Institute of High Performance Computing

navigation

Powerding Discoveries!

Research

----

Computing Science (CS) Department

>>Department Members

Researchers in the CS Department develop techniques that draw out the efficiency, insight and intelligence in computing to power scientific discovery and technological advances. R&D activities are undertaken to:

  • Develop techniques & tools to enhance human productivity in software optimisation for heterogeneous architectures:
    • Refactoring for optimisation
    • GPGPU techniques and applications
    • Runtime adaptation techniques

  • Develop reliable, adaptive and automated computing platforms for distributed applications and services:
    • Cloud computing for engineering and data analytics services
    • Fault tolerance techniques for engineering and data workflow
    • Cloud economics and service management

  • Create capabilities in multi-domain, large-scale data modelling, analytics and simulation:
    • Large-scale data analytics for life sciences and enterprises
    • Applied cross-domain-driven learning
    • Smart automation for analytics design

  • Develop 3D & 4D geometrical modelling and shape analysis algorithms:
    • Anatomical modelling & simulation
    • Computational geometrical modelling & analysis
    • Automatic finite element mesh generation technologies
    • Virtual surgical training

  • Create intelligent systems that interact in a believable, intuitive and socially appropriate manner:
    • Computational modelling of human cognition and emotion
    • Human-computer interaction and computer-assisted negotiation
    • Social awareness and social cognition for AI agents
The CS Department consists of five Capability Groups that oversees the five key research areas:

High Performance Computing (HPC)

Distributed Computing (DC)

Cross-disciplinary Data-intensive Analytics (CDA)

Geometrical Modelling (GM)

Computational Social Cognition (CSC)

High Performance Computing

The High Performance Computing (HPC) group aims to design and develop a generic programming and system framework that efficiently utilises heterogeneous architectures and accelerators to solve computational problems faster.

Research Directions / Challenges
Our R&D in productivity-enhancing tools enables HPC developers to perform source code transformations:

  • Tool to visualise code structure as a design aid to the developer
  • Tools for integrated code transformation as an alternative to manual editing
  • Computer-aided source code transformation for migrating legacy codes to HPC systems
  • Memory-hierarchy optimisation techniques for GPUs
  • Runtime adaptive tuning framework for new generation of HPC systems
We also provide consultancy and training to enable HPC developers to make better use of next-generation computing platforms:
  • Collaborate with researchers and industry partners to optimise and scale promising applications with significant performance gain
  • Establish programming know-how for multicore and GPUs
  • Train and develop talent pool of HPC application engineers
  • Provide a knowledge-base on various parallel programming models such as MPI, OpenMP, CUDA, etc.

Figure 1: Visualising code structure as a design aid to the developer. Above is an example showing data dependencies of a simple set of nested loops performing Gaussian elimination on a matrix.


Distributed Computing for Cloud/Grid Services

The Distributed Computing group aims to establish new paradigms in the way computing technologies and platforms are being used. The research areas involve:

  • Design and implementation of end-to-end web-based solutions to power scientific discovery and to enhance efficiency in solving industry problems
  • New service-oriented paradigms and techniques in cloud computing such as software-as-a-service, MapReduce programming, etc.
  • Development of collaborative technologies to enable researchers and engineers to cooperate with each other remotely and effectively

Research Directions / Challenges

With the increasing prevalence of HPC-as-a-Service (HaaS), the group undertakes research to investigate how we can transform traditional HPC usage into pervasive, scalable and commercially-viable services. In particular, we are concerned about how HPC services can be established, valued, managed and used in a scalable and adaptive manner. Research activities explore:

  • New ways in which parallel and distributed HPC services can be mapped into a cloud-based service marketplace
  • New paradigms in negotiating service contracts and pricing services in a marketplace
  • Effective techniques in managing, adapting and scheduling services in a multi-provider ecosystem that help to maximise service quality and utility to both clients and providers
  • Reliable and cost-effective methods of automating the execution of data-aware workflow in hybrid clouds

Figure 2: Research in Distributed Computing group.


Figure 3: Fusion Cloud showing the submission portlet, remote visualisation portlet, data Management portlet and administration portlet with job characteristic visualisation


Cross-disciplinary Data-intensive Analytics

Over the past two decades, we have witnessed an exponential growth of data in a multitude of fields such as the life sciences and enterprises. While the amount of data has grown through better acquisition devices, the methodologies to analyse and derive insight from these vast amount of data has not kept up, resulting in a data deluge.

The Cross-disciplinary Data-intensive Analytics (CDA) group researches on analytics methodologies that reduce time- and space-complexity of computations, and establishes frameworks and integrate tools to generate insight quickly and accurately from large-scale data and apply them to real problems.

Research Directions / Challenges
  • To investigate methodologies that reduces time- and space-complexities in order to address the challenge of large data sets
  • To establish simple and reusable analytics framework to capture the know-how of domain experts and make their job more efficient through reuse
  • To design automated and optimised analytics workflow so that non-domain experts are confident in using the analytics workflow, through appropriate prompts and advice from the workflow designer system

As the research frontier increasingly focuses on the interfaces of different disciplines, the multi-disciplinary approach is all the more important. Hence, there is a need to move from single-domain to cross-domain analytics. For example, in translational research, it is often necessary to investigate both clinical as well as biological data.


Figure 4: Smart automated analytics design for enabling layman to perform large-scale cross-domain data analytics


Geometrical Modelling

The Geometrical Modelling group engages in R&D activities to bring the science of computational geometry and digital modelling to various domains such as computational science and biomedical engineering. The aim is to build a full range of capabilities to answer difficult challenges in modelling shapes on computers. These include:

  • Interpretive techniques to handle and convert raw multi-domain digital information into appropriate input for simulation-driven and scientific applications
  • Fundamental algorithms in the manipulation of discrete digital models, such as mesh reconstruction, simplification & enhancement, and deformation
  • Computational geometry techniques to extract surface and volumetric mesh properties which can provide domain -specific insights

target="_blank">

Figure 5: Geometrical Modelling capabilities in computational science and biomedical domains.

Research Directions / Challenges

  • Develop fundamental algorithms for processing 3D/4D mesh models, especially for human anatomical model in biomedical applications
  • Establish shape analytics methodologies for analysing mesh properties of 3D/4D anatomical digital models to extract clinical insight
  • Utilising realistic virtual environments and scenery for education, training simulations or augmentation of live operative procedures.

Computational Social Cognition

The Computational Social Cognition group is a new, cognitive science endeavour with an initial focus on designing architectures and building models to explore the interplay between social and cognitive processes underlying socially appropriate, coherent, and understandable behaviour.

Computational models of these processes can be used to generate such behaviours in agent-based artificial systems which could also interact in a believable, intuitive and socially appropriate manner with human users.

The group involves an interdisciplinary team of psychologists, computer science and artificial intelligence researchers, roboticists, cognitive and social psychologists, and computational linguists.

Figure 6: Research inComputational Social Cognition group.


Dr. Rick GOH Siow Mong
Department Director



This page is last updated at: 11-Mar-2012