Blair Adamache - Electrical & Computer Engineering

advertisement
IBM Big Data Projects
with Ontario Universities
July 16, 2014
0
© 2012 IBM Corporation
Focus Area
Phase
Focus
#Project
SME
Phase1
FastStart
7
30%
Phase2
Academic-led
24
70%
Phase3
Industry and
Academic-led
11
90%
Status: 25 running/initiated, 13 scheduled 3Q, 2 Q114, 2 TBD*
Institution
Health
20
Energy (* new Mining)
8
Water
5
Cities
4
Agile
5 + 6 multi = 13
#Projects
Platform
McMaster University
4 (all Agile)
University of Ottawa
3 (1 Agile)
Queen’s University
4 (0 Agile)
University of Ontario
3 (0 Agile)
University of Toronto
12 (3 Agile)
University of Waterloo
6 (1 Agile)
Western University (1 IBM lead Mining)
9 (4 Agile)
Carleton University (IBM Lead)
1 (0 Agile)
TOTAL
1
#Project
42
#Project
Blue Gene/Q
15
Cloud
13
Agile
13
Multi-Platform
6 (1 non SOSCIP)
* 2 projects deferred pending Sustainability plan + resources
** 1 project on hold pending UofT/UHN/IBM IP agreement
NANOPHOTONIC DEVICES FOR EARLY DISEASE DETECTION
Create a computational model of nanophotonic devices that could improve the early detection of disease at a cellular level.
HEALTH
PANDEMIC MODELING, PREDICTION, AND CONTROL
Based on citizen behaviors and local GTA health care policies, create models to help decision makers control a pandemic
outbreak of a contagious disease.
INFECTIOUS DISEASE MODELING SOFTWARE
Develop a software package that enables easier mathematical modeling of certain factors of infectious disease, such as infection
rates, incubation periods, and initial conditions for a new infection.
COMMON SIGNATURES IN LUNG DISEASE
Design an algorithm to identify common signatures in lung cancer patients, advancing diagnosis, prognosis and treatment
capabilities.
INTEGRATED RADIOLOGY AND PATHOLOGY
Leverage cloud-based smart platform solutions to integrate medical imaging from radiology, pathology, and oncology
departments, providing a patient-centred, holistic approach to the diagnosis and treatment of cancer.
DETECTING RADIATION EXPOSURE
Develop software that will enable rapid, accurate detection of radiation exposure for mass numbers of people.
REAL TIME ANALYSIS OF HUMAN BRAIN NETWORKS
Apply stream analytics to functional MRI data to analyze brain activity in near real time to improve patient experience and reduce
medical costs and timelines.
ARTEMIS EXPANSION
Build a cloud-based service using streaming analytics to predict the health status of individuals, particularly premature children.
Bring sophistication of urban teaching hospitals to remote communities and extend early detection of infections to adult ICU.
BUILDING AND CERTIFYING SAFE AND SECURE INSULIN PUMPS
Build a toolkit for software certification. First application will target medical devices.
DATA PRIVACY AND SHARING MEDICAL DATA
Develop new healthcare privacy and security framework to address both patient care and medical studies
ANALYTICS AS A SERVICE
Develop new toolkit to support ultra large scale services for big data analytics. Initial target is health applications but toolkit will
be applicable to other domains.
2
ENERGY
3
RENEWABLE SOLAR ENERGY
Develop a new, low-cost, paint-on, solar cell-based renewable energy.
COMBUSTION SYSTEM SIMULATION
Develop a high-performance computing algorithm that simulates combustion systems to improve the design of gasturbine engines in transportation and power-generation applications.
CREATING SUSTAINABLE ENERGY FROM ARTIFICIAL PHOTOSYNTHESIS
Examine artificial photosynthesis and water-splitting at a nanoscale level to better understand the chemical and
physical properties of these processes. This level of understanding will contribute to the development of clean,
renewable, and sustainable sources of energy in Ontario.
SMART METER DATA ANALYTICS
Develop software for small/medium enterprises which will help to identify smart ways to reduce energy
consumption.
WIND-FORECASTING MODEL
Develop a wind-forecasting model to enable renewable wind energy generation in Ontario to be more proactive and
operationally cost effective.
A MODEL FOR SHORT AND LONG TERM ENERGY PLANNING
Improve models for power distribution reliability building on IBM’s weather forecast model.
WATER
REAL-TIME, REMOTE SENSOR, WATER SHED DATA ANALYSIS
Design communications software that will transmit real-time data from remote
sensors in the Grand River Watershed. The data will improve both the
understanding of water behavior, as well as water management tactics.
CLIMATE CHANGE IMPACT ON WATER RESOURCES
Develop a 3D hydrological model that represents the impact of the climatechange forecast on the quality and quantity of the surface and subsurface water
resources in the Grand River Watershed.
WATER QUALITY MONITORING
Create a low-cost, easy-to-use, real-time sensor system for water
quality monitoring, including biological and chemical contamination detection.
MISSION CRITICAL INFRASTRUCTURE MONITORING
Create disaster management response systems which have a clear
understanding of interdependencies of all aspects of risk mitigation, disaster
preparedness and post disaster planning. Develop reliable monitoring systems
for the real-time detection of mission-critical infrastructure failures.
REAL-TIME DRINKING WATER MANAGEMENT/
MONITORING
Create a real-time water data processing system to aid in water contamination
alerts, ensuring enough water is supplied to the public, and to control
residential water use.
4
CITIES
SMART URBAN SYSTEM DESIGN
5
Research into transportation and urban activity systems in the Greater
Toronto Area (GTA), improving the decision-making ability of urban planning
designers in Ontario.
AUTOMATIC DETECTION OF MAN-MADE OBJECTS FROM
IMAGE DATA
Develop a cloud-based tool that will automatically identify features (buildings,
roads, forest types, etc.) from high-resolution image data for use in areas like
urban planning or forest management.
CLIMATE MODEL
Create new detailed climate projection and drive hydrological models to
assess impact of global warming using dynamic downscaling specific to the
Grand River watershed.
MAKING HIGH-PERFORMANCE COMPUTING ACCESSIBLE
AGILE
Design a program that enables software application developers, with minimal hardware skills, to leverage agile, highperformance computing, resulting in faster development cycles.
6
ASTRONOMICAL DATA MINING
Leverage agile, high-performance computing to boost the processing capability of data obtained from Ontario’s
world-leading astronomical radio telescope at the Algonquin Radio Observatory.
REAL-TIME NETWORK CAPACITY ADJUSTMENT
Leverage agile computing to develop an accelerated ray-tracing algorithm, allowing network operators to adjust
network capacity in real-time based on changes in network state and improving quality of service.
IMPROVING SMART GRID DATA EXCHANGE
Improve the security and efficiency of handling massive data exchanges in smart grid infrastructures through
hardware accelerated computing.
DESIGN PATTERNS FOR HETEROGENEOUS COMPUTING
Develop a set of design patterns that will make high performance computing more accessible to software
developers creating complex applications.
HARDWARE ACCELERATION THROUGH AGILE COMPUTING
Using agile computing, develop a hardware-based system that will accelerate the ability of the IT industry to
solve optimization problems, such as routing and scheduling of airplanes and urban transportation systems.
SME Lead
7
PATIENT-CENTRED UNIVERSAL HEALTH RECORDS
Develop a cloud-based solution to aggregate, analyze, and standardize
patient health records.
WEATHER PROJECTIONS FOR SMART CITIES
Integrate high resolution weather projections with cities infrastructure
(buildings, transportation networks, etc.) to improve their design,
sustainability, and resiliency.
PREDICTING LEUKEMIA INHIBITORS
Develop a tool that simulates molecular behaviour to accelerate the
selection of drugs for the treatment of leukemia.
IBM Lead
8
IN SILICO PROTEIN SYNTHESIZER
Perform genetic analysis to sequence proteins that aid in drug design. This
will help to better understand diseases and treatments.
ANALYZING GEOSPATIAL PATTERNS IN THE CLOUD: APPLICATION
TO MINERAL EXPLORATION AND MINING IN CANADA
Aggregate and perform statistical analysis, data-mining, scoring and ranking,
using Monte Carlo methods and Bayesian statistics to identify areas of
promise with requiring extraction of new core samples.
NEW
PROJECTS
CYTOGENETIC DECISION-SUPPORT TOOL
Design and implement a decision-support tool to assist
cytogeneticists in the selection of appropriate probes to improve
speed and accuracy of DNA microarray testing for the diagnosis and
treatment of chromosome/cell related disorders.
PHOTODYNAMIC CANCER THERAPY
Develop new, minimally invasive, photodynamic therapy for the
treatment of head and neck cancers.
MISSION-CRITICAL INFRASTRUCTURE MONITORING
Develop real-time detection of mission critical infrastructure
failures, such as loss of energy supply, water contamination, dam
failure, or the collapse of structures such as buildings or bridges.
NEWEST
PROJECTS
PREDICTIVE CARDIOTOXICITY USING MACHINE LEARNED
MODELING FROM SIMPLE BIOLOGICAL INPUTS
Use techniques that identify novel ECG patterns via advanced
mathematical techniques that assess dynamic
alterations in cardiac conduction and repolarization along with
alterations in vascular and autonomic function.
CHARACTERIZATION OF PROTEIN-DRUG INTERACTION
NETWORKS FOR RARE-DISEASE REPURPOSING (FDA)
Structural alignment of test sets of millions of compounds,
molecular dynamics simulations of involving tens or hundreds of
proteins from the FDA’s Rare Disease Repurposing Database.
QUANTIFIED PSYCHIATRY: TIME SERIES PREDICTIVE
MODELING IN MENTAL HEALTH DIAGNOSTICS
Perform analysis of pharmaceutical clinical trial data for
antidepressant medications and gather behavioural markers to
segment psychopathology. Analysis will generate insight into
patterns underlying patient behaviour, and understand placebo
response in the context of these trials.
FPGAs and Big Data
Dimensions of Parallelism
Direct data injest

Network

Storage
Pipeline
Parallelism
I/O attachment, or
Coherent attachment via CAPI
Multiple pipelines
to Host CPU
Multiple kernels/functions
Performance
Floating Point Performance
Power Consumption
GFLOPS
Power (W)
300
200
100
0
GPU
FPGA
7000
6000
5000
4000
3000
2000
1000
0
40
30
~2×
20
10
GFLOPS/W
400
CPU
Enabling Technologies
(single-precision fused-multiple-add)
Lime
0
CPU
GPU
FPGA
POWER8 CAPI
© 2012 IBM Corporation
Real-time fMRI Brain Analytics
Mark Daley, Western University (London, ON)
 The problem: brain activity scans take days to analyze
 The solution: a real-time analytics engine
FPGA replaces 48 x86 cores and
implements superior motion correction
algorithm
IBM InfoSphere Streams on Power 7
constructs graphs of brain networks 40x
faster than single process on x86
Graph updates every 0.6-0.8s
Planning replacement of CPU-based
graph analytics with Power 8 and
CAPI-attached FPGA accelerator
Results in seconds instead of days!
© 2012 IBM Corporation
Download