Pioneering and Democratizing Scalable HPC+AI - HPC Advisory ...
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Pioneering and Democratizing Scalable HPC+AI
Nick Nystrom Paola Buitrago
Interim Director, PSC Director, AI & Big Data, PSC
nystrom@psc.edu paola@psc.edu
2019 Stanford Conference · Stanford · February 15, 2019
©1 2019 Pittsburgh Supercomputing Center © 2019 Pittsburgh Supercomputing Center
Outline
Motivation & Vision
Realizing the Vision: Bridges and Bridges-AI
Exemplars of Success
Summary
2
What is PSC?
National service provider for research and discovery Active member in the CMU and Pitt communities
• Bridges, Anton 2, Brain Image Library, • Research collaborations
Open Compass, XSEDE, Olympus • Colocation for lower cost and greater capability
Brain Image
Bridges Anton 2 Library
PSC is a joint effort of
Carnegie Mellon University
Research institution advancing knowledge and the University of Pittsburgh.
Networking and security
through converged HPC, AI, and Big Data 33 years of leadership in HPC, • Networking & security service provider
• ~30 active funded projects HPDA, and computational science.
• Research networking
21 HPC systems, 10 of which
were the first or unique.
Pioneering the convergence
Education and training of AI + HPC + data.
• Lead national & local workshops Advise and support industry
• Support courses at CMU and elsewhere • Training, access to advanced resources,
• Teaching, thesis committees, interns collaborative research
3
Research Needs Converged HPC, AI, and Data
Pan-STARRS telescope
http://pan-starrs.ifa.hawaii.edu/public/
Social networks and the Internet Video Wearable Sensors Detecting Cancer The Human BioMolecular Atlas Program
Wikipedia Commons F. De Roose et al., https://research.googleblog.com/2017/ https://commonfund.nih.gov/hubmap
https://techxplore.com/news/2016-12-smart-contact- 03/assisting-pathologists-in-
lens-discussed-electron.html detecting.html
Genome sequencers
(Wikipedia Commons)
Collections Legacy documents Environmental sensors: Water temperature Library of Congress stacks
BlueTides astrophysics simulation Horniman museum: http://www.horniman.ac.uk/ Wikipedia Commons
profiles from tagged hooded seals https://www.flickr.com/photos/danlem2001/6922113091/
http://bluetides-project.org/ get_involved/blog/bioblitz-insects-reviewed http://www.arctic.noaa.gov/report11/biodiv_whales_walrus.html
Structured, regular, Unstructured, irregular, heterogeneous
homogeneous
4
Enabling the Creation of Knowledge
Objectives Common Goal
Enable data-intensive applications & workflows Enable the creation of knowledge
• Deliver HPC Software as a Service • Democratize HPC, Big Data, and AI
(Science Gateways) • Enable research areas that have not
• Deliver Big Data as a Service (BDaaS) previously used HPC
• Provide scalable deep learning, machine learning, • Advance previously traditional fields
and graph analytics through machine learning and data
• Support very large in-memory databases analytics
• Facilitate data assimilation from instruments and • Couple applications in novel ways
the Internet
Scale beyond the laptop and to
interdisciplinary, collaborative teams
5
The Rapid Growth of AI
From: Artificial Intelligence Index: 2018 Annual Report (Stanford University, 2018)
6
Outline
Motivation & Vision
Realizing the Vision: Bridges and Bridges-AI
Exemplars of Success
Summary
7
• Available at no cost for open research and
courses and by arrangement to industry
• Easier access for CMU and Pitt faculty through
the Pittsburgh Research Computing Initiative
• 29,036 Intel Xeon CPU cores
• 216 NVIDIA GPUs: 64 K80, 64 P100, 88 V100
• 17 PB storage (10 PB persistent, 7.3 PB local)
• 277 TB memory (RAM), up to 12 TB per node
• 44M core-hours, 173k GPU-AI-hours,
442k GPU-hours, and 343k TB-hours
allocated quarterly
• Serving ~1,850 projects and ~7500 users at
393 institutions, spanning 119 fields of study
• Bridges-AI: NVIDIA DGX-2 Enterprise AI
system + 9 HPE 8-Volta Apollo 6500 Gen10
servers: total of 88 V100 GPUs
Bridges converges HPC, AI, and Big Data to empower new research communities, bring desktop convenience
to advanced computing, expand remote access, and help researchers to work more intuitively.
• Funded by NSF award #OAC-1445606 ($20.9M), Bridges emphasizes usability, flexibility, and interactivity
• Available at no charge for open research and coursework and by arrangement to industry
• Popular programming languages and applications: Python, Jupyter, R, MATLAB, Java, Spark, Hadoop, …
• 856 compute nodes containing Intel Xeon CPUs and 128GB (800), 3TB (42), and 12TB (4) of RAM each
• 216 NVIDIA Tesla GPUs: 64 K80, 64 P100, (new) 88 V100 configured to balance capability & capacity
• Dedicated nodes for persistent databases, gateways, and distributed services
• The world’s first deployment of the Intel Omni-Path Architecture fabric
8
Acquisition and operation of Bridges are made
possible by the National Science Foundation
through award #OAC-1445606 ($20.9M):
Bridges: From Communities and Data to
Workflows and Insight
delivered Bridges, and is now
delivering Bridges GPU-AI
All trademarks, service marks, trade names, trade dress, product names, and logos appearing herein are the property of their respective owners.
9
Bridges Makes Advanced Computing Easy
Make HPC accessible to all research communities
Converge HPC, AI, and Big Data
Support the widest range of science with an extremely rich
computing environment
• 3 tiers of memory: 12 TB, 3 TB, and 128 GB
• Powerful, flexible CPUs and GPUs
• Familiar, easy-to-use user environment:
Elements not available in traditional supercomputers
– Interactivity
– Popular languages and frameworks:
Python, Anaconda, R, MATLAB, Java, Spark, Hadoop
– AI frameworks: TensorFlow, Caffe2, PyTorch, etc.
– Containers (e.g., NGC) and virtual machines (VMs)
– Databases
– Gateways and distributed (web) services
– Large collection of applications and libraries
10
10Conceptual Architecture
Users,
XSEDE,
campuses,
Web Server Database Data Transfer Login instruments
nodes nodes nodes nodes
Intel Omni-Path
Parallel File Management
Architecture
System nodes
fabric
Bridges-AI
Introduced in
RSM Nodes
ESM Nodes LSM Nodes
128GB RAM NVIDIA DGX-2 Operations Year 3
12TB RAM 3TB RAM (16 V100 GPUs)
800 nodes,
4 nodes 42 nodes 9x HPE A6500
48 with GPUs (9x 8 V100 GPUs)
1120 Storage Building Blocks, Project &
implementing the parallel Pylon community
storage system (10 PB usable) datasets
4 HPE Integrity
Superdome X (12TB)
compute nodes … Large-
… each with 2 gateway nodes memory
Java &
4 MDS nodes Python
2 front-end nodes 42 HPE ProLiant DL580 (3
Representative 2 boot nodes TB) compute nodes
uses for AI 8 management nodes
12 HPE ProLiant DL380
database nodes User interfaces for
6 “core” Intel® OPA edge switches:
fully interconnected, 6 HPE ProLiant DL360 AIaaS, BDaaS
2 links per switch web server nodes
Robust paths to 20 “leaf” Intel® OPA edge switches
parallel storage Intel® OPA cables
Distributed training, Spark, etc.
Purpose-built Intel® Omni-Path Deep
32 HPE Apollo 2000 (128GB) GPU nodes
Architecture topology for Learning
with 2 NVIDIA Tesla P100 GPUs each
data-intensive HPC
Bridges-AI Maximum-Scale Deep Learning
32 RSM nodes, each with
2 NVIDIA Tesla P100 GPUs NVIDIA DGX-2 and
9 HPE Apollo 6500
Gen10 nodes:
16 RSM nodes, each with 2 NVIDIA Tesla K80 GPUs 88 NVIDIA Tesla
748 HPE Apollo 2000 (128GB) V100 GPUs
compute nodes
16 HPE Apollo 2000 (128GB) GPU nodes
ML, inferencing, DL development, with 2 NVIDIA Tesla K80 GPUs each Bridges Virtual Tour:
Spark, HPC AI (Libratus)
Simulation (including AI-enabled) https://psc.edu/bvt
12
12Accessing Bridges: No Cost for Research & Education and
Cost-Recovery Rates for Corporate Use
The following annual allocations are renewable and extendable, also at no cost for research and education.
Open Research Industry
PSC Corporate Program
Startup Research Education
Cost No charge No charge No charge Cost recovery rates
CPU-hours 50k Up to ~107 Up to ~106 Up to ~18M
GPU-hours 2500 Up to ~105 Up to ~104 Up to ~180k
GPU-AI hours 1500 Up to ~105 Up to ~104 Up to ~69k
TB-hours 1000 Up to ~104 Up to ~104 Up to ~137k
Developer Yes Yes (Yes) Yes
Accepted Any time Quarterly Any time Any time
Awarded ~1-2 days Quarterly ~1-3 days ASAP
13Interactivity
Interactivity is the feature most frequently
requested by nontraditional HPC communities.
– Interactivity provides immediate
feedback for doing exploratory
data analytics and testing hypotheses.
– Bridges offers interactivity through a combination of shared,
dedicated, and persistent resources to maximize availability
while accommodating diverse needs.
14High-Productivity Programming
Supporting languages that communities already use is vital for them to
apply HPC to their research questions. This applies to both traditional and
nontraditional HPC communities.
15Gateways and Tools for Building Them
Gateways provide easy-to-use access to Bridges’ HPC and data resources, allowing users to
launch jobs, orchestrate complex workflows, and manage data from their browsers.
– Provide “HPC Software-as-a-Service”
– Extensive use of VMs, databases, and distributed services
Galaxy (PSU, Johns Hopkins) The Causal Web (Pitt, CMU) Neuroscience Gateway (SDSC)
https://galaxyproject.org/ http://www.ccd.pitt.edu/tools/
16Databases and Distributed/Web Services
Dedicated database nodes power persistent relational
and NoSQL databases
– Support data management and data-driven workflows
– SSDs for high IOPs; HDDs for high capacity
(examples)
Dedicated web server nodes
– Enable distributed, service-oriented architectures
– High-bandwidth connections to XSEDE and the Internet
17Bridges-AI: Overview
• 1 NVIDIA DGX-2 Volta introduces Tensor Cores to
Tightly couples 16 NVIDIA Tesla V100 (Volta) GPUs accelerate neural networks, yielding
extremely high peak performance
at 2.4 TB/s bisection bandwidth, to provide maximum
for appropriate applications.
capability for the most demanding of AI challenges
Bridges-AI providea massive
• 9 Hewlett Packard Enterprise Apollo 6500 Gen10 servers aggregate performance:
• 9.9Pf/s mixed-precision tensor
Each with 8 NVIDIA Tesla V100 GPUs connected by
• 251Tf/s 32-bit
NVLink 2.0, to balance great AI capability and capacity
• 125Tf/s 64-bit
• Bridges-AI is integrated with Bridges and allocated through
XSEDE as resource “Bridges GPU-AI”, analogous to Bridges GPU, RM, LM, and Pylon
• Bridges-AI adds 9.9 Pf/s of mixed-precision tensor, 1.24 Pf/s of fp32, and 0.62 Pf/s of fp64.
(Totals: 9.9 Pf/s tensor, 3.93 Pf/s fp32, 1.97 Pf/s fp64).
• The $1.786M supplement includes additional staffing to support solutions and scaling
• Deployment: Bridges-AI deployed on time. PSC ran an Early User Program from November-
December 2018, and production operations began January 1, 2019.
18The Heart of Bridges-AI: NVIDIA Volta
New Streaming Multiprocessor (SM) architecture, introducing
Tensor Cores, independent thread scheduling, combined L1 data cache and
shared memory unit, and 50% higher energy efficiency over Pascal.
Tensor Cores accelerate deep learning training and inference, providing up to 12× and
NVIDIA Tesla V100 SXM2 Module 6× higher peak flops respectively over the P100 GPUs currently available in XSEDE.
with Volta GV100 GPU
NVLink 2.0 delivering 300 GB/s total bandwidth per GV100, nearly 2× higher than P100.
HBM2 bandwidth and capacity increases: 900 GB/s and up to 32GB.
Enhanced Unified Memory and Address Translation Services improve accuracy of
memory page migration by providing new access counters.
Cooperative Groups and New Cooperative Launch APIs expand the programming model
to allow organizing groups of communicating threads.
Volta-Optimized Software includes new versions of frameworks and libraries optimized
Training ResNet-50 with ImageNet: to take advantage of the Volta architecture: TensorFlow, Caffe2, MXNet, CNTK, cuDNN,
V100 : 1075 images/sa cuBLAS, TensorRT, etc.
P100 : 219 images/sb
K80 : 52 images/sb
a. https://devblogs.nvidia.com/tensor-core-ai-performance-milestones/
b. https://www.tensorflow.org/performance/benchmarks
19Balancing AI Capability & Capacity: HPE Apollo 6500
Bridges-AI adds 9 HPE Apollo 6500 Gen10 servers
Each HPE Apollo 6500 couples 8 NVIDIA Tesla V100 SXM2 GPUs
– 40,960 CUDA cores and 5,120 tensor cores
Performance: 1 Pf/s mixed-precision tensor, 125 Tf/s 32b, 64 Tf/s 64b
Memory: 128 GB HBM2, 7.2 TB/s aggregate memory bandwidth
HPE Apollo 6500 Gen10 Server
2 × Intel Xeon Gold 6148 CPUs and 192 GB of DDR4-2666 RAM
– 20c, 2.4–3.7 GHz, 27.5 MB L3, 3 UPI links
4 × 2 TB NVMe SSDs for user and system data
1 × Intel Omni-Path host channel adapter
Hybrid cube-mesh topology connecting the 8 V100 GPUs and 2 Xeon
CPUs, using NVLink 2.0 between the GPUs and PCIe3 to the CPUs
HPE Apollo 6500 Gen10
hybrid cube-mesh topology
20Maximum DL Capability: NVIDIA DGX-2
Couples 16 NVIDIA Tesla V100 SXM2 GPUs
– 81,920 CUDA cores and 10,240 tensor cores
Performance: 2 Pf/s mixed-precision tensor, 251 Tf/s 32b, 125 Tf/s 64b
Memory: 512 GB HBM2, 14.4 TB/s aggregate memory bandwidth
2 × Intel Xeon Platinum 8168 CPUs and 1.5 TB of DDR4-2666 RAM
– 24c, 2.7–3.7 GHz, 33 MB L3, 3 UPI links
NVIDIA DGX-2
2 × 960 GB NVMe SSDs host the Ubuntu Linux OS
8 × 3.84 TB NVMe SSDs (aggregate ~30 TB)
8 × Mellanox ConnectX adapters for EDR InfiniBand & 100 Gb/s Ethernet
The NVSwitch tightly couples the 16 V100 GPUs for capability & scaling
– Each of the 12 NVSwitch chips is an 18×18-port, fully-connected crossbar
NVIDIA DGX-2 with NVSwitch – 50 GB/s/port and 900 GB/s/chip bidirectional bandwidths
internal topology
– 2.4 TB/s system bisection bandwidth
21Deep Learning Frameworks on Bridges 22
Containers
Interoperability
with clouds and
Containers enable reproducible, cloud-interoperable workflows other resources
and simplify deployment of applications and frameworks
– PSC is a key partner of the Critical Assessment of Metagenome Interpretation
(CAMI) project for reproducible evaluation of metagenomics tools
– CAMI and the DOE Joint Genome Institute defined the biobioxes standard for
Docker containers encapsulating bioinformatics tools
Docker images can be converted to Singularity images and run on Bridges
– Certain vetted Docker containers are also supported
23Community Datasets
Some unique, others
with local caching for
• Hosting mature corpus of data and data tools efficiency and to drive
interdisciplinary
for an open science community research
– Accessible by multiple users, multiple groups.
– Provision of reusable data management tools
– Facilitate collaboration
– Offload data management
• Interoperable with HPC capabilities
– High speed data transfer
– High performance compute capabilities
• Support copies, maintenance, guarantee
integrity
• Data resource not subject to project
limitations
24The Expanding Ecosystem of Bridges
Hybrid on-prem
10s of PB data/AI/HPDA + Cloud
2.2 PB
Human BioMolecular Atlas Big Data for Better Health
Dedicated resources +
10 PB cloud use of Bridges
Brain Image Library Campus Clusters
25Big Data for Better Health (BD4BH)
Implementing, applying, and evaluating machine
learning methods for predicting patient outcomes of
breast and lung cancer
University of Pittsburgh Department of Biomedical
Informatics (Gregory Cooper), CMU Machine Learning
(Ziv Bar-Joseph) and Computational Biology (Robert
Murphy), and PSC (Nick Nystrom, Alex Ropelewski)
Dedicated 2.2 PB file system (/pghbio) attached to Bridges
for long-term data management & collaboration
Big Data research training opportunities: summer
program for Lincoln University students
26The Brain Image Library brainimagelibrary.org
Confocal Fluorescence Microscopy:
multispectral, subcellular resolution, highly quantitative
Will contain whole-brain volumetric images of mouse, rat, and
other mammals, targeted experiments highlighting connectivity
between cells, spatial transcriptomic data, and metadata
describing essential information about the experiments.
Supported by the National Institute of Mental Health of the
NIH under award number R24MH114793 ($5M).
Alex Ropelewski (PSC), Marcel Bruchez (CMU Biology),
Simon Watkins (Pitt Cell Biology & Center for Biologic Imaging)
Integrated with Bridges to support additional advanced analytics and
development of AI/ML techniques.
A. M. Watson et al., Ribbon scanning confocal for high-speed high-resolution volume
imaging of brain. PLoS ONE 12 (2017) doi: https://doi.org/10.1371/journal.pone.0180486.
27Human Biomolecular Atlas Program (HuBMAP)
Hybrid on-prem data/AI/HPDA + Cloud
“The Human BioMolecular Atlas Program (HuBMAP) aims
to facilitate research on single cells within tissues by
supporting data generation and technology development
to explore the relationship between cellular organization
and function, as well as variability in normal tissue
organization at the level of individual cells.” —NIH
The PSC+Pitt team was awarded development of the Infrastructure Component (IC) for the
HuBMAP HIVE (Integration, Visualization & Engagement)
– To receive data from Tissue Mapping Centers at Florida (lymphatic system), CalTech (endothelium),
Vanderbilt, Stanford, and UCSD (kidney, urinary tract, and lung)
– Supporting Tools Components at CMU and Harvard
– Supporting Mapping Components at Indiana University Bloomington and New York Genome Center
– Interfacing with the Collaboration Component at U. of South Dakota
– Supporting Transformative Technology Development centers at CalTech (single-cell transcriptomics),
Stanford (genomic imaging), Purdue (sub-cellular mass spec), and Harvard (proteomics)
28Outline
Motivation & Vision
Realizing the Vision: Bridges and Bridges-AI
Exemplars of Success
Summary
29AI for Strategic Reasoning
Tuomas Sandholm and Noam Brown, Carnegie Mellon University
An AI for making decisions with imperfect information:
Beating Top Pros in Heads-Up No-Limit Texas Hold’em Poker
Imperfect-info games require different
Prof. Tuomas Sandholm watching one of the
algorithms, but apply to important
world’s best players compete against Libratus.
classes of real-world problems:
– Medical treatment planning
– Negotiation
– Strategic pricing
– Auctions
– Military allocation problems
Libratus improved upon previous best
Heads-up no-limit Texas hold’em is the main algorithms by incorporating real-time
benchmark for games with imperfect information: improvements in its strategy.
– 10161 situations
Libratus was the first program to beat top humans
Beat 4 top pros playing 120,000 hands over 20 days
Libratus won decisively: 99.98% statistical significance
30AI for Strategic Reasoning
Tuomas Sandholm and Noam Brown, Carnegie Mellon University
“The best AI's ability to do strategic
reasoning with imperfect information has
now surpassed that of the best humans.”
—Professor Tuomas Sandholm,
—Carnegie Mellon University
1. N. Brown, T. Sandholm, Safe and Nested Subgame Solving for Imperfect-
Information Games, in NIPS 2017, I. Guyon et al., Eds. (Curran Associates, Awarded Best Paper at NIPS 2017
Inc., Long Beach, California, 2017), pp. 689-699.
2. N. Brown, T. Sandholm, Superhuman AI for heads-up no-limit poker: Libratus Companion paper in Science
beats top professionals. Science (2017) doi: 10.1126/science.aao1733.
Bridges enabled this breakthrough through 19 million core-hours of computing and 2.6 PB of data
in the knowledge base that Libratus generated.
Libratus, under the Chinese name Lengpudashi, or “cold poker master”, also won a 36,000-hand exhibition in
China in April 2017 against a team of six strong Chinese poker players. Further demonstrated at IJCAI 17
(Melbourne, August 2017) and NIPS 2017 (Long Beach, December 2017).
31Impact on the National Interest
Prof. Sandholm launched two startups
on Libratus’ algorithms:
Strategic Machine Inc. and Strategy
Robot.
In August 2018, Strategy Robot
received a 2-year contract for up to
$10M from the Pentagon’s Defense
Innovation Unit.
https://www.wired.com/story/poker-playing-robot-goes-to-pentagon/
32Materials Discovery for Energy Applications
Chris Wolverton, Northwestern University
AI-Driven HPC
Materials Discovery Through Data Driven Structural Search and
Heusler Nanostructures
Discovery of high-pressure compounds
– Materials discovery using density functional theory and the minima hopping
structure prediction method
– Discovery of FeBi2, the first iron-bismuth compound
– Discovery of two superconducting compounds in the
Cu-Bi system, CuBi and Cu11Bi7
Discovery of a new form of TiO2
– Employed machine learning to explore new TiO2 polymorphs
– Identified a new TiO2 hexagonal nano sheet (HNS)
– The HNS has a tunable band-gap and could be used for photocatalytic water
splitting and H2 production
33Severe Thunderstorm Prediction with Big Visual Data
James Z. Wang et al., Penn State
Applying machine learning to detect severe storm-
causing clouds
– Leveraging the vast historical archive of satellite imagery,
radar data, and weather report data from the NOAA to train
statistical models including deep neural networks on Bridges’
CPUs and GPUs
– Achieved high accuracy in detection of cloud patterns
– Developed fundamental statistical methods for data analysis
Detection of severe storm causing comma-shaped clouds
– Increasing the prediction lead time using deep models and from satellite images
GPUs
1. Zheng et al., Detecting Comma-shaped Clouds for Severe Weather Forecasting
using Shape and Motion, IEEE Transactions on Geosciences and Remote
Sensing, under 2nd-round review, 2018.
2. J. Ye, P. Wu, J. Z. Wang, J. Li, Fast Discrete Distribution Clustering Using
Wasserstein Barycenter With Sparse Support. IEEE Transactions on Signal Detection and categorization of bow echoes
Processing 65, 2317-2332 (2017) doi: 10.1109/TSP.2017.2659647. from weather radar data
34Fermilab Using Bridges to Prep for CMS @ HL-LHC
The High-Luminosity Large Hadron Collider (HL-LHC) will increase
luminosity by 10×, resulting in ~1EB of data.
The Compact Muon Solenoid (CMS) experiment will allow study of
the Standard Model, extra dimensions, and dark matter.
Fermilab is now using Bridges to integrate HPC into their workflow,
in preparation for HL-LHC coming online in 2026.
CMS Detector. From CERN, Event display of heavy-ion collision registered at the CMS Estimated CPU resources required for CMS into the HL-LHC era, using the
https://home.cern/science/experiments/cms detector on Nov. 8, 2018 (image: Thomas McCauley). current computing model with parameters projected out for the next 12
From https://cms.cern/news/2018-heavy-ion-collision- years. From A Roadmap for HEP Software and Computing R&D for the 2020s,
run-has-started. HPE Software Foundation.
Learn more: https://www.psc.edu/news-publications/2930-psc-supplies-computation-to-large-hadron-collider-group
35Unsupervised Deep Learning Reveals Prognostically Relevant Subtypes of Glioblastoma
Jonathan D. Young, Chunhui Cai, and Xinghua Lu, Univ. of Pittsburgh
Showed that a deep learning model can be trained to
represent biologically and clinically meaningful abstractions
of cancer gene expression data
Data: The Cancer Genome Atlas (1.2 PB)
Hypotheses: Hierarchical structures emerging from deep
learning on gene expression data relate to the cellular signal
system, and the first hidden layer represents signals related
to transcription factor activation. [1]
– Model selection indicates ~1,300 units in the first hidden
layer, consistent with ~1,400 human transcription factors.
– Consensus clustering on the third hidden layer led to
“One of these clusters contained all of the glioblastoma
discovery of clusters of glioblastoma multiforme with samples with G-CIMP, a known methylation
differential survival. phenotype driven by the IDH1 mutation and
associated with favorable prognosis, suggesting that
J. D. Young, C. Cai, X. Lu, Unsupervised deep learning reveals prognostically
· the hidden units in the 3rd hidden layer
representations captured a methylation signal
relevant subtypes of glioblastoma. BMC Bioinformatics 18, 381 (2017) without explicitly using methylation data as input.”
doi: 10.1186/s12859-017-1798-2. —Jonathan D. Young, Chunhui Cai, and Xinghua Lu
36Modeling of Imaging and Genetics using a Deep Graphical Model
Kayhan Batmanghelich, University of Pittsburgh
Causal Generative Domain Adaptation Networks
– A deep learning model trained with image data from
one hospital (“domain”) may fail to produce reliable
predictions in a different hospital where the data
distribution is different
– A generative domain adaptation network (G-DAN),
implemented using PyTorch, is able to understand
distribution changes and generate new domains
– Incorporating causal structure into the model – a
causal G-DAN (CG-DAN) can reduce its complexity
and accordingly improve the transfer efficiency
M. Gong, K. Zhang, B. Huang, C. Glymour, D. Tao, and K. Batmanghelich,
“Causal Generative Domain Adaptation Networks,” arXiv:1804.04333, 2018,
http://arxiv.org/abs/1804.04333.
37Multimodal Automatic Speech Recognition (ASR)
Florian Metze (CMU) et al.
2017 Jelinek Summer Workshop on Speech and Language Technology (JSALT)
38Deep Learning for Text-Based Prediction in Finance
Bryan Routledge and Vitaliy Merso, Carnegie Mellon University
Studying firm and investment fund financial disclosure
using Deep Learning Natural Language Processing
models
– Results presented at the Doctoral Consortium at the Text as
Data 2018 conference
– An early version linking the text of earnings announcements
to market reactions was been presented at the SEC Doctoral
Symposium 2018
“Given the large sizes of our corpora (hundreds of millions of words) and
the computational requirements of the modern Deep Learning models,
· our work would be impossible without the support from Bridges.”
—Brian Routledge, CMU
Many words used by investment funds in letters to their
shareholders are highly context-dependent. For example, the
word “subprime” can be either a very strong signal of a letter
describing a booming market or a very weak one, depending
on what other words appear around it.
39Exploring and Generating Data with Generative Adversarial Networks
Giulia Fanti, Zinan Lin, Carnegie Mellon University
Privacy-preserving dataset generation
– Fanti & Lin’s recent research aims to understand
fundamentally how Generative Adversarial Networks
(GANs) internally represent complex data structures and
to harness these observations to use GANs for privacy-
preserving dataset generation
– GANs are a new class of data-driven, neural network
based generative models that excel in high dimensions.
This work has led to two papers accepted to NIPS 2018: CelebA samples generated from DCGAN (upper) and
– “The power of two samples in generative adversarial PacDCGAN2 (lower) show PacDC-GAN2 generates
more diverse and sharper images.
networks” proposes “packing”, a principled approach to
improving the quality of generated images
– “Robustness of conditional GANs to noisy labels” earned a 1. Z. Lin, A. Khetan, G. Fanti, and S. Oh, “PacGAN: The
Spotlight Award at NIPS 2018, proposing a novel, power of two samples in generative adversarial
theoretically sound, and practical GAN architecture that networks,” arXiv:1712.04086, 2017.
2. K. Thekumparampil, A. Khetan, Z. Lin, and S. Oh,
consistently improves upon baseline approaches to “Robustness of conditional GANs to noisy labels,”
learning conditional generators where the labels are forthcoming in NIPS 2018, 2018 (Spotlight Award).
corrupted by random noise
40Towards a Deeper Understanding of Generative Image Models in Vision
Ying Nian Wu, UCLA
Learning interpretable latent representations:
a deformable generator model disentangles
appearance and geometric information into
two independent latent vectors
– The appearance generator produces the
appearance information, including color,
illumination, identity or category, of an image
– The geometric generator produces Each dimension of the appearance latent vector encodes appearance
displacement of the coordinates of each pixel
information such as color, illumination, and gender. In the fist line,
from left to right, the color of background varies from black to white,
and performs geometric warping, such as and the gender changes from a woman to a man. In the second line,
stretching and rotation, on the appearance the moustache of the man becomes thicker when the corresponding
dimension of Z approaches zero, and the hair of the woman becomes
generator to obtain the final synthesized denser when the corresponding dimension of Z increases. In the third
image. line, from left to right, the skin color changes from dark to white. In
the fourth line, from left to right, the illumination lighting changes
The model can learn both representations from the left-side of the face to the right-side of the face.
from image data in an unsupervised manner.
41Towards Real-time Video Object Detection Using Adaptive Scaling
Ting-Wu (Rudy) Chin, Ruizhou Ding, and Diana Marculescu, Carnegie Mellon University
Exploiting Resolution to Tune Accuracy and Speed
– The AdaScale project is about exploiting the resolution of the image
“as a knob” to improve the accuracy and speed of the deep neural
network-based object detection system.
Without AdaScale With AdaScale
The qualitative results of detection
accuracy achieved by AdaScale.
1. T.-W. Chin, R. Ding, and D. Marculescu, “AdaScale: Towards Real-Time
The performance of AdaScale on Video Object Detection Using Adaptive Scaling,” in SysML 2019, 2019
various baselines. [Online]. Available: https://www.sysml.cc/papers.html#
42Mapping Energy Infrastructure Using Deep Learning and Large Remote Sensing Datasets
Jordan Malof, Duke University
Satellite image Building mappings
Extracting high-quality information about energy
systems from overhead imagery with deep learning
– Precise locations of buildings (energy consumption)
– Small-scale solar arrays (energy generation)
– Improved speed and performance by expanding the receptive
field of neural networks only during label inference
B. Huang et al., “Large-scale semantic classification:
outcome of the first year of Inria aerial image labeling
benchmark,” in IEEE International Geoscience and Remote
Performance Computation time
(higher is better) (lower is better)
Sensing Symposium – IGARSS 2018, 2018.
https://hal.inria.fr/hal-01767807
Aerial photograph Solar mappings
Increasing receptive field size (in pixels)
43Understanding Public Space Use in Market Square
Javier Argota Sánchez-Vaquerizo, Carnegie Mellon University
The Project in Figures
– 4 cams
– 5 weeks of data collection (Aug 24 to Sep 28, 2018)
– 3200 hours of video processed
– 250 million detections
– 12 categories: pedestrians, trolleys, seats, tables,
sun umbrellas, tents, cars, pickups, vans, trucks,
bikes, motorcycles
Motivations
– Public safety
– Pedestrian flow and crowd management
– Vehicular traffic affection
– Venues and events impact assessment
Technology Capabilities
– Number of people, vehicles and objects detected Insights
– Segmentation – Weather (rain) affection on attendance
– Location, Trajectory, Speed – Uneven distribution of pedestrians in the space
– Prediction – Events and venues positive impact on attendance
– Anonymity from scratch – Short duration of visits
4445
Pedestrians
Trolleys
Seats
Tables
Sun umbrellas
Tents
Cars
Pickups
Vans
Trucks
Bikes
Motorcycles
46Fast and Accurate Object Detection in High-Resolution Video Using GPUs
Vic Ruzicka and Franz Franchetti, Carnegie Mellon University
Object detection in computer vision traditionally works
with relatively low-resolution images. However, the
resolution of recording devices is increasing, requiring
new methods for processing high-resolution data.
Ruzicka & Franchetti’s attention pipeline method
uses two-staged evaluation of each image or video
frame under rough and refined resolution to limit Example of a crowded 4K video frame annotated
the total number of necessary evaluations. with Ruzicka & Franchetti’s method.
Both stages use the fast object detection model YOLO v2.
Their distributed-GPU code maintains high accuracy while reaching performance of
3-6 fps on 4k video and 2 fps on 8k video. This outperforms the individual base-line
approaches, while allowing the user to set the trade-off between accuracy and performance.
Best Paper Finalist at IEEE High Performance Extreme Computing Conference (HPEC) 2018
47Fast and Accurate Object Detection in High-Resolution Video Using GPUs
Vic Růžička and Franz Franchetti, Carnegie Mellon University
48Distributed Learning for Large-Scale Multi-Robot Path Planning in Complex Environments
Guillaume Sartoretti, Carnegie Mellon University
Multi-agent path finding (MAPF)
– An essential component of many large-scale, real-world robot
deployments, from aerial swarms to warehouse automation.
– Most state-of-the-art MAPF algorithms still rely on centralized
planning, scaling poorly past a few hundred agents.
– Such planning approaches are maladapted to real-world
deployments, where noise and uncertainty often require paths
be recomputed online, which is impossible when planning
times are in seconds to minutes. Example problem where 100 simulated robots (white dots) must
Pathfinding via Reinforcement + Imitation Learning compute individual, collision-free paths in a large factory-like
environment. Reproduced from [1].
– Using Bridges-GPU, Sartoretti trained and tested PRIMAL, a novel
1. G. Sartoretti et al., “PRIMAL: Pathfinding via
framework for MAPF that combines reinforcement and imitation Reinforcement and Imitation Multi-Agent Learning,”
learning to teach fully-decentralized policies, where agents 2018. http://arxiv.org/abs/1809.03531.
reactively plan paths online in a partially-observable world while
exhibiting implicit coordination.
– In low obstacle-density environments, PRIMAL outperforms state-of-the-art MAPF planners in certain cases,
even though these have access to the whole state of the system. They also deployed PRIMAL on physical and
simulated robots in a factory mockup scenario, showing how robots can benefit from their online, local-
information-based, decentralized MAPF approach.
49Automation in data discovery
Automation in data curation and generation
Measuring and improving data quality
Integrating datasets and enabling interoperability
Biomedical data discovery and reuse
•… Data privacy, security and algorithmic bias
The future of scientific data and how we work together
INVITED SPEAKERS
KEYNOTES
Tom M. Mitchell Glen de Vries Robert F. Murphy Natasha Noy
Interim Dean and President and Ray and Stephanie Staff Scientist
E. Fredkin University Co-founder Lane Professor Google AI
Professor Medidata Solutions Head of
School of Computational Biology
Computer Science School of
Carnegie Mellon Computer Science
University Carnegie Mellon
University
Deadline for Abstracts: February 22
https://events.library.cmu.edu/aidr2019/
502018 HPCwire Awards 51
Outline
Motivation & Vision
Realizing the Vision: Bridges and Bridges-AI
Exemplars of Success
Summary
52Summary
PSC’s approach to scalable, converged HPC+AI is enabling breakthroughs
across an extremely broad range of research areas.
These resources – Bridges, including Bridges-AI, are available at no charge
for research and education
– Bridges-AI builds on Bridges’ strength in converged HPC, AI, and Big Data to
provide a unique platform for AI and AI-enabled simulation.
To request a free research/education allocation, visit:
https://psc.edu/about-bridges/apply
53Thank you!
Questions?
54You can also read
