You are here

Abstracts

(32 results)
Name
Ismail Akhalwaya
Abstract Title
The role of IBM Research SA in the SKA Consortium
Abstract
The young IBM South Africa Research Lab has a vision to pursue research projects with the potential of great scientific and social impact of local relevance. With the SKA being the pre-eminent big science project in South Africa with frontier science, strong human capacity development and technical skills & infrastructure upliftment, it is natural that IBM Research should get seriously involved. This poster places IBM Research's role in the pipeline from instrument to scientific breakthrough. Amongst other efforts: on the human capacity development front, the IBM lab is already supervising post-graduate university students on SKA related topics; on the research front, IBM Research is developing computationally feasible Bayesian Interferometry solutions, investigating unsupervised learning for a Serendipity Machine, and exploring constrained complexity modelling in Deep Learning for fast and accurate Radio Galaxy Zoo classification. Some of the computing platforms that we employ include the open-source-based IBM Bluemix cloud offering with built-in distributed deep learning. The very same platform includes a powerful and collaborative interactive notebook interface which is planned to be deployed in a citizen science project, bringing SKA research to the people of Mzansi and the world.
Name
Richard Armstrong
Abstract Title
MeerKAT's real-time radio eye.
Abstract
The dynamic radio sky is a unique window into extreme astrophysical phenomena, such as relativistic flows, cataclysmic explosions and all types of accretion onto black holes -- an important research area for astrophysics in the upcoming decades. However, our view of the dynamic radio sky has thus far relied predominantly on 1.) archival searches of data, obviating the possibility of multi-wavelength follow-up and classification, and 2.) biased triggers from instruments at other wavelengths. MeerKAT will be the world's most sensitive radio telescope once science operations begin at the end of next year, and will provide an unrivalled view of the dynamic radio sky through a dedicated, real-time search for transients and variables during all scientific operations. In this talk, I will discuss the MeerKAT's fast, real-time imaging system, which will push the boundaries of transient research in terms of both temporal resolution (cadence of images) and completeness (time-on-sky).
Name
Sarah Blyth
Abstract Title
Data visualisation and analytics for LADUMA
Abstract
In order to meet the various science goals of the LADUMA survey, we will require visualisation tools which can display our large three-dimensional dataset effectively and in a timely manner, thereby enabling astronomers to better understand and analyse the data. In this talk I will present some of the shortfalls of the currently available visual-analytic platforms and discuss the requirements and other wishlist items for LADUMA science.
Name
Sarah Blyth
Abstract Title
Analysing the LADUMA dataset
Abstract
The LADUMA survey aims to investigate galaxy evolution back in time over two-thirds of the age of the Universe by studying the neutral hydrogen content of galaxies. The survey will involve a single pointing observation of a nominal 5000 hours with the MeerKAT, encompassing the extended-Chandra Deep Field-South. These will be some of the deepest HI measurements on any radio telescope to date and will present various challenges for our data analysis due to the sheer size of our final dataset. In this talk I will outline some of the analyses we intend to do and the tools and methods, some of which still need to be developed, that we will require to reach our science goals.
Name
Luke Chamandy
Abstract Title
Simulating cosmological magnetic fields
Abstract
Magnetic fields exist almost everywhere in the universe, and play a crucial role in many astrophysical processes. The evolution of magnetic fields within galaxies over cosmic time is governed by magnetohydrodynamics (MHD). Current cosmological simulations that include MHD are computationally demanding and are probably not able to resolve the smallest scales needed for a realistic treatment of magnetic fields. We are in the process of developing a hybrid scheme, wherein output from a semi-analytic galaxy evolution model (GALFORM), which in turn uses output from a cosmological dark matter simulation (MILLENIUM), is used as input for galactic dynamo mean-field simulations. These simulations solve a suitably simplified set of MHD equations to compute the magnetic field for each galaxy. Thus we hope to simulate the temporal and spatial evolution of the magnetic fields of >10^7 galaxies over cosmological timescales and volumes. The goal is to produce mock MeerKAT/SKA radio polarization data sets and images, which can then be compared with observations. The project will be carried out in phases, with each phase introducing more physics.
Name
Steve Crawford
Abstract Title
SALT Data Archive and the need for Multi-wavelength strategy
Abstract
The Southern African Large Telescope is the most cost effective 10m-class, general observatory in the world. It is the premiere optical facility in Africa and has been in regular science operations since 2011. While modest compared to the data processing requirements of MeerKAT, the SALT Data archive already contains 100 TB of data but will grow into a peta-scale archive with the next generation of instruments. A public interface was developed in collaboration with SA3 and India-VO, which has already produced at least one paper from SALT observations. However, the challenges of the SALT data archive are in supporting a wide diversity of programs observed with the many different modes for different investigators, providing access to high quality data products, and long term service and support for the archive. As such, development of the data reduction pipeline has supported almost all science publications and enabled fast response science, and we are currently investigating migrating to cloud-based architecture. SALT is already supporting MeerKAT key programs and other radio surveys with approximately 20% of the time already allocated to these proposals. Many of the MeerKAT key program science objects will rely on the availability and integration of optical data from SALT, other SAAO facilities, and observations around the world. As approximately 80% of papers that contain radio observations also contain optical observations, it will be absolutely critical to support the integration of optical data into the MeerKAT key science programs. The key challenge is producing an integrated data service that is able to provide seamless access to the multi-wavelength data sets necessary to achieve these science goals.
Name
Catherine Cress
Abstract Title
Astronomy at the Centre for High Performance Computing
Abstract
I will describe the facilities available at the CHPC and give an overview of activities related to Astronomy at the CHPC. Activities include work on simulations, data mining and various post-processing tools required for extraction of information.
Name
Daniel Cunnama
Abstract Title
Mocking Astrophysics
Abstract
The Mocking Astrophysics initiative was created with the aim of providing testing and validation of mock skies to underpin the scientific interpretation of the latest generation of massive astronomical surveys. I will provide a brief update on the Mocking Astrophysics project and the upcoming Comparison Cape Town workshop in July.
Name
Romeel Dave
Abstract Title
Big Data Challenges in Simulations of Galaxy Formation
Abstract
I will discuss recent cosmological hydrodynamic simulations run at UWC, compare them to recent simulations around the world, and outline plans to scale our simulations up to sizes that will require Big Data capabilities to handle. I will discuss tools we developed primarily at UWC to allow easy access to simulation data products, now being integrated into the "yt" simulation analysis package. I will discuss progress towards making mock radio observational images from simulations, in order to enable the closest possible comparisons between MeerKAT key project observations and state-of-the-art models.
Name
Radwan Deeb
Abstract Title
SAP HANA Voa
Abstract
SAP HANA Vora is an in-memory query engine which leverages and extends the Apache Spark execution framework to provide enriched interactive analytics on Hadoop. In this presentation we introduce SAP HANA Vora and show how SAP newest technology can not only meet digital enterprise needs but can also provide an innovative research platform using the power of in-memory to analyze Big Data in Hadoop.
Name
Goran Dragosavac
Abstract Title
Big Data - proceed with caution!
Abstract
While collection and analysis on big data hold great promise, quantity doesn't always translate to quality. There are many issues that needs to be resolved, starting from data relevancy, algorithmic efficiency, data management issues, component design, data ownership, privacy and security and most importantly – knowing end-goal, and having clear vision of business value be generated using big data and big-data technologies. This, presentation hopes to share more light on some of these challenges that comes with attempts to wrestle with big data.
Name
Ed Elson
Abstract Title
Optimising LADUMA stacking strategies
Abstract
HI stacking is a useful method for probing the HI content of distant galaxies. The method is well established, yet has its limitations. All HI stacking experiments suffer from double counting: the final stacked spectrum overestimates the true total HI content of the galaxy sample. Observers are well aware of this, yet have been unable to properly account for it. In this work I present a series of mock HI stacking experiments based on a new suite of simulated HI data products. The simulated products are used to create stacked HI spectra and to then decompose them into their mass components. The results clearly show most experiments to suffer severely from source confusion, with only a small fraction of the stacked signal actually coming from the target galaxies. LADUMA will stack galaxies from z~0.5 to beyond unity. This is a completely unexplored region of the Universe. Knowing how to properly extract galaxy spectra for stacking will go a very long way in determining the accuracy and reliability of the results. Our simulations can be used to optimise the observing and stacking strategies in order to maximise the scientific returns from LADUMA at z>0.5.
Name
José Fonseca
Abstract Title
New prospects for multi-wavelength surveys
Abstract
Next-generation cosmological surveys will probe ever larger volumes of the Universe, including the largest scales, near and beyond the horizon. On these scales, the galaxy power spectrum carries signatures of local primordial non-Gaussianity (PNG) and horizon-scale General Relativistic (GR) effects. However, cosmic variance limits detection of horizon-scale effects. Combining different surveys via the multi-tracer method allows us to beat down cosmic variance. We forecated that the combination of two contemporaneous surveys, a large neutral hydrogen intensity mapping survey in SKA Phase 1 and a Euclid-like photometric survey, will provide unprecedented constraints on PNG, achieve σ(fNL) ≃ 1.37−0.48, as well as detection of the GR effects with ∼ 7% accuracy. This not only opens new windows to study large scale effects but also poses questions about which are the best multi-tracer estimators as well as computational strategies for cosmological parameter estimation.
Name
Bradley Frank
Abstract Title
The Calibration and Imaging Pipeline for MeerKAT - An Overview
Abstract
The MeerKAT radio telescope will be one of the most sensitive radio telescopes in the world, and will embark on an exciting campaign comprising of several large imaging surveys during the next few years. The scientific prospects are matched by the challenges that we will face when calibrating and imaging MeerKAT data. Fortunately, many of the MeerKAT surveys have significant overlap in their calibration and imaging requirements, and we aim to use these synergies to develop a consolidated set of requirements that will inform the design of a flexible and comprehensive pipeline. In my talk I will outline the pipeline design related to the basic data flow from interferometric detection to scientific product. I will also provide an overview of the diverse scientific requirements of the large imaging surveys.
Name
David Gilbank
Abstract Title
Bridging the Gap Between Current Large Optical Surveys and the Future
Abstract
With a view towards the type of multiwavelength collaborations necessary to maximally exploit forthcoming radio surveys, I will give an overview of state-of-the-art in optical surveys from a couple of recent projects. - I will discuss experience gained from conducting one of the first new generation optical surveys, the ~1000 deg^2 Red-sequence Cluster Survey 2 (RCS-2), and the challenges faced by scaling up to the even larger imaging data sets coming in the next few years. - I will also describe some of the work recently started at SAAO using our new computing cluster to dissect large numbers of galaxies in parallel.
Name
Lauren Hunt
Abstract Title
Fast navigation of large universe simulations using trees
Abstract
As simulations get larger, we face greater problems in trying to navigate the data in detail. Spatial navigation is a difficult problem, both in the sense that objects within a simulation do not know about their environment and because the volumes that must be navigated are very large. Trying to compute the contribution from particles near an arbitrary location can be time consuming. I am working with GADGET simulations and need to trace a ray through these simulations. Defining a ray and all of the positions along it is easy, but it is challenging to efficiently determine which particles are near enough to have an effect on any given position, especially if I want to look at the details. To search through the entire table, or even a subset of it is not efficient, or even easy. I am implementing a much faster way using a data structure called a “kdtree" which is commonly used by game designers to determine what a bullet will hit in a shooting game. This method can link nearby particles extremely fast after an initial time outlay (to build the tree). The use of this method is facilitated with the use of existing packages such as scipy’s spatial package and a method called “ckdtree”. I have implemented this on a volume composed of 3 million particles and the next step will be to test it on a larger volume up to high redshifts.
Name
Gyula I. G. Jozsa
Abstract Title
TiRiFiC and FAT
Abstract
Blind HI surveys with MeerKAT will result in large number of spatially resolved, spectroscopic observations of individual galaxies, while targeted galaxy observations will reach an unprecedented depth of information about the HI distribution in galaxies. One branch of scientific analyses of HI data cubes aims at deriving the surface density profile and the rotation curve of the observed galaxies through tilted-ring modelling, which can be assumed to become a standard part of HI analysis pipelines. Traditional methods are, however, limited by resolution effects, the source's complexity at the sensitivity reached, and/or the sheer number of objects to be analysed. TiRiFiC (Tilted-Ring-Fitting-Code) and FAT (Fully Automated TiRiFiC) are software packages aimed at overcoming these problems by allowing for more complex models to be directly and automatically fitted to the data. This poses a computational challenge. I will present the method, current development, and future plans and challenges.
Name
Chris Koen
Abstract Title
Some statistical aspects of working with "big" data
Abstract
A few statistical issues particular to very large datasets are discussed.
Name
Michelle Kuttel
Abstract Title
Computational tools for Data Intensive Astronomy
Abstract
We have an established collaboration between Computer Scientists and Astronomers at the University of Cape Town with the goal of designing and developing effective computational tools to support research in Astronomy, with a focus on large data sets. In this talk, I will outline projects that I am involved in, specifically visualisation tools and parallel computing approaches to accelerate data analysis.
Name
Francois Luus
Abstract Title
SETI on Apache Spark
Abstract
The recently announced $100M Breakthrough Listen Initiative is reinvigorating efforts in the Search for Extra-Terrestrial Intelligence (SETI) by adopting a Silicon Valley approach to expertise, funding, and adhering to open data, open science and open source solutions. IBM has been making inroads in Big Data Analytics for SETI through an exploratory collaboration with the SETI Institute, NASA and the Swinburne University, to demonstrate the capabilities of open source platforms such as Apache Spark for analyzing large Allen Telescope Array measurement sets. This talk will feature an overview of the IBM-SETI Institute collaboration with a discussion of novel RFI analysis using Doppler shift; a statistical analysis of narrow-band paradigms; and wide-band detection and communication regime exploration. The implications of distributed computing solutions such as Spark on Big Data Analytics and CPU-only Deep Learning are also covered to facilitate a conversation on the suitability of constrained complexity modeling in radio astronomy analytics.
Name
Lindsay Magnus
Abstract Title
Using Docker to create a data analysis software suite
Abstract
Docker is typically used to package single applications to allow, amongst other things, ease of installation and portability. In this talk we look at the use of Docker to rather package together a collection of data analysis software packages into a single easily distributable suite of applications. This approach has been used by the Systems Engineering and Commissioning teams to process the initial MeerKAT data for qualification and acceptance of the MeerKAT receptors. Using Docker has allowed a diverse team, with varying operating systems and software knowledge, to be able to work with a centrally version controlled stack of applications to process the Data.
Name
Vanessa McBride
Abstract Title
MeerLICHT: an optical eye on the radio sky
Abstract
MeerLICHT is a 65cm optical telescope that will operate from Sutherland from early 2017. It will provide optical images of the sky in multiple filters simultaneous with radio observations by MeerKAT, which will be crucial to interpretation of both new and known radio transients. I will discuss the MeerLICHT project with specific focus on the data processing, storage and visualisation challenges.
Name
Shazrene Mohamed
Abstract Title
Physics of evolved stars: Linking realistic 3D models to detailed observations
Abstract
Evolved stars, e.g., Wolf-Rayet stars and red (super)giants lose copious amounts of mass and momentum through powerful, dense stellar winds. The interaction of these outflows with their surroundings results in highly structured and complex circumstellar environments, often featuring knots, arcs, shells and spirals. Recent improvements in computational power and techniques have led to the development of detailed, multi-dimensional simulations that have given new insight into the origin of these structures, and better understanding of the physical mechanisms driving their formation. In this talk, we describe 3D smoothed particle hydrodynamics models of the circumstellar environment around carbon-rich red giant, R Scl, and compare the results to high resolution observations from ALMA. We discuss how similar techniques can be used to interpret and make predictions for observations of related systems in the MeerKAT/SKA era.
Name
Arrykrishna Mootoovaloo
Abstract Title
FEMS - Fast and Easy Model Selection
Abstract
In the Bayesian formalism, the Bayesian Evidence, also known as the marginal likelihood, is the most common method for calculating the strength of a model. The higher the evidence the better the model. However, this is computationally expensive and we have come up with the idea of doing model comparison/selection in a fast and easy way. In this talk, we shall provide a brief introduction to Bayesian Statistics, which will then be followed by an overview of our work. We will use the Gaussian Linear Model in which case the Bayesian Evidence can be calculated analytically, to show that our method is also effective.
Name
Iniyan Natarajan
Abstract Title
Bayesian Inference for Radio Observations (BIRO) - Working in the visibilities
Abstract
The imaging of interferometric data is fraught with difficulty because it introduces deconvolution artifacts, correlates noise and is difficult to automate. In contrast, a Bayesian analysis directly in the visibility domain - which we call BIRO - is conceptually simple and mathematically consistent, but computationally challenging. We argue, however, that Bayesian visibility analysis is a complementary goal worth pursuing because of its ability to shed light on systematics which is hard to accomplish in traditional imaging, if at all it can be done. Moreover, a GPU-enabled BIRO (using montblanc) shows promise when applied to targeted observations. I will motivate for and outline the BIRO framework, illustrating it with examples of sky-telescope coupling and super-resolution in WSRT and EVN observations.
Name
Simon Perkins
Abstract Title
Montblanc: GPU-Accelerated Radio Interferometer Measurement Equations
Abstract
Montblanc is a Python package implementing the Radio Interferometer Measurement Equation (RIME). The RIME is a highly parallel equation and is therefore suited to GPU implementation. A familiar Numpy interface is provided to users, offering them the ability to modify input arrays and inspect output arrays. Expensive compute is directed to available GPUs through the PyCUDA package. Montblanc computes model visibilities on the GPU and was originally designed to compute chi-squared values from observed visibilities transferred onto the GPU. Recently, work has been done to allow Montblanc to act as a simulator by supporting the streaming of visibilities off the GPU. It currently supports Direction-Dependent Effects through a holographic beam cube and Direction-Independent Effects via a time, antenna and channel varying NumPy array. The architecture allows for the addition of new source types and currently supports point, gaussian and sersic sources.
Name
Adrianna Pińska
Abstract Title
Adapting existing visualisation tools for use with big data
Abstract
Many visualisation packages in the astronomical software ecosystem are limited by design decisions which did not anticipate the need to scale to the multi-terabyte-sized images which we expect the SKA project to produce. Adapting these tools for larger data sizes is a challenging task. In my presentation I will give an overview of some of the issues to be considered, including support of newer data formats such as HDF5 and use of parallel algorithms.
Name
Bart Scheers
Abstract Title
Back-end systems for time-domain astronomy
Abstract
With the advent of time-domain astronomy, telescope designs become more and more integrated with back-end hardware and software systems. A considerable part of the science output relies on the computation, data storage and access facilities. Apart from their own key-science goals, most instruments should be capable of carrying out all-sky surveys, searching for transient and variable sources and cataloguing repeated measurements of millions of sources. Therefore, scientific discoveries depend increasingly on the ability to efficiently grind massive amounts of experimental data using database technologies: the database as a new lab. However, databases have difficulties in supporting aggregations of petabytes or providing responses to ad-hoc queries in near-real time. In this talk I will discuss the general functionality of automated software pipelines that process streams of calibrated telescope data and the back-end systems that store the data. The main operations are inserting new data into database, associating new sources with known sources, inspecting the data for interesting sources. I will highlight fast methods of database insertion, the usage of statistical sky models for the complex source association algorithms and ways to speed up this process, where the set of operations involved serve as performance benchmarks. During the telescope's lifetime a large archive may be built up, which should be available for complementary science. Therefore, it is of importance that this is taken into account into the design phase as well. Finally, I will mention some database distribution approaches in a large cluster that we are currently investigating.
Name
Mattia Vaccari
Abstract Title
Multi-Walevelength Data Fusion & Data Mining for MeerKAT Extragalactic Survey Science
Abstract
The increasing size and complexity of contemporary and near-future multi-wavelength surveys have recently revoluzionized Galaxy and AGN evolution studies. While fully automated data reduction and statistical data analysis will be required for most future wide-area projects such as the SKA and the LSST, currently available tools are often inadequate to address the upcoming data deluge in a timely manner. This is particularly important for the MeerKAT Key Survey Project which will have a limited amount of time to deliver on its science goals before the advent of the SKA. I will argue how IDIA should enable work on a few key aspects which will be necessary to timely deliver MeerKAT Extragalactic (LADUMA/MIGHTEE) Survey Science, and namely: - Multi-Wavelength Data Fusion & Metadata Characterization - Multi-Wavelength Source Detection and Measurement - Data Mining for Source Identification and Classification - Next-Generation Stacking Approaches to go Beyond Confusion - Citizen Science Approaches e.g. via the Zooniverse Platform
Name
Patricia Whitelock
Abstract Title
The Virtual Observatory and Multiwavelength Astronomy
Abstract
I will discuss broad developments in data access and interoperability within multiwavelength astronomy. The emphasis will be on the challenges presented by developments in SA and internationally. The LSST, in particular, offers special challenges which must be dealt with in parallel to those of SKA. The International Virtual Observatory Alliance (IVOA) meeting to be held in South Africa next month will offer us a special opportunity to find out how the major astronomy projects, in space and on the ground, are anticipating dealing with interoperability.
Name
Patrick Woudt
Abstract Title
The data challenges of radio transient surveys - a ThunderKAT perspective
Abstract
I will give an overview of the ThunderKAT large survey project on MeerKAT. One of ThunderKAT's primary aims to find, report and characterise radio transients in real-time. I will describe the classes of known and expected astrophysical transients in the MeerKAT data stream, and highlight work done on the current generation of precursor facilities in preparing for the data challenges of real-time studies of the transient sky.
Name
Jonathan Zwart
Abstract Title
A Survey of Guerrilla Radio Astronomy
Abstract
What we want. The traditional toolkit. Thinking outside the toolbox. Bayesvertorial. Lessons from the Cosmic Microwave Background. A candidate for 4th Generation Calibration. Complementarity of imaging and visibility domain analyses. Statistical techniques as a limiting factor. Sky and telescope couple. Simulations are key. Avoiding detection. Undercover astronomy. What can be done. What can't be done. Visibilities again. Confusion and the wall(s). Not just total-power continuum. Simulations again. Where next?