1. Data-centric view of algorithms displayed in the ACES Vislab
CDGC Research
Applications
Reduce Energy Requirements of Computer Programs
Security Applications that Need Real-Time Analysis of Provenance Graphs
Machine learning
Mesh Generation
Algorithms & Modeling
Galois System
Data-centric Programming
Inexact Computing
Funding: NSF, IBM, Intel, DARPA
Research Staff:
Andrew Lenharth
Sreepathi Pai
Graduate Students:
10 PhD Students from
ICES, CS, and ECE
Keshav Pingali
Director
Brief description of some of the research activities in PADAS:
Fast algorithms for robust (i.e., automatic, black box) analysis for medical images.
Typically, problem-specific algorithms
Top left image: MRI FLAIR Image of patient with Grade IV glioma.
Top right image: Estimated segmentatio to Grey matter, white matter, cerebrospinal fluid, and tumor
Bottom image:
Spatial frequency of adult glioblastoma with 3D surface rendering (top) and 2D multiplanar slices (bottom). These tumors tend to appear much more frequently in some brain regions, and this depends on age. The Statistical atlas was constructed with registration algorithms.
Data-centric view of algorithms displayed in the ACES Vislab

2. Performance of Galois system:
Galois Project - CDGC
Goal:
Make it easier for applications programmers to use heterogeneous parallel computers (CPUs, GPUs, FPGAs)
Strategy:
Raise abstraction level of programming
Implement compilers and runtime systems to get efficient execution
Focus:
Irregular applications, particularly those involving large graphical models
Application areas: graph analytics, machine-learning, mesh generation
Software:
Galois system
Data-centric programming model implemented in C++
Used by BAE, HP, and university groups all over the world
Funding:
NSF, IBM, Intel, and other companies
Roughly $5 million in past 10 years
Performance of Galois system:
512-core SGI Ultraviolet

3. RIPE Project - CDGC
Rapid Identification and Prevention of Exfiltration (RIPE)
Goal:
Use CDGC-developed Galois graph processing system for security applications that need real-time analysis of provenance graphs
Identifying clusters in social networks
Funded by DARPA Transparent Computing Program ( )
Sub-contractor to BAE Systems
$5.5 million total, UT portion is $1 million

4. Proteus Project - CDGC Goal: Funded by DARPA BRASS program (2016-2020)
Images can often be rendered inexactly to reduce
energy requirements without losing semantic content
Goal:
Exploit inexact computing and machine learning to reduce energy requirements of programs
Funded by DARPA BRASS program ( )
Building Resource Adaptive Software Systems
Joint project with Rice, MIT, and Chicago
$8 million total, UT portion is $1.725 million