arXiv subjects

Ryosuke Goto

Publications and source records attributed to Ryosuke Goto.

11 recordsLinked to original sources

arXiv2024

On permutation-invariant neural networks

Conventional machine learning algorithms have traditionally been designed under the assumption that input data follows a vector-based format, with an emphasis on vector-centric paradigms. However, as the demand for tasks involving set-based inputs has grown, there has been a paradigm shift in the research community towards addressing these challenges. In recent years, the emergence of neural network architectures such as Deep Sets and Transformers has presented a significant advancement in the treatment of set-based data. These architectures are specifically engineered to naturally accommodate sets as input, enabling more effective representation and processing of set structures. Consequently, there has been a surge of research endeavors dedicated to exploring and harnessing the capabilities of these architectures for various tasks involving the approximation of set functions. This comprehensive survey aims to provide an overview of the diverse problem settings and ongoing research efforts pertaining to neural networks that approximate set functions. By delving into the intricacies of these approaches and elucidating the associated challenges, the survey aims to equip readers with a comprehensive understanding of the field. Through this comprehensive perspective, we hope that researchers can gain valuable insights into the potential applications, inherent limitations, and future directions of set-based neural networks. Indeed, from this survey we gain two insights: i) Deep Sets and its variants can be generalized by differences in the aggregation function, and ii) the behavior of Deep Sets is sensitive to the choice of the aggregation function. From these observations, we show that Deep Sets, one of the well-known permutation-invariant neural networks, can be generalized in the sense of a quasi-arithmetic mean.

cs.LG↗

arXiv2023

Outfit Completion via Conditional Set Transformation

In this paper, we formulate the outfit completion problem as a set retrieval task and propose a novel framework for solving this problem. The proposal includes a conditional set transformation architecture with deep neural networks and a compatibility-based regularization method. The proposed method utilizes a map with permutation-invariant for the input set and permutation-equivariant for the condition set. This allows retrieving a set that is compatible with the input set while reflecting the properties of the condition set. In addition, since this structure outputs the element of the output set in a single inference, it can achieve a scalable inference speed with respect to the cardinality of the output set. Experimental results on real data reveal that the proposed method outperforms existing approaches in terms of accuracy of the outfit completion task, condition satisfaction, and compatibility of completion results.

cs.LG↗

arXiv2020

Anime-to-Real Clothing: Cosplay Costume Generation via Image-to-Image Translation

Cosplay has grown from its origins at fan conventions into a billion-dollar global dress phenomenon. To facilitate imagination and reinterpretation from animated images to real garments, this paper presents an automatic costume image generation method based on image-to-image translation. Cosplay items can be significantly diverse in their styles and shapes, and conventional methods cannot be directly applied to the wide variation in clothing images that are the focus of this study. To solve this problem, our method starts by collecting and preprocessing web images to prepare a cleaned, paired dataset of the anime and real domains. Then, we present a novel architecture for generative adversarial networks (GANs) to facilitate high-quality cosplay image generation. Our GAN consists of several effective techniques to fill the gap between the two domains and improve both the global and local consistency of generated images. Experiments demonstrated that, with two types of evaluation metrics, the proposed GAN achieves better performance than existing methods. We also showed that the images generated by the proposed method are more realistic than those generated by the conventional methods. Our codes and pretrained model are available on the web.

cs.CV↗

arXiv2020

A Computer-Aided Diagnosis System Using Artificial Intelligence for Hip Fractures -Multi-Institutional Joint Development Research-

[Objective] To develop a Computer-aided diagnosis (CAD) system for plane frontal hip X-rays with a deep learning model trained on a large dataset collected at multiple centers. [Materials and Methods]. We included 5295 cases with neck fracture or trochanteric fracture who were diagnosed and treated by orthopedic surgeons using plane X-rays or computed tomography (CT) or magnetic resonance imaging (MRI) who visited each institution between April 2009 and March 2019 were enrolled. Cases in which both hips were not included in the photographing range, femoral shaft fractures, and periprosthetic fractures were excluded, and 5242 plane frontal pelvic X-rays obtained from 4,851 cases were used for machine learning. These images were divided into 5242 images including the fracture side and 5242 images without the fracture side, and a total of 10484 images were used for machine learning. A deep convolutional neural network approach was used for machine learning. Pytorch 1.3 and Fast.ai 1.0 were used as frameworks, and EfficientNet-B4, which is pre-trained ImageNet model, was used. In the final evaluation, accuracy, sensitivity, specificity, F-value and area under the curve (AUC) were evaluated. Gradient-weighted class activation mapping (Grad-CAM) was used to conceptualize the diagnostic basis of the CAD system. [Results] The diagnostic accuracy of the learning model was accuracy of 96. 1 %, sensitivity of 95.2 %, specificity of 96.9 %, F-value of 0.961, and AUC of 0.99. The cases who were correct for the diagnosis showed generally correct diagnostic basis using Grad-CAM. [Conclusions] The CAD system using deep learning model which we developed was able to diagnose hip fracture in the plane X-ray with the high accuracy, and it was possible to present the decision reason.

physics.med-ph↗

arXiv2018

Outfit Generation and Style Extraction via Bidirectional LSTM and Autoencoder

When creating an outfit, style is a criterion in selecting each fashion item. This means that style can be regarded as a feature of the overall outfit. However, in various previous studies on outfit generation, there have been few methods focusing on global information obtained from an outfit. To address this deficiency, we have incorporated an unsupervised style extraction module into a model to learn outfits. Using the style information of an outfit as a whole, the proposed model succeeded in generating outfits more flexibly without requiring additional information. Moreover, the style information extracted by the proposed model is easy to interpret. The proposed model was evaluated on two human-generated outfit datasets. In a fashion item prediction task (missing prediction task), the proposed model outperformed a baseline method. In a style extraction task, the proposed model extracted some easily distinguishable styles. In an outfit generation task, the proposed model generated an outfit while controlling its styles. This capability allows us to generate fashionable outfits according to various preferences.

cs.CV↗

arXiv2017

The Stellar Mass, Star Formation Rate and Dark Matter Halo Properties of LAEs at $z\sim2$

We present average stellar population properties and dark matter halo masses of $z \sim 2$ \lya emitters (LAEs) from SED fitting and clustering analysis, respectively, using $\simeq$ $1250$ objects ($NB387\le25.5$) in four separate fields of $\simeq 1$ deg$^2$ in total. With an average stellar mass of $10.2\, \pm\, 1.8\times 10^8\ {\mathrm M_\odot}$ and star formation rate of $3.4\, \pm\, 0.4\ {\mathrm M_\odot}\ {\rm yr^{-1}}$, the LAEs lie on an extrapolation of the star-formation main sequence (MS) to low stellar mass. Their effective dark matter halo mass is estimated to be $4.0_{-2.9}^{+5.1} \times 10^{10}\ {\mathrm M_\odot}$ with an effective bias of $1.22^{+0.16}_{-0.18}$ which is lower than that of $z \sim 2$ LAEs ($1.8\, \pm\, 0.3$), obtained by a previous study based on a three times smaller survey area, with a probability of $96\%$. However, the difference in the bias values can be explained if cosmic variance is taken into account. If such a low halo mass implies a low HI gas mass, this result appears to be consistent with the observations of a high \lya escape fraction. With the low halo masses and ongoing star formation, our LAEs have a relatively high stellar-to-halo mass ratio (SHMR) and a high efficiency of converting baryons into stars. The extended Press-Schechter formalism predicts that at $z=0$ our LAEs are typically embedded in halos with masses similar to that of the Large Magellanic Cloud (LMC); they will also have similar SHMRs to the LMC, if their SFRs are largely suppressed after $z \sim 2$ as some previous studies have reported for the LMC itself.

astro-ph.GA↗

arXiv2016

Ly$α$ Emitters with Very Large Ly$α$ Equivalent Widths, EW$_{\rm 0}$(Ly$α$) $\simeq 200-400$ Å, at $z\sim 2$

We present physical properties of spectroscopically confirmed Ly$α$ emitters (LAEs) with very large rest-frame Ly$α$ equivalent widths EW$_{\rm 0}$(Ly$α$). Although the definition of large EW$_{\rm 0}$(Ly$α$) LAEs is usually difficult due to limited statistical and systematic uncertainties, we identify six LAEs selected from $\sim 3000$ LAEs at $z\sim 2$ with reliable measurements of EW$_{\rm 0}$ (Ly$α$) $\simeq 200-400$ Å given by careful continuum determinations with our deep photometric and spectroscopic data. These large EW$_{\rm 0}$(Ly$α$) LAEs do not have signatures of AGN, but notably small stellar masses of $M_{\rm *} = 10^{7-8}$ $M_{\rm \odot}$ and high specific star-formation rates (star formation rate per unit galaxy stellar mass) of $\sim 100$ Gyr$^{-1}$. These LAEs are characterized by the median values of $L({\rm Lyα})=3.7\times 10^{42}$ erg s$^{-1}$ and $M_{\rm UV}=-18.0$ as well as the blue UV continuum slope of $β= -2.5\pm0.2$ and the low dust extinction $E(B-V)_{\rm *} = 0.02^{+0.04}_{-0.02}$, which indicate a high median Ly$α$ escape fraction of $f_{\rm esc}^{\rm Lyα}=0.68\pm0.30$. This large $f_{\rm esc}^{\rm Lyα}$ value is explained by the low {\sc Hi} column density in the ISM that is consistent with FWHM of the Ly$α$ line, ${\rm FWHM (Lyα)}=212\pm32$ km s$^{-1}$, significantly narrower than those of small EW$_{\rm 0}$(Ly$α$) LAEs. Based on the stellar evolution models, our observational constraints of the large EW$_{\rm 0}$ (Ly$α$), the small $β$, and the rest-frame He{\sc ii} equivalent width imply that at least a half of our large EW$_{\rm 0}$(Ly$α$) LAEs would have young stellar ages of $\lesssim 20$ Myr and very low metallicities of $Z<0.02 Z_\odot$ regardless of the star-formation history.

astro-ph.GA↗

arXiv2016

A Close Comparison between Observed and Modeled Lyα Lines for z ~ 2.2 Lyman Alpha Emitters

We present the results of a Lya profile analysis of 12 Lya emitters (LAEs) at z = 2.2 with high-resolution Lya spectra. We find that all 12 objects have a Lya profile with the main peak redward of the systemic redshift defined by nebular lines, and five have a weak, secondary peak blueward of the systemic redshift (blue bump). The average velocity offset of the red main peak (the blue bump, if any) with respect to the systemic redshift is Delta_v_Lya,r = 174+/- 19 km s-1 (Delta_v_Lya,b = -316+/-45 km s-1), which is smaller than (comparable to) that of Lyman-break galaxies (LBGs). The outflow velocities inferred from metal absorption lines in three individual and one stacked spectra are comparable to those of LBGs. The expanding shell model constructed by Verhamme et al. (2006) reproduces not only the Lya profiles but also other observed quantities including the outflow velocity and the FWHM of nebular lines for the non-blue bump objects. On the other hand, the model predicts too high FWHMs of nebular lines for the blue bump objects, although this discrepancy may disappear if we introduce additional Lya photons produced by gravitational cooling. We show that the small Delta_v_Lya,r values of our sample can be explained by low neutral-hydrogen column densities of log(NHI) = 18.9 cm-2 on average. This value is more than one order of magnitude lower than those of LBGs but is consistent with recent findings that LAEs have high ionization parameters and low Hi gas masses. This result suggests that low NHI values, giving reduced numbers of resonant scattering of Lya photons, are the key to the strong Lya emission of LAEs.

astro-ph.GA↗

arXiv2015

Physical conditions of the interstellar medium in star-forming galaxies at z~1.5

We present results from Subaru/FMOS near-infrared (NIR) spectroscopy of 118 star-forming galaxies at $z\sim1.5$ in the Subaru Deep Field. These galaxies are selected as [OII]$λ$3727 emitters at $z\approx$ 1.47 and 1.62 from narrow-band imaging. We detect H$α$ emission line in 115 galaxies, [OIII]$λ$5007 emission line in 45 galaxies, and H$β$, [NII]$λ$6584, and [SII]$λλ$6716,6731 in 13, 16, and 6 galaxies, respectively. Including the [OII] emission line, we use the six strong nebular emission lines in the individual and composite rest-frame optical spectra to investigate physical conditions of the interstellar medium in star-forming galaxies at $z\sim$1.5. We find a tight correlation between H$α$ and [OII], which suggests that [OII] can be a good star formation rate (SFR) indicator for galaxies at $z\sim1.5$. The line ratios of H$α$/[OII] are consistent with those of local galaxies. We also find that [OII] emitters have strong [OIII] emission lines. The [OIII]/[OII] ratios are larger than normal star-forming galaxies in the local Universe, suggesting a higher ionization parameter. Less massive galaxies have larger [OIII]/[OII] ratios. With evidence that the electron density is consistent with local galaxies, the high ionization of galaxies at high redshifts may be attributed to a harder radiation field by a young stellar population and/or an increase in the number of ionizing photons from each massive star.

astro-ph.GA↗

arXiv2014

What is the physical origin of strong Lya emission? II. Gas Kinematics and Distribution of Lya Emitters

We present a statistical study of velocities of Lya, interstellar (IS) absorption, and nebular lines and gas covering fraction for Lya emitters (LAEs) at z~2. We make a sample of 22 LAEs with a large Lya equivalent width (EW) of > 50A based on our deep Keck/LRIS observations, in conjunction with spectroscopic data from the Subaru/FMOS program and the literature. We estimate the average velocity offset of Lya from a systemic redshift determined with nebular lines to be dv_Lya=234+-9 km s-1. Using a Kolmogorv-Smirnov test, we confirm the previous claim of Hashimoto et al. (2013) that the average dv_Lya of LAEs is smaller than that of LBGs. Our LRIS data successfully identify blue-shifted multiple IS absorption lines in the UV continua of four LAEs on an individual basis. The average velocity offset of IS absorption lines from a systemic redshift is dv_IS=204+-27 km s-1, indicating LAE's gas outflow with a velocity comparable to typical LBGs. Thus, the ratio, R^Lya_ IS = dv_Lya/dv_IS of LAEs, is around unity, suggestive of low impacts on Lya transmission by resonant scattering of neutral hydrogen in the IS medium. We find an anti-correlation between Lya EW and the covering fraction, f_c, estimated from the depth of absorption lines, where f_c is an indicator of average neutral hydrogen column density, N_HI. The results of our study support the idea that N_HI is a key quantity determining Lya emissivity.

astro-ph.CO↗

arXiv2013

The stellar mass function and efficiency of galaxy formation with a varying initial mass function

Several recent observational studies have concluded that the initial mass function (IMF) of stars varies systematically with galaxy properties such as velocity dispersion. In this paper, we investigate the effect of linking the circular velocity of galaxies, as determined from the Fundamental Plane and Tully-Fisher relations, to the slope of the IMF with parameterizations guided by several of these studies. For each empirical relation, we generate stellar masses of ~600,000 SDSS galaxies at z ~ 0.1, by fitting the optical photometry to large suites of synthetic stellar populations that sample the full range of galaxy parameters. We generate stellar mass functions and examine the stellar-to-halo mass relations using sub-halo abundance matching. At the massive end, the stellar mass functions become a power law, instead of the familiar exponential decline. As a result, it is a generic feature of these models that the central galaxy stellar-to-halo mass relation is significantly flatter at high masses (slope ~ -0.3 to -0.4) than in the case of a universal IMF (slope ~ -0.6). We find that regardless of whether the IMF varies systematically in all galaxies or just early types, there is still a well-defined peak in the central stellar-to-halo mass ratio at halo masses of ~ 10E12 solar masses. In general, the IMF variations explored here lead to significantly higher integrated stellar densities if the assumed dependence on circular velocity applies to all galaxies, including late-types; in fact the more extreme cases can be ruled out, as they imply an unphysical situation in which the stellar fraction exceeds the universal baryon fraction.

astro-ph.CO↗