National Institute of Advanced Industrial Science and Technology

About: National Institute of Advanced Industrial Science and Technology is a government organization based out in Tsukuba, Ibaraki, Japan. It is known for research contribution in the topics: Catalysis & Thin film. The organization has 22114 authors who have published 65856 publications receiving 1669827 citations. The organization is also known as: Sangyō Gijutsu Sōgō Kenkyū-sho.

Physics of Liquid Crystals

Abstract: Over the 100 years since its discovery, liquid crystals have been the intriguing subject for both academia and industries. The textbook of de Gennes The Physics of Liquid Crystals published in 1974 is still the bible for many LC researchers, but new subjects unmentioned in the book have also risen for these years. This chapter describes the story of the recent developments and the future perspectives in physics of liquid crystals, especially focusing on the contributions by Japanese research groups for the last decade.

Atomic layers of hybridized boron nitride and graphene domains

Abstract: (1) Department of Mechanical Engineering and Materials Science, Rice University, Houston, TX 77005, United States

Shape-engineerable and highly densely packed single-walled carbon nanotubes and their application as super-capacitor electrodes

Abstract: Shape-engineerable and highly densely packed single-walled carbon nanotubes and their application as super-capacitor electrodes

Sinking deltas due to human activities

Abstract: Many of the world's deltas are densely populated and intensively farmed. An assessment of recent publications indicates that the majority of these deltas have been subject to intense flooding over the past decade, and that this threat will grow as global sea-level rises and as the deltas subside. Many of the world's largest deltas are densely populated and heavily farmed. Yet many of their inhabitants are becoming increasingly vulnerable to flooding and conversions of their land to open ocean. The vulnerability is a result of sediment compaction from the removal of oil, gas and water from the delta's underlying sediments, the trapping of sediment in reservoirs upstream and floodplain engineering in combination with rising global sea level. Here we present an assessment of 33 deltas chosen to represent the world's deltas. We find that in the past decade, 85% of the deltas experienced severe flooding, resulting in the temporary submergence of 260,000 km2. We conservatively estimate that the delta surface area vulnerable to flooding could increase by 50% under the current projected values for sea-level rise in the twenty-first century. This figure could increase if the capture of sediment upstream persists and continues to prevent the growth and buffering of the deltas.

Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet

Abstract: The purpose of this study is to determine whether current video datasets have sufficient data for training very deep convolutional neural networks (CNNs) with spatio-temporal three-dimensional (3D) kernels. Recently, the performance levels of 3D CNNs in the field of action recognition have improved significantly. However, to date, conventional research has only explored relatively shallow 3D architectures. We examine the architectures of various 3D CNNs from relatively shallow to very deep ones on current video datasets. Based on the results of those experiments, the following conclusions could be obtained: (i) ResNet-18 training resulted in significant overfitting for UCF-101, HMDB-51, and ActivityNet but not for Kinetics. (ii) The Kinetics dataset has sufficient data for training of deep 3D CNNs, and enables training of up to 152 ResNets layers, interestingly similar to 2D ResNets on ImageNet. ResNeXt-101 achieved 78.4% average accuracy on the Kinetics test set. (iii) Kinetics pretrained simple 3D architectures outperforms complex 2D architectures, and the pretrained ResNeXt-101 achieved 94.5% and 70.2% on UCF-101 and HMDB-51, respectively. The use of 2D CNNs trained on ImageNet has produced significant progress in various tasks in image. We believe that using deep 3D CNNs together with Kinetics will retrace the successful history of 2D CNNs and ImageNet, and stimulate advances in computer vision for videos. The codes and pretrained models used in this study are publicly available1.