Search NASA⌕ Search

Engineering topics

Chen, B Y

Publications and source records attributed to Chen, B Y.

Fast neural network training on a cluster of GPUs for action recognition with high accuracy

In this work, we propose algorithms and techniques to accelerate training of deep neural networks for action recognition on a cluster of GPUs. The convergence analysis of our algorithm shows it is possible to reduce communication cost and at the same time minimize the number of iterations needed for convergence. We customize the Adam optimizer for our distributed algorithm to improve efficiency. In addition, we employ transfer-learning to further reduce training time while improving validation accuracy. For the UCF101 and HMDB51 datasets, the validation accuracies achieved are 93.1% and 67.9% respectively. With an additional end-to-end trained temporal stream, the validation accuracies achieved for UCF101 and HMDB51 are 93.47% and 81.24% respectively. As far as we know, these are the highest accuracies achieved with the two-stream approach using ResNet that does not involve computationally expensive 3D convolutions or pretraining on much larger datasets.

97 MATHEMATICS AND COMPUTING↗

FY10 Engineering Innovations, Research and Technology Report

This report summarizes key research, development, and technology advancements in Lawrence Livermore National Laboratory's Engineering Directorate for FY2010. These efforts exemplify Engineering's nearly 60-year history of developing and applying the technology innovations needed for the Laboratory's national security missions, and embody Engineering's mission to ''Enable program success today and ensure the Laboratory's vitality tomorrow.'' Leading off the report is a section featuring compelling engineering innovations. These innovations range from advanced hydrogen storage that enables clean vehicles, to new nuclear material detection technologies, to a landmine detection system using ultra-wideband ground-penetrating radar. Many have been recognized with R&D Magazine's prestigious R&D 100 Award; all are examples of the forward-looking application of innovative engineering to pressing national problems and challenging customer requirements. Engineering's capability development strategy includes both fundamental research and technology development. Engineering research creates the competencies of the future where discovery-class groundwork is required. Our technology development (or reduction to practice) efforts enable many of the research breakthroughs across the Laboratory to translate from the world of basic research to the national security missions of the Laboratory. This portfolio approach produces new and advanced technological capabilities, and is a unique component of the value proposition of the Lawrence Livermore Laboratory. The balance of the report highlights this work in research and technology, organized into thematic technical areas: Computational Engineering; Micro/Nano-Devices and Structures; Measurement Technologies; Engineering Systems for Knowledge Discovery; and Energy Manipulation. Our investments in these areas serve not only known programmatic requirements of today and tomorrow, but also anticipate the breakthrough engineering innovations that will be needed in the future.

42 ENGINEERING↗

Fabrication of shape controlled Fe{sub 3}O{sub 4} nanostructure

Shape-controlled Fe{sub 3}O{sub 4} nanostructure has been successfully prepared using polyethylene glycol as template in a water system at room temperature. Different morphologies of Fe{sub 3}O{sub 4} nanostructures, including spherical, cubic, rod-like, and dendritic nanostructure, were obtained by carefully controlling the concentration of the Fe{sup 3+}, Fe{sup 2+}, and the molecular weight of the polyethylene glycol. Transmission Electron Microscope images, X-ray powder diffraction patterns and magnetic properties were used to characterize the final product. This easy procedure for Fe{sub 3}O{sub 4} nanostructure fabrication offers the possibility of a generalized approach to the production of single and complex nanocrystalline oxide with tunable morphology.

77 NANOSCIENCE AND NANOTECHNOLOGY↗

Building Ultra-Low False Alarm Rate Support Vector Classifier Ensembles Using Random Subspaces

This paper presents the Cost-Sensitive Random Subspace Support Vector Classifier (CS-RS-SVC), a new learning algorithm that combines random subspace sampling and bagging with Cost-Sensitive Support Vector Classifiers to more effectively address detection applications burdened by unequal misclassification requirements. When compared to its conventional, non-cost-sensitive counterpart on a two-class signal detection application, random subspace sampling is shown to very effectively leverage the additional flexibility offered by the Cost-Sensitive Support Vector Classifier, yielding a more than four-fold increase in the detection rate at a false alarm rate (FAR) of zero. Moreover, the CS-RS-SVC is shown to be fairly robust to constraints on the feature subspace dimensionality, enabling reductions in computation time of up to 82% with minimal performance degradation.

42 ENGINEERING↗