Search

Graphs are a highly expressive data structure, but it is often difficult for humans to find patterns from a complex graph. Hence, generating human-interpretable sequences from graphs have gained interest, called graph2seq learning. It is expected that the compositionality in a graph can be associated to the compositionality in the output sequence in many graph2seq tasks. Therefore, applying compositionality-aware GNN architecture would improve the model performance. In this study, we adopt the multi-level attention pooling (MLAP) architecture, that can aggregate graph representations from multiple levels of information localities. As a real-world example, we take up the extreme source code summarization task, where a model estimate the name of a program function from its source code. We demonstrate that the model having the MLAP architecture outperform the previous state-of-the-art model with more than seven times fewer parameters than it., Comment: 8 pages, 1 figure, 2 tables

Graph neural networks (GNNs) have been widely used to learn vector representation of graph-structured data and achieved better task performance than conventional methods. The foundation of GNNs is the message passing procedure, which propagates the information in a node to its neighbors. Since this procedure proceeds one step per layer, the range of the information propagation among nodes is small in the lower layers, and it expands toward the higher layers. Therefore, a GNN model has to be deep enough to capture global structural information in a graph. On the other hand, it is known that deep GNN models suffer from performance degradation because they lose nodes' local information, which would be essential for good model performance, through many message passing steps. In this study, we propose multi-level attention pooling (MLAP) for graph-level classification tasks, which can adapt to both local and global structural information in a graph. It has an attention pooling layer for each message passing step and computes the final graph representation by unifying the layer-wise graph representations. The MLAP architecture allows models to utilize the structural information of graphs with multiple levels of localities because it preserves layer-wise information before losing them due to oversmoothing. Results of our experiments show that the MLAP architecture improves the graph classification performance compared to the baseline architectures. In addition, analyses on the layer-wise graph representations suggest that aggregating information from multiple levels of localities indeed has the potential to improve the discriminability of learned graph representations., Comment: 17 pages + 5 appendix pages, 16 figures + 2 appendix figures, 2 tables + 5 appendix tables

Program comprehension is a dominant process in software development and maintenance. Experts are considered to comprehend the source code efficiently by directing their gaze, or attention, to important components in it. However, reflecting the importance of components is still a remaining issue in gaze behavior analysis for source code comprehension. Here we show a conceptual framework to compare the quantified importance of source code components with the gaze behavior of programmers. We use "attention" in attention models (e.g., code2vec) as the importance indices for source code components and evaluate programmers' gaze locations based on the quantified importance. In this report, we introduce the idea of our gaze behavior analysis using the attention map, and the results of a preliminary experiment., Comment: 4 pages, 2 figures; APSIPA 2019 ACCEPTED

In 2006 June, the obscured low luminosity active galactic nucleus in the nearby Seyfert 1.9 galaxy NGC 4258 was observed with Suzaku for ~ 100 ks. Utilizing the XIS and the HXD, the nucleus emission was detected over 2 to 40 keV range, with an unabsorbed 2--10 keV luminosity of 8 x 10 40 erg / s, and varied by a factor of ~ 2 during the observation. Its 2--40 keV spectrum is reproduced by a single power law with photon index of ~ 2.0, absorbed by an equivalent hydrogen column of ~ 1.0 x 10 23 cm2. The spectrum within 4' of the nucleus required also a softer thin-thermal emission, as well as an intermediate hardness component attributable to integrated point sources. A weak neutral Fe-Kalpha florescence line was detected at an equivalent width of ~ 40 eV. The cold reflection component was not required by the data, with the reflector solid angle Omega seen from the nucleus constrained as Omega / 2 pi < 0.3 assuming a general case of 60 deg inclination. The results suggest that the cold reflecting material around the nucleus is localized along our line of sight, rather than forming a thick torus., Comment: PASJ, NGC4258, Suzaku, 12 pages, 11 figures

The black-hole binary Cygnus X-1 was observed for 17 ks with the Suzaku X-ray observatory in 2005 October, while it was in a low/hard state with a 0.7-300 keV luminosity of 4.6 x 10^37 erg/s. The XIS and HXD spectra, spanning 0.7-400 keV, were reproduced successfully incorporating a cool accretion disk and a hot Comptonizing corona. The corona is characterized by an electron temperature of ~100 keV, and two optical depths of ~0.4 and ~1.5 which account for the harder and softer continua, respectively. The disk has the innermost temperature of ~0.2 keV, and is thought to protrude half way into the corona. The disk not only provides seed photons to the Compton cloud, but also produces a soft spectral excess, a mild reflection hump, and a weakly broadened iron line. A comparison with the Suzaku data on GRO J1655-40 reveals several interesting spectral differences, which can mostly be attributed to inclination effects assuming that the disk has a flat geometry while the corona is grossly spherical. An intensity-sorted spectroscopy indicates that the continuum becomes less Comptonized when the source flares up on times scales of 1-200 s, while the underlying disk remains unchanged., Comment: 21 pages, 12 figures, accepted for Publications of the Astronomical Society of Japan

This paper reports the discovery of a bright X-ray transient source, Suzaku J1305-4913, in the south-west arm of the nearby Seyfert II galaxy NGC 4945. It was detected at a 0.5 -- 10 keV flux of $2.2 \times 10^{-12}$ erg cm$^{-2}$ s$^{-1}$ during the Suzaku observation conducted on 2006 January 15 -- 17, but was undetectable in a shorter observation on 2005 August 22 --23, with an upper limit of $1.7 \times 10^{-14}$ erg cm$^{-2}$ s$^{-1}$ (90% confidence level). At a distance of 3.7 Mpc, the bolometric luminosity of the source becomes $L_{\rm bol} = 4.4 \times 10^{39} \alpha$ erg s$^{-1}$, where $\alpha = (\cos 60^\circ / \cos i)$ and $i$ is the disk inclination. Therefore, the source is classified into so-called ultraluminous X-ray sources (ULXs). The time-averaged X-ray spectrum of the source is described by a multi-color disk model, with the innermost accretion disk temperature of $T_{\rm in} = 1.69_{-0.05}^{+0.06}$ keV. During the 2006 January observation, it varied by a factor of 2 in intensity, following a clear correlation of $L_{\rm bol} \propto T_{\rm in}^4$. It is inferred that the innermost disk radius $R_{\rm in}$ stayed constant at $R_{\rm in} = 79_{-3.9}^{+4.0} \alpha^{1/2}$ km, suggesting the presence of a standard accretion disk. Relating $R_{\rm in}$ with the last stable orbit around a non-rotating black hole yields a rather low black hole mass, $\sim 9 \alpha^{1/2}$ solar masses, which would imply that the source is shining at a luminosity of $\sim3 \alpha^{1/2} $ times the Eddington limit. These results can be better interpreted by invoking sub-Eddington emission from a rapidly spinning black hole with a mass of 20 -- 130 solar masses., Comment: 21 pages, 10 figures, Accepted for PASJ 2nd Suzaku special issue

Suzaku observed a nearby Seyfert 2 galaxy NGC4945, which hosts one of the brightest active galactic nuclei above 20 keV. Combining data from the X-ray CCD camera (XIS) and the Hard X-ray Detector (HXD), the AGN intrinsic nuclear emission and its reprocessed signals were observed simultaneously. The intrinsic emission is highly obscured with an absorbing column of $\sim 5 \times 10^{24}$ cm$^{-2}$, and was detectable only above $\sim 10$ keV. The spectrum below 10 keV is dominated by reflection continuum and emission lines from neutral/ionized material. Along with a neutral iron K$\alpha$ line, a neutral iron K$\beta$ and a neutral nickel K$\alpha$ line were detected for the first time from this source. The neutral lines and the cold reflection continuum are consistent with both originating in the same location. The Compton down-scattered shoulder in the neutral Fe-K$\alpha$ line is $\sim 10%$ in flux of the narrow core, which confirms that the line originates from reflection rather than transmission. The weakness of the Compton shoulder also indicates that the reflector is probably seen nearly edge-on. Flux of the intrinsic emission varied by a factor of $\sim 2$ within $\sim 20$ ks, which requires the obscuring material to be geometrically thin. Broadband spectral modeling showed that the solid angle of the neutral reflector is less than a few $\times 10^{-2} \times 2\pi$. All this evidence regarding the reprocessed signals suggests that a disk-like absorber/reflector is viewed from a near edge-on angle., Comment: 11 pages, 11 figures

The Galactic black-hole binary GRO J1655$-$40 was observed with Suzaku on 2005 September 22--23, for a net exposure of 35 ks with the X-ray Imaging Spectrometer (XIS) and 20 ks with the Hard X-ray Detector (HXD). The source was detected over a broad and continuous energy range of 0.7--300 keV, with an intensity of $\sim$50 mCrab at 20 keV. At a distance of 3.2 kpc, the 0.7--300 keV luminosity is $ \sim 5.1 \times 10^{36}$ erg s$^{-1}$ ($\sim 0.7$ % of the Eddington luminosity for a 6 $M_{\odot}$ black hole). The source was in a typical low/hard state, exhibiting a power-law shaped continuum with a photon index of $\sim 1.6$. During the observation, the source intensity gradually decreased by 25% at energies above $\sim 3$ keV, and by 35% below 2 keV. This, together with the soft X-ray spectra taken with the XIS, suggests the presence of an independent soft component that can be represented by emission from a cool ($\sim 0.2$ keV) disk. The hard X-ray spectra obtained with the HXD reveal a high-energy spectral cutoff, with an e-folding energy of $\sim 200$ keV. Since the spectral photon index above 10 keV is harder by $\sim 0.4$ than that observed in the softer energy band, and the e-folding energy is higher than those of typical reflection humps, the entire 0.7--300 keV spectrum cannot be reproduced by a single thermal Comptonization model, even considering reflection effects. Instead, the spectrum (except the soft excess) can be successfully explained by invoking two thermal-Comptonization components with different $y$-parameters. In contrast to the high/soft state spectra of this object in which narrow iron absorption lines are detected with equivalent widths of 60--100 eV, the present XIS spectra bear no such features beyond an upper-limit equivalent width of 25 eV., Comment: 32 pages, 12 figures; accepted for publication in Publications of the Astronomical Society of Japan (PASJ)

We obtained a wide-band spectrum of the Compton-thick Seyfert 2 galaxy Mrk 3 with Suzaku. The observed spectrum was clearly resolved into weak, soft power-law emission, a heavily absorbed power-law component, cold reflection, and many emission lines. The heavily absorbed component, absorbed by gas with a column density of 1.1x10^24 cm^-2, has an intrinsic 2--10 keV luminosity of ~1.6x10^43 erg s^-1, and is considered to be direct emission from the Mrk 3 nucleus. The reflection component was interpreted as reflection of the direct light off cold, thick material; the reflection fraction $R$ was 1.36+/-0.20. The cold material is inferred to be located > 1 pc from the central black hole of Mrk 3 due to the low ionization parameter of iron (xi < 1 erg cm s^-1) and the narrow iron line width (s < 22 eV). A Compton shoulder to the iron line was detected, but the intensity of the shoulder component was less than that expected from spherically distributed Compton-thick material. The weak, soft power-law emission is considered to be scattered light by ionized gas. The existence of many highly-ionized lines from O, Ne, Mg, Si, S, and Fe in the observed spectrum indicates that the ionized gas has a broad ionized structure, with xi=10--1000. The scattering fraction with respect to the direct light was estimated to be 0.9+/-0.2%, which indicates that the column density of the scattering region is about 3.6x10^22 cm^-2. This high-quality spectrum obtained by Suzaku can be considered a template for studies of Seyfert 2 galaxies., Comment: 24 pages, 8 figures, accepted in PASJ for publication

We present detailed time-averaged X-ray spectroscopy in the 0.5--10 keV band of the Seyfert~1.9 galaxy NGC 2992 with the Suzaku X-ray Imaging Spectrometers (XIS). We model the complex continuum in detail. There is an Fe K line emission complex that we model with broad and narrow lines and we show that the intensities of the two components are decoupled at a confidence level >3sigma. The broad Fe K line has an EW of 118 (+32,-61) eV and could originate in an accretion disk (with inclination angle greater than ~30 degrees). The narrow Fe Kalpha line has an EW of 163 (+47,-26) eV and is unresolved FWHM <4090 km/s) and likely originates in distant matter. The absolute flux in the narrow line implies that the column density out of the line-of-sight could be much higher than measured in the line-of-sight, and that the mean (historically-averaged) continuum luminosity responsible for forming the line could be a factor of several higher than that measured from the data. We also detect the narrow Fe Kbeta line with a high signal-to-noise ratio and describe a new robust method to constrain the ionization state of Fe responsible for the Fe Kalpha and Fe Kbeta lines that does not require any knowledge of possible gravitational and Doppler energy shifts affecting the line energies. For the distant line-emitting matter (e.g. the putative obscuring torus) we deduce that the predominant ionization state is lower than Fe VIII (at 99% confidence), conservatively taking into account residual calibration uncertainties in the XIS energy scale and theoretical and experimental uncertainties in the Fe K fluorescent line energies. From the limits on a possible Compton-reflection continuum it is likely that the narrow Fe Kalpha and Fe Kbeta lines originate in a Compton-thin structure., Comment: Abstract is abridged. Accepted for publication in the Suzaku special issue of PASJ (November 2006). 18 pages, 6 figures