PAISE 2019


1st Workshop on Parallel AI and Systems for the Edge

PAISE 2019 will be held in conjunction with IPDPS 2019
Friday, 24 May 2019
Rio de Janeiro, Brazil


Workshop Program

Friday, May 24th, 2019 at Buzios B room on the 2nd Floor

Session 1: 08:45 AM - 09:50 AM (70 min)

09:50 AM - 10:10 AM Coffee Break (20 min)

Session 2: 10:10 AM - 11:50 AM (105 min)

11:50 AM - 01:15 PM Lunch (85 min)

Session 3: 01:15 PM - 03:25 PM (130 min)

03:25 PM - 03:45 PM Coffee Break (20 min)

Session 4: 03:45 PM - 05:20 PM (95 min)

Invited Talks

• Keynote

Speaker
İlkay Altıntaş, San Diego Supercomputer Center (SDSC), UC San Diego
Title
Modeling Wildfire Behavior at the Continuum of Computing
Abstract
Modeling of the extent and dynamics of wildfires, and their socio-economic and human impacts, is a data-driven discipline with a vibrant scientific community of modeling, remote sensing, technology and social science experts, driven by the urgent societal need to mitigate the rising frequency and severity of wildfires. However, there are still challenges in integration of the scientific discoveries and products with operational and planning purposes, requiring a deeper knowledge network than what exists today. Previously NSF-funded WIFIRE project took the first steps to tackle this problem with a goal to create integrated systems for wildfire monitoring, simulation, and response. Today, the WIFIRE system provides an end-to-end management infrastructure from the data collection to modeling efforts using a continuum of computing methods that integrated edge, cloud and high-performance computing. Though this cyberinfrastructure, the WIFIRE project provides data driven knowledge for a wide range of public and private sector users enabling scientific, municipal and educational use. This talk will review some of our recent work on building this dynamic data driven cyberinfrastructure and impactful application solution architectures that showcase integration of a variety of existing technologies and collaborative expertise. The lessons learned from the development of the NSF WIFIRE cyberinfrastructure will be summarized. Open data issues, use of edge and cloud computing on top of high-speed network, reproducibility through containerization and automated workflow provenance will also be discussed in the context of WIFIRE.
Bio
Dr. İlkay Altıntaş is the Chief Data Science Officer at the San Diego Supercomputer Center (SDSC), UC San Diego, where she is also the Founder and Director for the Workflows for Data Science Center of Excellence and a Fellow of the Halicioglu Data Science Institute (HDSI). In her various roles and projects, she leads collaborative multi-disciplinary teams with a research objective to deliver impactful results through making computational data science work more reusable, programmable, scalable and reproducible. Since joining SDSC in 2001, she has been a principal investigator and a technical leader in a wide range of cross-disciplinary projects. Her work has been applied to many scientific and societal domains including bioinformatics, geoinformatics, high-energy physics, multi-scale biomedical science, smart cities, and smart manufacturing. She is a co-initiator of the popular open-source Kepler Scientific Workflow System, and the co-author of publications related to computational data science at the intersection of workflows, provenance, distributed computing, big data, reproducibility, and software modeling in many different application areas. She is also a popular MOOC instructor in the field of “big” data science, and reached out to hundreds of thousands of learners across any populated continent. Her Ph.D. degree is from the University of Amsterdam in the Netherlands with an emphasis on provenance of workflow-driven collaborative science. She is an associate research scientist at UC San Diego. Among the awards she has received are the 2015 IEEE TCSC Award for Excellence in Scalable Computing for Early Career Researchers and the 2017 ACM SIGHPC Emerging Woman Leader in Technical Computing Award.

• Invited Talk 1

Speaker
Chengxiang Yin, Department of Electrical Engineering and Computer Science, Syracuse University
Title
A Deep Recurrent Neural Network Based Predictive Control Framework for Reliable Distributed Stream Data Processing
Abstract
It is very hard to model, predict and control modern computing systems since they have become very complicated and highly dynamic. We aim to develop a novel predictive control framework based on Deep Recurrent Neural Network, which can learn a reliable way to control a Distributed Stream Data Processing System (DSDPS) from its collected data, rather than any mathematical model (such as queueing models), just as a human learns a new skill (such as cooking, swimming and driving). We, for the first time, propose to leverage emerging Deep Recurrent Neural Network (DRNN) for enabling predictive control in DSDPSs. In this talk, I will first introduce the basic idea of DRNN for prediction. Then I will describe a DRNN-based predictive control framework proposed particularly for DSDPS. I will also show some experimental results to justify its effectiveness and superiority. In addition, I will point out some future research directions. At last, I will point out some future research directions.
Bio
Chengxiang Yin is currently pursuing the Ph.D. degree at the Department of Electrical Engineering and Computer Science, Syracuse University, Syracuse, NY, USA. He received his B.S. degree from the School of Information and Electronics at Beijing Institute of Technology, Beijing, China, in 2016. His research interests include Deep Learning and Computer Vision.

• Invited Talk 2

Speaker
Tal Ben-Nun, ETH Zurich.
Title
Demystifying Parallel Deep Neural Network Inference.
Abstract
Deep Neural Networks (DNNs) are becoming an important tool in modern computing applications. Accelerating inference is a major challenge and techniques range from high-level compilers to low-level circuit design. The talk outlines deep learning from a computational perspective, followed by approaches for inference parallelization. We present trends in DNN architectures and the resulting implications on computation strategies. We then review the state of the art in DNN compilers, neural architecture tuning, specialized hardware architectures, and combinations thereof. Based on those approaches, we extrapolate potential directions for parallelism in decentralized deep learning on edge devices.
Bio
Dr. Tal Ben-Nun is a Computer Science Postdoc and member of the Scalable Parallel Computing Laboratory at ETH Zurich. Prior to this, he was part of the Distributed Computing (Computer Science), the X-ray scattering (Chemistry) and Parallel Systems Laboratories.

• Invited Talk 3

Speaker
Brian C. Van Essen, Lawrence Livermore National Laboratory (LLNL).
Title
Scalable Deep Learning and Computing at the Edge of Cognitive Simulation
Abstract
Our work on scalable deep learning has created new opportunities for training deep neural networks on scientific data sets with data samples too large to even fit on an accelerator. We will explore how this capability enables us to think about computing at the edge of large scale scientific instrument. Additionally, recent work on cognitive simulation is exploring methods for coupling machine learning with traditional scientific simulations. One challenge is tightly coupling machine learning inference with running scientific simulations, inside of the simulation's computing loop. We will present the idea that tightly-coupled machine learning inference and associated in-system training represents a new edge for parallel AI and computing.
Bio
Brian Van Essen is the informatics group leader and a computer scientist at the Center for Applied Scientific Computing at Lawrence Livermore National Laboratory (LLNL). He is pursuing research in large-scale deep learning for scientific domains and training deep neural networks using high-performance computing systems. He is project leader for the Livermore Big Artificial Neural Network open-source deep learning toolkit, and he co-leads an effort to map scientific, data-intensive machine learning applications to neuromorphic architectures. He joined LLNL in 2010 after earning his Ph.D. and M.S. in computer science and engineering at the University of Washington. He also has an M.S and B.S. in electrical and computer engineering from Carnegie Mellon University.

• Invited Talk 4

Speaker
Rajesh Sankaran, Argonne National Laboratory.
Title
Waggle: A Platform for AI @ Edge.
Abstract
The Waggle Platform is a research project at Argonne National Laboratory to design, develop, and deploy a novel wireless sensor platform with advanced edge computing capabilities to enable a new breed of sensor-driven environmental science and smart city research. The software and hardware designs from the Waggle project are used by the NSF funded Array Of Things, which is building a smart city and open data with urban sensors in Chicago. The innovative architecture leverages emerging technology in low-power processors, sensors, and cloud computing to build powerful and reliable sensor nodes that can actively analyze and respond to data. Cloud computing provides elastic resources for storing and computing on data. Waggle is designed from the ground up with security, privacy, extensability, and survivability as key design points. The Waggle reference platforms and software are launching points. All of the software is Open Source, and the software is modular, so researchers can add their own sensors, computing pipelines, and data analysis. The talk will discuss the various technical aspects of the Waggle platform and the Array of Things urban deployment.
Bio
Raj Sankaran is a member of the technical staff at Argonne National Laboratory. He has a PhD in Electrical and Computer Engineering and is interested in research and development in topics related to Edge Computing, Attentive Sensing, and Embedded Computing Systems. Through his research pursuits at Argonne, Raj collaborates closely with Environmental, Urban, High-Performance Computing and Weather/Climate researchers. Raj co-leads the Waggle Edge-Computing research platform and the tech-team of the Array of Things initiative.