Academic literature on the topic 'CSS hardware acceleration'

Machine learning algorithms are gaining ground in high energy physics for applications in particle and event identification, physics analysis, detector reconstruction, simulation and trigger. Currently, most data-analysis tasks at LHC experiments benefit from the use of machine learning. Incorporating these computational tools in the experimental framework presents new challenges. This paper reports on the implementation of the end-to-end deep learning with the CMS software framework and the scaling of the end-to-end deep learning with multiple GPUs. The end-to-end deep learning technique combines deep learning algorithms and low-level detector representation for particle and event identification. We demonstrate the end-to-end implementation on a top quark benchmark and perform studies with various hardware architectures including single and multiple GPUs and Google TPU.

Realistic modeling and deformation of soft tissue is one of the key technologies of virtual surgery simulation which is a challenging research field that stimulates the development of new clinical applications such as the virtual surgery simulator. In this paper we adopt the linear FEM (Finite Element Method) and sparse matrix compression stored in CSR (Compressed Sparse Row) format that enables fast modeling and deformation of soft tissue on GPU hardware with NVIDIA’s CUSPARSE (Compute Unified Device Architecture Sparse Matrix) and CUBLAS (Compute Unified Device Architecture Basic Linear Algebra Subroutines) library. We focus on the CGS (Conjugate Gradient Solver) which is the mainly time-consuming part of FEM, and transplant it onto GPU with the two libraries mentioned above. The experimental results show that the accelerating method in this paper can achieve realistic and fast modeling and deformation simulation of soft tissue.

Strapdown Inertial navigation (SINS) is a highly reliable navigation system for short term applications. SINS functions continuously, less hardware failures, renders high speed navigation solutions ranging from 50 Hz to 1000 Hz and exhibits low short-term errors. It provides efficient attitude, angular rate, acceleration, velocity and position solutions. But, the accuracy of SINS solution vitiates with time as the sensor (gyros & accelerometers) errors are integrated through the navigation equations. Average navigation grade SINS are capable of providing effective stand-alone navigation for shorter duration (few minutes) applications Stand-alone SINS capable of providing solutions for applications exceeding 10 minutes duration, are generally highly expensive ($0.1M to $2.0M). To cope with this limitation, a cost effective solution is the integrated navigation system wherein the unboundedly growing errors of SINS are contained with the help of external non-inertial navigation aids like GPS, Celestial Navigation System (CNS), Odometer, Doppler radars etc. The efficient methodology for integrated or multi-sensory navigation is the Federated Kalman Filter (FKF) scheme. In FKF architecture, a reference SINS solution is integrated independently with each of the aiding navigation systems in a bank of local Kalman filters. There are a number of different ways in which the local filter outputs may be combined to produce an integrated navigation solution. The no-reset, fusion-reset, zero-reset, and cascaded versions of federated integration have been used by different researcher and navigators over the years. All different schemes of FKF have certain pros and cons. Fusion-reset method although nearly optimal is less fault tolerant while no-resent scheme renders highly fault tolerant solutions but with sub-optimal solutions and compromised precision. To enhance the fault tolerance ability of fusion-reset scheme of FKF, additional parameters called weighting factors are introduced to tune the contribution of each local filter in the final data fusion. The presented scheme has been found nearly optimal and expressively fault tolerant.