs which are compiled on line from the source codes.

The invention discloses a neural network model calculation-oriented intermediate representation method and device. The method comprises the following steps of S1, analyzing an input model file to obtain topological structure information of a neural network; s2, constructing a logic calculation graph; s21, deducing physical layout information of each operator in the logic calculation graph; s22, deriving the element attribute of each operator in the logic calculation graph; s23, deducing description information of an input and output logic tensor of each operator in the logic calculation graph; s3, constructing a physical calculation graph; s31, generating a physical calculation graph; according to the meta-attribute-based intermediate representation for neural network model calculation disclosed by the invention, data parallelism, model parallelism and pipeline parallelism are originally supported from an operator level. According to the neural network model calculation-oriented intermediate representation method and device disclosed by the invention, the calculation expressions are taken as basic units, the tensors are taken as flowing data in the calculation graph formed by the whole calculation expressions, and the calculation process of the neural network model is realized in a composition manner.

The present disclosure relates to a method for implementing a stream processing computer architecture, including creating a stream computer processing (SCP) system by forming a super node cluster of processors representing physical computation nodes (''nodes''), communicatively coupling the processors via a local interconnection means (''interconnect''), and communicatively coupling the cluster to an optical circuit switch (OCS), via optical external links (''links''). The OCS is communicatively coupled to another cluster of processors via the links. The method also includes generating a stream computation graph including kernels and data streams, and mapping the graph to the SCP system, which includes assigning the kernels to the clusters and respective nodes, assigning data stream traffic between the kernels to the interconnection when the data stream is between nodes in the same cluster, and assigning traffic between the kernels to the links when the data stream is between nodes in different clusters. The method also includes configuring the OCSs to provide connectivity between mapped clusters.

Some embodiments provide a method for a compute manager that manages (i) virtual machines executing on host computers and (ii) physical computers. The method uses a first set of application programming interfaces (APIs) to communicate with a virtual machine (VM) executing on a host first computer via a hypervisor executing on the host first computer. The method uses the first set of APIs to communicate with a second computer via a smart network interface controller (NIC) of the second computer, wherein the smart NIC translates the first set of APIs into a different, second set of APIs for the second computer so that the compute manager manages the VM and the second computer with the same first set of APIs.

The invention discloses a method for verifying applicability of reactor physical computation programs. The method includes 1, analyzing sensitivity of existing critical experiment systems and novel nuclear reactor systems to obtain relative sensitivity coefficient vectors of effective multiplication coefficients; 2, computing correlation coefficients of the various critical experiment systems and the novel nuclear reactor systems and selecting the critical experiment systems meeting requirements of similarity limit; 3, computing relative adjustment amounts of multi-group cross sections to minimize integral deviation between computation results of Monte-Carlo programs for the effective multiplication coefficients of the selected critical experiment systems and measurement results; 4, adjusting computation results of Monte-Carlo programs for the effective multiplication coefficients of the novel nuclear reactor systems and utilizing the computation results as the optimal estimation values of the 'measurement results'. The relative sensitivity coefficient vectors of the effective multiplication coefficients are related to the multi-group cross sections. The method can be used for verifying the applicability of the to-be-verified nuclear reactor physical computation programs to designing and analyzing the novel nuclear reactor systems.

Aggregating and transforming data, and performing analytics thereupon, for application-specific optimization based on multiple data sources. The data is preferably ingressed automatically, and may originate from various public and/or private data sources. Data transformation preferably aligns the data aggregated from the various sources, to thereby allow meaningful referencing. Complex and non-aligned data can therefore be consolidated, such that it is readily digestible by simulation (or other) software. In an embodiment, risk of flooding for a supply chain is computed from the aggregated and transformed data, using data analytics based on physical computation for flood risk assessment, allowing the supply chain to be optimized with regard to threat of flooding and/or actual flooding. In another embodiment, risk of wild fire may be assessed. Other types of risk may also be assessed.

An exemplary surgical instrument tracking system includes at least one physical computing device that determines, based on endoscopic imagery of a surgical area and using a trained neural network, an observation for an object of interest depicted in the endoscopic imagery, associates, based on a probabilistic framework and kinematics of a robotically-manipulated surgical instrument located at the surgical area, the observation for the object of interest to the robotically-manipulated surgical instrument, and determines a physical position of the robotically-manipulated surgical instrument at the surgical area based on the kinematics of the robotically-manipulated surgical instrument and the observation associated with the robotically-manipulated surgical instrument.

The invention discloses an optimization method oriented to a task parallel programming model under a virtualization environment. Failing stealing operation in the task parallel programming model is obtained through a front-end monitoring part of a client virtual machine; the decision whether acceleration is executed is made through a rear-end acceleration part of a virtual machine monitor according to running states of an acceleration initiator and an accelerated candidate and information of physical CPUs where the acceleration initiator and the accelerated candidate are located. If acceleration is executed, remaining time slices of the acceleration initiator are provided for the accelerated candidate, and when the accelerated candidate is preempted by the virtual machine monitor (a time slice is used up or blocked), if the accelerated candidate is in the runnable state, an original scheduling path of the accelerated candidate is recovered. Optimization aiming at the virtualization environment is added to the existing task parallel programming model, waste caused by a virtual CPU of a thief thread for computational resources is reduced, scheduling delay for running of a virtual CPU of a useful thread is shortened, and physical computational resources are put into effective computation to the greatest extent.

Aggregating and transforming data, and performing analytics thereupon, for application-specific optimization based on multiple data sources. The data is preferably ingressed automatically, and may originate from various public and/or private data sources. Data transformation preferably aligns the data aggregated from the various sources, to thereby allow meaningful referencing. Complex and non-aligned data can therefore be consolidated, such that it is readily digestible by simulation (or other) software. In an embodiment, risk of flooding for a supply chain is computed from the aggregated and transformed data, using data analytics based on physical computation for flood risk assessment, allowing the supply chain to be optimized with regard to threat of flooding and/or actual flooding. In another embodiment, risk of wild fire may be assessed. Other types of risk may also be assessed.

Machine vision apparatus, methods and articles advantageously employ neural networks to parse or evaluate a three-dimensional environment which may, for example, be useful in robotics. Such may advantageously allow detection of objects (e.g., targets, obstacles, or even portions of the robot itself or neighboring robots) in a three-dimensional environment using limited physical computational resources, limited processing time, and/or with a high level of accuracy. Detection may include representation of a volume occupied by an object in the three-dimensional environment and/or a pose (i.e., position and orientation) of the object in the three-dimensional environment. Such may, for example, allow a robot to engage or otherwise interact with one or more target objects while avoiding obstacles in the three-dimensional environment, for instance while the robot operates autonomously and/or in real-time in the three-dimensional environment.

The present disclosure describes a sleep staging system. The sleep staging system comprising: one or more sensors configured to generate output signals conveying information related to one or more breathing parameters of subject; and one or more physical computer processors operatively connected with the one or more sensors, the one or more physical computer processors configured by computer readable instructions to: determine, based on the output signals, one or more breathing features of individual breaths of the subject; determine a distribution of the one or more breathing features over one or more time windows; detect presence of a breathing event based on the output signals; determine sleep states of the subject with a sleep stage classifier model based upon the distribution of the breathing features and the one or more breathing events; and provide feedback indicating the sleep states.

An image processing method includes the steps of: extracting a first two-dimensional feature point from a two-dimensional face image; on the basis of a geometric model of a standard face, deriving a three-dimensional feature point set including a first three-dimensional feature point corresponding to the first two-dimensional feature point; generating a three-dimensional face model including the three-dimensional feature point set; estimating an input pose of the two-dimensional face image to determine whether to update the three-dimensional feature point set; updating the three-dimensional feature point set by relocating the first three-dimensional feature point according to the first two-dimensional feature point, on the basis of the determination on whether to update the three-dimensional feature point set; and projecting the three-dimensional face model onto the two-dimensional face image. The present invention comprises a physical computer-readable recording medium in which a program for executing, in a computer, an image processing method is recorded.

The invention provides a neural network model calculation-oriented graph execution pipeline parallel method and device, and provides a neural network model calculation-oriented graph execution pipeline parallel method and device in a deep learning training system. Comprising a graph execution process in a neural network model-oriented calculation process and a process of cooperative work of each functional module. According to the neural network model calculation-oriented graph execution pipeline parallel method, graph execution bodies on a local machine are created according to a physical calculation graph compiled and generated by a deep learning framework, and a plurality of free memory blocks are distributed for each graph execution body. The whole calculation graph can participate in deep learning training tasks of different batches of data at the same time in a pipeline parallel mode, and the utilization rate of a memory and the parallel speed of the data are fully improved.