-
RAO-SS: A Prototype of Run-time Auto-tuning Facility for Sparse Direct Solvers
Authors:
Takahiro Katagiri,
Yoshinori Ishii,
Hiroki Honda
Abstract:
In this paper, a run-time auto-tuning method for performance parameters according to input matrices is proposed. RAO-SS (Run-time Auto-tuning Optimizer for Sparse Solvers), which is a prototype of auto-tuning software using the proposed method, is also evaluated. The RAO-SS is implemented with the Autopilot, which is middle-ware to support run-time auto-tuning with fuzzy logic function. The target…
▽ More
In this paper, a run-time auto-tuning method for performance parameters according to input matrices is proposed. RAO-SS (Run-time Auto-tuning Optimizer for Sparse Solvers), which is a prototype of auto-tuning software using the proposed method, is also evaluated. The RAO-SS is implemented with the Autopilot, which is middle-ware to support run-time auto-tuning with fuzzy logic function. The target numerical library is the SuperLU, which is a sparse direct solver for linear equations. The result indicated that: (1) the speedup factors of 1.2 for average and 3.6 for maximum to default executions were obtained; (2) the software overhead of the Autopilot can be ignored in RAO-SS.
△ Less
Submitted 20 August, 2024;
originally announced August 2024.
-
From Coupled Oscillators to Graph Neural Networks: Reducing Over-smoothing via a Kuramoto Model-based Approach
Authors:
Tuan Nguyen,
Hirotada Honda,
Takashi Sano,
Vinh Nguyen,
Shugo Nakamura,
Tan M. Nguyen
Abstract:
We propose the Kuramoto Graph Neural Network (KuramotoGNN), a novel class of continuous-depth graph neural networks (GNNs) that employs the Kuramoto model to mitigate the over-smoothing phenomenon, in which node features in GNNs become indistinguishable as the number of layers increases. The Kuramoto model captures the synchronization behavior of non-linear coupled oscillators. Under the view of c…
▽ More
We propose the Kuramoto Graph Neural Network (KuramotoGNN), a novel class of continuous-depth graph neural networks (GNNs) that employs the Kuramoto model to mitigate the over-smoothing phenomenon, in which node features in GNNs become indistinguishable as the number of layers increases. The Kuramoto model captures the synchronization behavior of non-linear coupled oscillators. Under the view of coupled oscillators, we first show the connection between Kuramoto model and basic GNN and then over-smoothing phenomenon in GNNs can be interpreted as phase synchronization in Kuramoto model. The KuramotoGNN replaces this phase synchronization with frequency synchronization to prevent the node features from converging into each other while allowing the system to reach a stable synchronized state. We experimentally verify the advantages of the KuramotoGNN over the baseline GNNs and existing methods in reducing over-smoothing on various graph deep learning benchmark tasks.
△ Less
Submitted 5 March, 2024; v1 submitted 6 November, 2023;
originally announced November 2023.
-
CLRerNet: Improving Confidence of Lane Detection with LaneIoU
Authors:
Hiroto Honda,
Yusuke Uchida
Abstract:
Lane marker detection is a crucial component of the autonomous driving and driver assistance systems. Modern deep lane detection methods with row-based lane representation exhibit excellent performance on lane detection benchmarks. Through preliminary oracle experiments, we firstly disentangle the lane representation components to determine the direction of our approach. We show that correct lane…
▽ More
Lane marker detection is a crucial component of the autonomous driving and driver assistance systems. Modern deep lane detection methods with row-based lane representation exhibit excellent performance on lane detection benchmarks. Through preliminary oracle experiments, we firstly disentangle the lane representation components to determine the direction of our approach. We show that correct lane positions are already among the predictions of an existing row-based detector, and the confidence scores that accurately represent intersection-over-union (IoU) with ground truths are the most beneficial. Based on the finding, we propose LaneIoU that better correlates with the metric, by taking the local lane angles into consideration. We develop a novel detector coined CLRerNet featuring LaneIoU for the target assignment cost and loss functions aiming at the improved quality of confidence scores. Through careful and fair benchmark including cross validation, we demonstrate that CLRerNet outperforms the state-of-the-art by a large margin - enjoying F1 score of 81.43% compared with 80.47% of the existing method on CULane, and 86.47% compared with 86.10% on CurveLanes.
△ Less
Submitted 15 May, 2023;
originally announced May 2023.
-
End-to-End Monocular Vanishing Point Detection Exploiting Lane Annotations
Authors:
Hiroto Honda,
Motoki Kimura,
Takumi Karasawa,
Yusuke Uchida
Abstract:
Vanishing points (VPs) play a vital role in various computer vision tasks, especially for recognizing the 3D scenes from an image. In the real-world scenario of automobile applications, it is costly to manually obtain the external camera parameters when the camera is attached to the vehicle or the attachment is accidentally perturbed. In this paper we introduce a simple but effective end-to-end va…
▽ More
Vanishing points (VPs) play a vital role in various computer vision tasks, especially for recognizing the 3D scenes from an image. In the real-world scenario of automobile applications, it is costly to manually obtain the external camera parameters when the camera is attached to the vehicle or the attachment is accidentally perturbed. In this paper we introduce a simple but effective end-to-end vanishing point detection. By automatically calculating intersection of the extrapolated lane marker annotations, we obtain geometrically consistent VP labels and mitigate human annotation errors caused by manual VP labeling. With the calculated VP labels we train end-to-end VP Detector via heatmap estimation. The VP Detector realizes higher accuracy than the methods utilizing manual annotation or lane detection, paving the way for accurate online camera calibration.
△ Less
Submitted 31 August, 2021;
originally announced August 2021.
-
Leveraging Temporal Joint Depths for Improving 3D Human Pose Estimation in Video
Authors:
Naoki Kato,
Hiroto Honda,
Yusuke Uchida
Abstract:
The effectiveness of the approaches to predict 3D poses from 2D poses estimated in each frame of a video has been demonstrated for 3D human pose estimation. However, 2D poses without appearance information of persons have much ambiguity with respect to the joint depths. In this paper, we propose to estimate a 3D pose in each frame of a video and refine it considering temporal information. The prop…
▽ More
The effectiveness of the approaches to predict 3D poses from 2D poses estimated in each frame of a video has been demonstrated for 3D human pose estimation. However, 2D poses without appearance information of persons have much ambiguity with respect to the joint depths. In this paper, we propose to estimate a 3D pose in each frame of a video and refine it considering temporal information. The proposed approach reduces the ambiguity of the joint depths and improves the 3D pose estimation accuracy.
△ Less
Submitted 4 November, 2020;
originally announced November 2020.
-
Sensitivity of quantum PageRank
Authors:
Hirotada Honda
Abstract:
In this paper, we discuss the sensitivity of quantum PageRank. By using the finite dimensional perturbation theory, we estimate the change of the quantum PageRank under a small analytical perturbation on the Google matrix. In addition, we will show the way to estimate the lower bound of the convergence radius as well as the error bound of the finite sum in the expansion of the perturbed PageRank.
In this paper, we discuss the sensitivity of quantum PageRank. By using the finite dimensional perturbation theory, we estimate the change of the quantum PageRank under a small analytical perturbation on the Google matrix. In addition, we will show the way to estimate the lower bound of the convergence radius as well as the error bound of the finite sum in the expansion of the perturbed PageRank.
△ Less
Submitted 27 June, 2019;
originally announced July 2019.
-
Full-body High-resolution Anime Generation with Progressive Structure-conditional Generative Adversarial Networks
Authors:
Koichi Hamada,
Kentaro Tachibana,
Tianqi Li,
Hiroto Honda,
Yusuke Uchida
Abstract:
We propose Progressive Structure-conditional Generative Adversarial Networks (PSGAN), a new framework that can generate full-body and high-resolution character images based on structural information. Recent progress in generative adversarial networks with progressive training has made it possible to generate high-resolution images. However, existing approaches have limitations in achieving both hi…
▽ More
We propose Progressive Structure-conditional Generative Adversarial Networks (PSGAN), a new framework that can generate full-body and high-resolution character images based on structural information. Recent progress in generative adversarial networks with progressive training has made it possible to generate high-resolution images. However, existing approaches have limitations in achieving both high image quality and structural consistency at the same time. Our method tackles the limitations by progressively increasing the resolution of both generated images and structural conditions during training. In this paper, we empirically demonstrate the effectiveness of this method by showing the comparison with existing approaches and video generation results of diverse anime characters at 1024x1024 based on target pose sequences. We also create a novel dataset containing full-body 1024x1024 high-resolution images and exact 2D pose keypoints using Unity 3D Avatar models.
△ Less
Submitted 6 September, 2018;
originally announced September 2018.
-
Geometric Analysis of Observability of Target Object Shape Using Location-Unknown Distance Sensors
Authors:
Hiroshi Saito,
Hirotada Honda
Abstract:
We geometrically analyze the problem of estimating parameters related to the shape and size of a two-dimensional target object on the plane by using randomly distributed distance sensors whose locations are unknown. Based on the analysis using geometric probability, we discuss the observability of these parameters: which parameters we can estimate and what conditions are required to estimate them.…
▽ More
We geometrically analyze the problem of estimating parameters related to the shape and size of a two-dimensional target object on the plane by using randomly distributed distance sensors whose locations are unknown. Based on the analysis using geometric probability, we discuss the observability of these parameters: which parameters we can estimate and what conditions are required to estimate them. For a convex target object, its size and perimeter length are observable, and other parameters are not observable. For a general polygon target object, convexity in addition to its size and perimeter length is observable. Parameters related to a concave vertex can be observable when some conditions are satisfied. We also propose a method for estimating the convexity of a target object and the perimeter length of the target object.
△ Less
Submitted 15 May, 2017;
originally announced July 2017.
-
Anomalous infrared spectra of hybridized phonons in type-I clathrate Ba$_8$Ga$_{16}$Ge$_{30}$
Authors:
K. Iwamoto,
S. Kushibiki,
H. Honda,
S. Kajitani,
T. Mori,
H. Matsumoto,
N. Toyota,
K. Suekuni,
M. A. Avila,
T. Takabatake
Abstract:
The optical conductivity spectra of the rattling phonons in the clathrate Ba$_8$Ga$_{16}$Ge$_{30}$ are investigated in detail by use of the terahertz time-domain spectroscopy. The experiment has revealed that the lowest-lying vibrational mode of a Ba(2)$^{2+}$ ion consists of a sharp Lorentzian peak at 1.2 THz superimposed on a broad tail weighted in the lower frequency regime around 1.0 THz. With…
▽ More
The optical conductivity spectra of the rattling phonons in the clathrate Ba$_8$Ga$_{16}$Ge$_{30}$ are investigated in detail by use of the terahertz time-domain spectroscopy. The experiment has revealed that the lowest-lying vibrational mode of a Ba(2)$^{2+}$ ion consists of a sharp Lorentzian peak at 1.2 THz superimposed on a broad tail weighted in the lower frequency regime around 1.0 THz. With decreasing temperature, an unexpected linewidth broadening of the phonon peak is observed, together with monotonic softening of the phonon peak and the enhancement of the tail structure. These observed anomalies are discussed in terms of impurity scattering effects on the hybridized phonon system of rattling and acoustic phonons.
△ Less
Submitted 1 November, 2012; v1 submitted 31 May, 2012;
originally announced June 2012.
-
Low temperature line-width broadening in optical-conductivity spectra of the off-center rattling phonons in type-I clathrate Ba$_8$Ga$_{16}$Sn$_{30}$
Authors:
T. Mori,
K. Iwamoto,
S. Kushibiki,
H. Honda,
H. Matsumoto,
N. Toyota,
K. Suekuni,
M. A. Avila,
T. Takabatake
Abstract:
With a terahertz time-domain spectrometer (0.3 - 3.0 THz) we have measured the optical conductivity of the type-I clathrate Ba$_8$Ga$_{16}$Sn$_{30}$ at temperatures from 300 K down to 7 K. Independent six spectra superimposed on the Drude conductivity are identified to infrared active vibrational modes of guest Ba ions and the cages. While the spectra of five higher-frequency modes depend hardly o…
▽ More
With a terahertz time-domain spectrometer (0.3 - 3.0 THz) we have measured the optical conductivity of the type-I clathrate Ba$_8$Ga$_{16}$Sn$_{30}$ at temperatures from 300 K down to 7 K. Independent six spectra superimposed on the Drude conductivity are identified to infrared active vibrational modes of guest Ba ions and the cages. While the spectra of five higher-frequency modes depend hardly on temperature, the lowest-lying spectrum with a peak at 0.72 THz due to the Ba(2) ion's off-centering vibration in the oversized cage changes with temperature characteristically. With lowering temperature, the spectral shape of this so-called rattling phonon continues to become so broad that the line-width amounts to be comparable to the peak frequency. Furthermore, below about 100 K, the single broad peak tends to split into two subpeaks. While this splitting can be explained by assuming a multi-well anharmonic potential, the strong enhancement of the line-width broadening toward low temperature, cannot be understood, since the Boltzmann factor generally sharpens the low-temperature spectra.
△ Less
Submitted 23 July, 2010;
originally announced July 2010.
-
Multi-physics Extension of OpenFMO Framework
Authors:
Toshiya Takami,
Jun Maki,
Jun'ichi Ooba,
Yuuichi Inadomi,
Hiroaki Honda,
Ryutaro Susukita,
Koji Inoue,
Taizo Kobayashi,
Rie Nogita,
Mutsumi Aoyagi
Abstract:
OpenFMO framework, an open-source software (OSS) platform for Fragment Molecular Orbital (FMO) method, is extended to multi-physics simulations (MPS). After reviewing the several FMO implementations on distributed computer environments, the subsequent development planning corresponding to MPS is presented. It is discussed which should be selected as a scientific software, lightweight and reconfi…
▽ More
OpenFMO framework, an open-source software (OSS) platform for Fragment Molecular Orbital (FMO) method, is extended to multi-physics simulations (MPS). After reviewing the several FMO implementations on distributed computer environments, the subsequent development planning corresponding to MPS is presented. It is discussed which should be selected as a scientific software, lightweight and reconfigurable form or large and self-contained form.
△ Less
Submitted 18 July, 2007;
originally announced July 2007.
-
Open-architecture Implementation of Fragment Molecular Orbital Method for Peta-scale Computing
Authors:
Toshiya Takami,
Jun Maki,
Jun-ichi Ooba,
Yuichi Inadomi,
Hiroaki Honda,
Taizo Kobayashi,
Rie Nogita,
Mutsumi Aoyagi
Abstract:
We present our perspective and goals on highperformance computing for nanoscience in accordance with the global trend toward "peta-scale computing." After reviewing our results obtained through the grid-enabled version of the fragment molecular orbital method (FMO) on the grid testbed by the Japanese Grid Project, National Research Grid Initiative (NAREGI), we show that FMO is one of the best ca…
▽ More
We present our perspective and goals on highperformance computing for nanoscience in accordance with the global trend toward "peta-scale computing." After reviewing our results obtained through the grid-enabled version of the fragment molecular orbital method (FMO) on the grid testbed by the Japanese Grid Project, National Research Grid Initiative (NAREGI), we show that FMO is one of the best candidates for peta-scale applications by predicting its effective performance in peta-scale computers. Finally, we introduce our new project constructing a peta-scale application in an open-architecture implementation of FMO in order to realize both goals of highperformance in peta-scale computers and extendibility to multiphysics simulations.
△ Less
Submitted 10 January, 2007;
originally announced January 2007.
-
ASCA Observations of Temperature Structure and Metal Distribution in the Perseus Cluster of Galaxies
Authors:
H. Ezawa,
N. Y. Yamasaki,
T. Ohashi,
Y. Fukazawa,
M. Hirayama,
H. Honda,
T. Kamae,
K. Kikuchi,
R. Shibata
Abstract:
Large-scale distributions of hot-gas temperature and Fe abundance in the Perseus cluster have been studied with multi-pointing observations by the GIS instrument on ASCA. Within a radius of 20' from the cluster center, the energy spectra requires two temperature components, in which the cool component indicates kT ~ 2 keV and the hot-component temperature shows a significant decline from about 8…
▽ More
Large-scale distributions of hot-gas temperature and Fe abundance in the Perseus cluster have been studied with multi-pointing observations by the GIS instrument on ASCA. Within a radius of 20' from the cluster center, the energy spectra requires two temperature components, in which the cool component indicates kT ~ 2 keV and the hot-component temperature shows a significant decline from about 8 keV to 6 keV toward the center. In the outer region of the cluster, the temperature shows a fluctuation with an amplitude of about 2 keV and suggest that a western region at ~16' from the cluster center is relatively hotter. As for the Fe abundance, a significant decline with radius is detected from 0.44 solar at the center to ~0.1 solar at a 50' offset region. If observed Fe-K line intensity within 4' from the center is suppressed by a factor of 2 due to the resonance scattering effect, the corrected Fe mass density follows the galaxy distribution. Finally, our results do not support the large-scale velocity gradients previously reported from the same GIS data.
△ Less
Submitted 24 May, 2001; v1 submitted 15 May, 2001;
originally announced May 2001.
-
Magnetic Dipoles and Quantum Coherence in Muscle Contraction
Authors:
Kuniyuki Hatori,
Hajime Honda,
Koichiro Matsuno
Abstract:
An actin filament contacting myosin molecules as a functional unit of muscle contraction induces magnetic dipoles along the filament when ATP molecules to be hydrolyzed are available there. The induced magnetic dipoles are coherent over the entire filament, though they are fluctuating altogether as constantly being subject to the ambient thermal agitations.
An actin filament contacting myosin molecules as a functional unit of muscle contraction induces magnetic dipoles along the filament when ATP molecules to be hydrolyzed are available there. The induced magnetic dipoles are coherent over the entire filament, though they are fluctuating altogether as constantly being subject to the ambient thermal agitations.
△ Less
Submitted 9 April, 2001;
originally announced April 2001.
-
X-Ray Probing of the Central Regions of Clusters of Galaxies
Authors:
Kazuo Makishima,
Hajime Ezawa,
Yasushi Fukazawa,
Hirohiko Honda,
Yasushi Ikebe,
Tuneyoshi Kamae,
Ken'ichi Kikuchi,
Kyoko Matsushita,
Kazuhiro Nakazawa,
Takaya Ohashi,
Tadayuki Takahashi,
Takayuki Tamura,
Haiguang Xu
Abstract:
Results of ASCA X-ray study of central regions of medium-richness clusters of galaxies are summarized, emphasizing differences between cD and non-cD clusters. The intra-cluster medium (ICM) is likely to consist of two (hot and cool) phases within $\sim 100$ kpc of a cD galaxy, where the ICM metallicity is also enhanced. In contrast, the ICM in non-cD clusters appears to be isothermal with little…
▽ More
Results of ASCA X-ray study of central regions of medium-richness clusters of galaxies are summarized, emphasizing differences between cD and non-cD clusters. The intra-cluster medium (ICM) is likely to consist of two (hot and cool) phases within $\sim 100$ kpc of a cD galaxy, where the ICM metallicity is also enhanced. In contrast, the ICM in non-cD clusters appears to be isothermal with little metallicity gradient right to the center. The gravitational potential exhibits a hierarchical nesting around cD galaxies, while a total mass-density profile with a central cusp is indicated for a non-cD cluster Abell~1060. The iron-mass-to-light ratio of the ICM decreases toward the center in both types of clusters, although it is radially constant in peripheral regions. The silicon-to-iron abundance ratio in the ICM increases with the cluster richness, but remains close to the solar ratio around cD galaxies. These overall results are interpreted without appealing to the popular cooling-flow hypothesis. Instead, an emphasis is put on the halo-in-halo structure formed around cD galaxies.
△ Less
Submitted 4 April, 2001;
originally announced April 2001.
-
Temperature Variation in the Cluster of Galaxies Abell 115 Studied with ASCA
Authors:
R. Shibata,
H. Honda,
M. Ishida,
T. Ohashi,
K. Yamashita
Abstract:
Abell 115 exhibits two distinct peaks in the surface brightness distribution. ASCA observation shows a significant temperature variation in this cluster, confirmed by a hardness ratio analysis and spectral fits. A linking region between main and sub clusters shows a high temperature compared with other regions. Two possibilities are examined as the cause of the temperature variation: cooling flo…
▽ More
Abell 115 exhibits two distinct peaks in the surface brightness distribution. ASCA observation shows a significant temperature variation in this cluster, confirmed by a hardness ratio analysis and spectral fits. A linking region between main and sub clusters shows a high temperature compared with other regions. Two possibilities are examined as the cause of the temperature variation: cooling flows in the main cluster and a shock heating due to the collision of the subcluster into the main system. Spectral fits with cooling flow models to the main-cluster data show a mass-deposition rate less than 419 solar-mass/yr. Temperatures in the main cluster, the linking region, and the subcluster are estimated by correcting for the effects of X-ray telescope response as 4.9 (+0.7/-0.6), 11 (+12/-4), and 5.2 (+1.4/-1.0) keV, respectively. The high temperature in the linking region implies that Abell 115 is indeed a merger system, with possible contribution from cooling flows on the temperature structure.
△ Less
Submitted 22 June, 1999;
originally announced June 1999.
-
The Structure of the X-Ray Emitting Gas in the Hydra-A Cluster of Galaxies
Authors:
Y. Ikebe,
K. Makishima,
H. Ezawa,
Y. Fukazawa,
M. Hirayama,
H. Honda,
Y. Ishisaki,
K. Kikuchi,
H. Kubo,
T. Murakami,
T. Ohashi,
T. Takahashi,
K. Yamashita
Abstract:
The temperature and abundance structure in the intracluster medium (ICM) of the Hydra-A cluster of galaxies is studied with ASCA and ROSAT. The effect of the large extended outskirts in the point-spread function of the X-Ray Telescope on ASCA is included in this analysis. In the X-ray brightness profile, the strong central excess above a single beta-model, identified in the Einstein and ROSAT da…
▽ More
The temperature and abundance structure in the intracluster medium (ICM) of the Hydra-A cluster of galaxies is studied with ASCA and ROSAT. The effect of the large extended outskirts in the point-spread function of the X-Ray Telescope on ASCA is included in this analysis. In the X-ray brightness profile, the strong central excess above a single beta-model, identified in the Einstein and ROSAT data, is also found in the harder energy band (>4keV). A simultaneous fit of five annular spectra taken with the GIS instrument shows a radial distribution of the temperature and metal abundance. A significant central enhancement in the abundance distribution is found, while the temperature profile suggests that the ICM is approximately isothermal with the temperature of ~3.5keV. The ROSAT PSPC spectrum in the central 1'.5 region indicates a significantly lower temperature than the GIS result. A joint analysis of the GIS and PSPC data reveals that the spectra can be described by a two temperature model as well as by a cooling flow model. In both cases, the hot phase gas with the temperature of ~3.5keV occupies more than 90% of the total emission measure within 1'.5 from the cluster center. The estimated mass of the cooler (0.5-0.7keV) component is ~2-6 x 10^9 M_solar, which is comparable to the mass of hot halos seen in non-cD ellipticals. The cooling flow model gives the mass deposition rate of 60+-30 M_solar/yr, an order of magnitude lower than the previous estimation.
△ Less
Submitted 25 January, 1997; v1 submitted 20 January, 1997;
originally announced January 1997.