-
CATGNN: Cost-Efficient and Scalable Distributed Training for Graph Neural Networks
Authors:
Xin Huang,
Weipeng Zhuo,
Minh Phu Vuong,
Shiju Li,
Jongryool Kim,
Bradley Rees,
Chul-Ho Lee
Abstract:
Graph neural networks have been shown successful in recent years. While different GNN architectures and training systems have been developed, GNN training on large-scale real-world graphs still remains challenging. Existing distributed systems load the entire graph in memory for graph partitioning, requiring a huge memory space to process large graphs and thus hindering GNN training on such large…
▽ More
Graph neural networks have been shown successful in recent years. While different GNN architectures and training systems have been developed, GNN training on large-scale real-world graphs still remains challenging. Existing distributed systems load the entire graph in memory for graph partitioning, requiring a huge memory space to process large graphs and thus hindering GNN training on such large graphs using commodity workstations. In this paper, we propose CATGNN, a cost-efficient and scalable distributed GNN training system which focuses on scaling GNN training to billion-scale or larger graphs under limited computational resources. Among other features, it takes a stream of edges as input, instead of loading the entire graph in memory, for partitioning. We also propose a novel streaming partitioning algorithm named SPRING for distributed GNN training. We verify the correctness and effectiveness of CATGNN with SPRING on 16 open datasets. In particular, we demonstrate that CATGNN can handle the largest publicly available dataset with limited memory, which would have been infeasible without increasing the memory space. SPRING also outperforms state-of-the-art partitioning algorithms significantly, with a 50% reduction in replication factor on average.
△ Less
Submitted 2 April, 2024;
originally announced April 2024.
-
The ESO Science Archive Facility: Status, Impact, and Prospects
Authors:
Martino Romaniello,
Magda Arnaboldi,
Mauro Barbieri,
Nausicaa Delmotte,
Adam Dobrzycki,
Nathalie Fourniol,
Wolfram Freudling,
Jorge Grave,
Laura Mascetti,
Alberto Micol,
Jörg Retzlaff,
Nicolas Rosse,
Tomas Tax,
Myha Vuong,
Olivier Hainaut,
Marina Rejkuba,
Michael Sterzik
Abstract:
Scientific data collected at ESO's observatories are freely and openly accessible online through the ESO Science Archive Facility. In addition to the raw data straight out of the instruments, the ESO Science Archive also contains four million processed science files available for use by scientists and astronomy enthusiasts worldwide. ESO subscribes to the FAIR (Findable, Accessible, Interoperable,…
▽ More
Scientific data collected at ESO's observatories are freely and openly accessible online through the ESO Science Archive Facility. In addition to the raw data straight out of the instruments, the ESO Science Archive also contains four million processed science files available for use by scientists and astronomy enthusiasts worldwide. ESO subscribes to the FAIR (Findable, Accessible, Interoperable, Reusable) guiding principles for scientific data management and stewardship. All data in the ESO Science Archive are distributed according to the terms of the Creative Commons Attribution 4.0 International licence (CC BY 4.0).
△ Less
Submitted 31 October, 2023;
originally announced October 2023.
-
Voyager: MTD-Based Aggregation Protocol for Mitigating Poisoning Attacks on DFL
Authors:
Chao Feng,
Alberto Huertas Celdran,
Michael Vuong,
Gerome Bovet,
Burkhard Stiller
Abstract:
The growing concern over malicious attacks targeting the robustness of both Centralized and Decentralized Federated Learning (FL) necessitates novel defensive strategies. In contrast to the centralized approach, Decentralized FL (DFL) has the advantage of utilizing network topology and local dataset information, enabling the exploration of Moving Target Defense (MTD) based approaches.
This work…
▽ More
The growing concern over malicious attacks targeting the robustness of both Centralized and Decentralized Federated Learning (FL) necessitates novel defensive strategies. In contrast to the centralized approach, Decentralized FL (DFL) has the advantage of utilizing network topology and local dataset information, enabling the exploration of Moving Target Defense (MTD) based approaches.
This work presents a theoretical analysis of the influence of network topology on the robustness of DFL models. Drawing inspiration from these findings, a three-stage MTD-based aggregation protocol, called Voyager, is proposed to improve the robustness of DFL models against poisoning attacks by manipulating network topology connectivity. Voyager has three main components: an anomaly detector, a network topology explorer, and a connection deployer. When an abnormal model is detected in the network, the topology explorer responds strategically by forming connections with more trustworthy participants to secure the model. Experimental evaluations show that Voyager effectively mitigates various poisoning attacks without imposing significant resource and computational burdens on participants. These findings highlight the proposed reactive MTD as a potent defense mechanism in the context of DFL.
△ Less
Submitted 14 February, 2024; v1 submitted 12 October, 2023;
originally announced October 2023.
-
Semantic Search using Spreading Activation based on Ontology
Authors:
Ngo Minh Vuong
Abstract:
Currently, the text document retrieval systems have many challenges in exploring the semantics of queries and documents. Each query implies information which does not appear in the query but the documents related with the information are also expected by user. The disadvantage of the previous spreading activation algorithms could be many irrelevant concepts added to the query. In this paper, a pro…
▽ More
Currently, the text document retrieval systems have many challenges in exploring the semantics of queries and documents. Each query implies information which does not appear in the query but the documents related with the information are also expected by user. The disadvantage of the previous spreading activation algorithms could be many irrelevant concepts added to the query. In this paper, a proposed novel algorithm is only activate and add to the query named entities which are related with original entities in the query and explicit relations in the query.
△ Less
Submitted 9 May, 2019;
originally announced May 2019.
-
The SAMI Galaxy Survey: Data Release One with Emission-line Physics Value-Added Products
Authors:
Andrew W. Green,
Scott M. Croom,
Nicholas Scott,
Luca Cortese,
Anne M. Medling,
Francesco D'Eugenio,
Julia J. Bryant,
Joss Bland-Hawthorn,
J. T. Allen,
Rob Sharp,
I-Ting Ho,
Brent Groves,
Michael J. Drinkwater,
Elizabeth Mannering,
Lloyd Harischandra,
Jesse van de Sande,
Adam D. Thomas,
Simon O'Toole,
Richard M. McDermid,
Minh Vuong,
Katrina Sealey,
Amanda E. Bauer,
S. Brough,
Barbara Catinella,
Gerald Cecil
, et al. (26 additional authors not shown)
Abstract:
We present the first major release of data from the SAMI Galaxy Survey. This data release focuses on the emission-line physics of galaxies. Data Release One includes data for 772 galaxies, about 20% of the full survey. Galaxies included have the redshift range 0.004 < z < 0.092, a large mass range (7.6 < log(Mstellar/M$_\odot$) < 11.6), and star-formation rates of 10^-4 to 10^1\ M$_\odot$/yr. For…
▽ More
We present the first major release of data from the SAMI Galaxy Survey. This data release focuses on the emission-line physics of galaxies. Data Release One includes data for 772 galaxies, about 20% of the full survey. Galaxies included have the redshift range 0.004 < z < 0.092, a large mass range (7.6 < log(Mstellar/M$_\odot$) < 11.6), and star-formation rates of 10^-4 to 10^1\ M$_\odot$/yr. For each galaxy, we include two spectral cubes and a set of spatially resolved 2D maps: single- and multi-component emission-line fits (with dust extinction corrections for strong lines), local dust extinction and star-formation rate. Calibration of the fibre throughputs, fluxes and differential-atmospheric-refraction has been improved over the Early Data Release. The data have average spatial resolution of 2.16 arcsec (FWHM) over the 15~arcsec diameter field of view and spectral (kinematic) resolution R=4263 (sigma=30km/s) around Halpha. The relative flux calibration is better than 5\% and absolute flux calibration better than $\pm0.22$~mag, with the latter estimate limited by galaxy photometry. The data are presented online through the Australian Astronomical Observatory's Data Central.
△ Less
Submitted 26 July, 2017;
originally announced July 2017.
-
AAO Starbugs: software control and associated algorithms
Authors:
Nuria P. F. Lorente,
Minh V. Vuong,
Keith Shortridge,
Tony J. Farrell,
Scott Smedley,
Sungwook E. Hong,
Carlos Bacigalupo,
Michael Goodwin,
Kyler Kuehn,
Christophe Satorre
Abstract:
The Australian Astronomical Observatory's TAIPAN instrument deploys 150 Starbug robots to position optical fibres to accuracies of 0.3 arcsec, on a 32 cm glass field plate on the focal plane of the 1.2 m UK-Schmidt telescope. This paper describes the software system developed to control and monitor the Starbugs, with particular emphasis on the automated path-finding algorithms, and the metrology s…
▽ More
The Australian Astronomical Observatory's TAIPAN instrument deploys 150 Starbug robots to position optical fibres to accuracies of 0.3 arcsec, on a 32 cm glass field plate on the focal plane of the 1.2 m UK-Schmidt telescope. This paper describes the software system developed to control and monitor the Starbugs, with particular emphasis on the automated path-finding algorithms, and the metrology software which keeps track of the position and motion of individual Starbugs as they independently move in a crowded field. The software employs a tiered approach to find a collision-free path for every Starbug, from its current position to its target location. This consists of three path-finding stages of increasing complexity and computational cost. For each Starbug a path is attempted using a simple method. If unsuccessful, subsequently more complex (and expensive) methods are tried until a valid path is found or the target is flagged as unreachable.
△ Less
Submitted 8 August, 2016;
originally announced August 2016.
-
First Light Results from the Hermes Spectrograph at the AAT
Authors:
Andrew Sheinis,
Borja Anguiano,
Martin Asplund,
Carlos Bacigalupo,
Sam Barden,
Michael Birchall,
Joss Bland-Hawthorn,
Jurek Brzeski,
Russell Cannon,
Daniela Carollo,
Scott Case,
Andrew Casey,
Vladimir Churilov,
Couch Warrick,
Robert Dean,
Gayandhi De Silva,
Valentina D'Orazi,
Ly Duong,
Tony Farrell,
Kristin Fiegert,
Kenneth Freeman,
Frost Gabriella,
Luke Gers,
Michael Goodwin,
Doug Gray
, et al. (38 additional authors not shown)
Abstract:
The High Efficiency and Resolution Multi Element Spectrograph, HERMES, is a facility-class optical spectrograph for the Anglo-Australian Telescope (AAT). It is designed primarily for Galactic Archaeology, the first major attempt to create a detailed understanding of galaxy formation and evolution by studying the history of our own galaxy, the Milky Way. The goal of the GALAH survey is to reconstru…
▽ More
The High Efficiency and Resolution Multi Element Spectrograph, HERMES, is a facility-class optical spectrograph for the Anglo-Australian Telescope (AAT). It is designed primarily for Galactic Archaeology, the first major attempt to create a detailed understanding of galaxy formation and evolution by studying the history of our own galaxy, the Milky Way. The goal of the GALAH survey is to reconstruct the mass assembly history of the Milky Way through a detailed chemical abundance study of one million stars. The spectrograph is based at the AAT and is fed by the existing 2dF robotic fiber positioning system. The spectrograph uses volume phase holographic gratings to achieve a spectral resolving power of 28,000 in standard mode and also provides a high-resolution mode ranging between 40,000 and 50,000 using a slit mask. The GALAH survey requires an SNR greater than 100 for a star brightness of V ?= 14 in an exposure time of one hour. The total spectral coverage of the four channels is about 100 nm between 370 and 1000 nm for up to 392 simultaneous targets within the 2-degree field of view. HERMES has been commissioned over three runs, during bright time in October, November, and December 2013, in parallel with the beginning of the GALAH pilot survey, which started in November 2013. We present the first-light results from the commissioning run and the beginning of the GALAH survey, including performance results such as throughput and resolution, as well as instrument reliability.
△ Less
Submitted 31 August, 2015;
originally announced September 2015.
-
March of the Starbugs: Configuring Fibre-bearing Robots on the UK-Schmidt Optical Plane
Authors:
Nuria P. F. Lorente,
Minh Vuong,
Christophe Satorre,
Sungwook E. Hong,
Keith Shortridge,
Michael Goodwin,
Kyler Kuehn
Abstract:
The TAIPAN instrument, currently being developed for the Australian Astronomical Observatory's UK Schmidt telescope at Siding Spring Observatory, makes use of the AAO's Starbug technology to deploy 150 science fibres to target positions on the optical plane. This paper describes the software system for controlling and deploying the fibre-bearing Starbug robots. The TAIPAN software is responsible f…
▽ More
The TAIPAN instrument, currently being developed for the Australian Astronomical Observatory's UK Schmidt telescope at Siding Spring Observatory, makes use of the AAO's Starbug technology to deploy 150 science fibres to target positions on the optical plane. This paper describes the software system for controlling and deploying the fibre-bearing Starbug robots. The TAIPAN software is responsible for allocating each Starbug to its next target position based on its current position and the distribution of targets, finding a collision-free path for each Starbug, and then simultaneously controlling the Starbug hardware in a closed loop, with a metrology camera used to determine the position of each Starbug in the field during reconfiguration. The software is written in C++ and Java and employs a DRAMA middleware layer (Farrell et al. 1995).
△ Less
Submitted 15 March, 2015;
originally announced March 2015.
-
CYCLOPS2: the fibre image slicer upgrade for the UCLES high resolution spectrograph
Authors:
Anthony Horton,
C. G. Tinney,
Scott Case,
Tony Farrell,
Luke Gers,
Damien Jones,
Jon Lawrence,
Stan Miziarski,
Nick Staszak,
David Orr,
Minh Vuong,
Lew Waller,
Ross Zhelem
Abstract:
CYCLOPS2 is an upgrade for the UCLES high resolution spectrograph on the Anglo-Australian Telescope, scheduled for commissioning in semester 2012A. By replacing the 5 mirror Coudé train with a Cassegrain mounted fibre-based image slicer CYCLOPS2 simultaneously provides improved throughput, reduced aperture losses and increased spectral resolution. Sixteen optical fibres collect light from a 5.0 ar…
▽ More
CYCLOPS2 is an upgrade for the UCLES high resolution spectrograph on the Anglo-Australian Telescope, scheduled for commissioning in semester 2012A. By replacing the 5 mirror Coudé train with a Cassegrain mounted fibre-based image slicer CYCLOPS2 simultaneously provides improved throughput, reduced aperture losses and increased spectral resolution. Sixteen optical fibres collect light from a 5.0 arcsecond^2 area of sky and reformat it into the equivalent of a 0.6 arcsecond wide slit, delivering a spectral resolution of R = 70000 and up to twice as much flux as the standard 1 arcsecond slit of the Coudé train. CYCLOPS2 also adds support for simultaneous ThAr wavelength calibration via a dedicated fibre. CYCLOPS2 consists of three main components, the fore-optics unit, fibre bundle and slit unit. The fore optics unit incorporates magnification optics and a lenslet array and is designed to mount to the CURE Cassegrain instrument interface, which provides acquisition, guiding and calibration facilities. The fibre bundle transports the light from the Cassegrain focus to the UCLES spectrograph at Coudé and also includes a fibre mode scrambler. The slit unit consists of the fibre slit and relay optics to project an image of the slit onto the entrance aperture of the UCLES spectrograph. CYCLOPS2 builds on experience with the first generation CYCLOPS fibre system, which we also describe in this paper. We present the science case for an image slicing fibre feed for echelle spectroscopy and describe the design of CYCLOPS and CYCLOPS2.
△ Less
Submitted 3 January, 2013;
originally announced January 2013.
-
Data Provenance: Use Cases for the ESO archive, and Interactions with the Virtual Observatory
Authors:
J. D. Santander-Vela,
A. Delgado,
N. Delmotte,
M. Vuong
Abstract:
In the Virtual Observatory era, where we intend to expose scientists (or software agents on their behalf) to a stream of observations from all existing facilities, the ability to access and to further interpret the origin, relationships, and processing steps on archived astronomical assets (their Provenance) is a requirement for proper observation selection, and quality assessment. In this artic…
▽ More
In the Virtual Observatory era, where we intend to expose scientists (or software agents on their behalf) to a stream of observations from all existing facilities, the ability to access and to further interpret the origin, relationships, and processing steps on archived astronomical assets (their Provenance) is a requirement for proper observation selection, and quality assessment. In this article we present the different use cases Data Provenance is needed for, the challenges inherent to building such a system for the ESO archive, and their link with ongoing work in the International Virtual Observatory Alliance (IVOA).
△ Less
Submitted 2 February, 2010;
originally announced February 2010.
-
Determination of the gas-to-dust ratio in nearby dense clouds using X-ray absorption measurements
Authors:
MyHa Vuong,
Thierry Montmerle,
Nicolas Grosso,
Eric Feigelson,
Laurent Verstraete,
Hideki Ozawa
Abstract:
We present a comparison of the gas and dust properties of the dense interstellar matter in six nearby star-forming regions (d<500 pc): rho Oph, Cha I, R CrA, IC 348, NGC 1333, and Orion. We measure from Chandra and XMM-Newton observations the X-ray absorption toward pre-main sequence stars (PMS) without accretion disks (i.e., Class III sources) to obtain the total hydrogen column density N_{H,X}…
▽ More
We present a comparison of the gas and dust properties of the dense interstellar matter in six nearby star-forming regions (d<500 pc): rho Oph, Cha I, R CrA, IC 348, NGC 1333, and Orion. We measure from Chandra and XMM-Newton observations the X-ray absorption toward pre-main sequence stars (PMS) without accretion disks (i.e., Class III sources) to obtain the total hydrogen column density N_{H,X}. For these sources we take from the literature the corresponding dust extinction in the near-infrared, A_J, or when unavailable we derive it from SED fitting using the available DENIS, 2MASS, ISOCAM and other data. We then compare N_{H,X} and A_J for each object, up to unprecedently high extinction. For the rho Oph dark cloud with a relatively large sample of 20 bona-fide Class III sources, we probe the extinction up to A_J <~ 14 (A_V <~ 45), and find a best-fit linear relation N_{H,X}/A_J = 5.6 (+/- 0.4)x10^{21} cm^{-2} mag^{-1}, adopting standard ISM abundances. The other regions reveal a large dispersion in the N_{H,X}/A_J ratio for each source but for lack of adequate IR data these studies remain limited to moderate extinctions (A_J <~ 1.5 or A_V <~5). For rho Oph, the N_{H,X}/A_J ratio is significantly lower (>~2 sigma) than the galactic value, derived using the standard extinction curve (R_V = 3.1). This result is consistent with the recent downwards revision of the metallicity of the Sun and stars in the solar vicinity. We find that the rho Oph dense cloud has the same metallicity than the local ISM when assuming that the galactic gas-to-dust ratio remains unchanged. The difference between galactic and local values of the gas-to-dust ratio can thus be attributed entirely to a difference in metallicity.
△ Less
Submitted 23 June, 2003;
originally announced June 2003.
-
Low mass T Tauri and young brown dwarf candidates in the Chamaeleon II dark cloud found by DENIS
Authors:
My Ha Vuong,
Laurent Cambresy,
Nicolas Epchtein
Abstract:
We define a sample designed to select low-mass T Tauri stars and young brown dwarfs using DENIS data in the Chamaeleon II molecular cloud. We use a star count method to construct an extinction map of the Chamaeleon II cloud. We select our low-mass T Tauri star and young brown dwarf candidates by their strong infrared colour excess in the I-J/J-K_s colour-colour dereddened diagram. We retain only…
▽ More
We define a sample designed to select low-mass T Tauri stars and young brown dwarfs using DENIS data in the Chamaeleon II molecular cloud. We use a star count method to construct an extinction map of the Chamaeleon II cloud. We select our low-mass T Tauri star and young brown dwarf candidates by their strong infrared colour excess in the I-J/J-K_s colour-colour dereddened diagram. We retain only objects with colours I-J>2, and spatially distributed in groups around the cloud cores. This provides a sample of 70 stars of which 4 are previously known T Tauri stars. We have carefully checked the reliability of all these objects by visual inspection on the DENIS images. Thanks to the association of the optical I-band to the infra-red J and K_s bands in DENIS, we can apply this selection method to all star formation regions observed in the southern hemisphere. We also identify six DENIS sources with X-ray sources detected by ROSAT. Assuming that they are reliable low-mass candidates and using the evolutionary models for low-mass stars, we estimate the age of these sources between 1 Myr and < 10 Myr.
△ Less
Submitted 24 September, 2001;
originally announced September 2001.
-
Dehydrogenation of polycyclic aromatic hydrocarbons in the diffuse interstellar medium
Authors:
My Ha Vuong,
Bernard H. Foing
Abstract:
We present a model for the hydrogenation states of Polycyclic Aromatic Hydrocarbons (PAHs) in the diffuse interstellar medium. First, we study the abundance of hydrogenation and charge states of PAHs due to photo-ionization, photo-dissociation in the interstellar UV field, electron recombination and chemical reactions between PAH cations and H or H_2. For PAH cations, we find that the dehydrogen…
▽ More
We present a model for the hydrogenation states of Polycyclic Aromatic Hydrocarbons (PAHs) in the diffuse interstellar medium. First, we study the abundance of hydrogenation and charge states of PAHs due to photo-ionization, photo-dissociation in the interstellar UV field, electron recombination and chemical reactions between PAH cations and H or H_2. For PAH cations, we find that the dehydrogenation effects are dominant. The hydrogenation state of PAHs depends strongly on the H density, the size of the molecule and UV field. In diffuse clouds with low H density and normal UV radiation, PAHs containing less than 40 C are completely or strongly dehydrogenated whereas at high H density, they are normally hydrogenated. The partially dehydrogenated species dominate in intermediate density clouds. PAHs above 40 C are quite stable and are fully hydrogenated, which would favor their spectroscopic search in near IR surveys of Diffuse Interstellar Bands (DIBs).
△ Less
Submitted 19 October, 2000;
originally announced October 2000.
