Accelerated Global Human Settlement Discovery

Speaker:  Jibonananda Sanyal – Oak Ridge, TN, United States
Topic(s):  Information Systems, Search, Information Retrieval, Database Systems, Data Mining, Data Science

Abstract

Understanding where people live is fundamental to understanding what people do and what their social needs are with respect to energy security; policy and urban development; resiliency; disaster and emergency response; intelligence and security; humanitarian support, as well as understanding the behavioral social dynamics. The computational research needs are driven by ongoing efforts where initial investigations have been conducted to explore the application of machine learning --- including deep learning --- toward global scale human settlement detection from high resolution satellite imagery. The approaches undertaken run well on single GPUs and small clusters, and are well positioned to scale on a resource such as the Titan supercomputer. The high resolution determination of settlements necessitates being able to dynamically produce very large datasets with fine temporal resolutions to adequately capture changes and fluxes. This problem is multi-petascale computationally and multi-petabytes in imagery data size. This research has produced a portable, scalable, and configurable machine-learning and image-processing pipeline that can run at continental to global scales at high resolutions.

The benefits of a portable, scalable, and configurable machine-learning and image-processing pipeline are several. The framework allows the rapid processing of massive amounts of imagery in a short period of time. The speed allows responsive support for FEMA’s disaster response in the western United States, to meet specific needs of the intelligence community, and strategic resourcing for polio and malaria eradication in sub-Saharan Africa by the Gates Foundation. 

This talk focuses on the key image products and computational components were developed in this work:
 
A scalable shallow and deep machine learning training and execution framework for high-resolution geospatial imagery (0.3-1 meter) 
Very high-resolution human settlement layer for the United States and sub-Saharan Africa, based on current scope of funded projects, and on-demand settlement detection capabilities for the rest of the world, and, 
A hybrid workflow for training on project-procured multiple nVIDIA’s DGX-1 deep learning machines and subsequent at-scale execution of the derived models on Titan. 

About this Lecture

Number of Slides:  40
Duration:  50 - 60 minutes
Languages Available:  English
Last Updated: 

Request this Lecture

To request this particular lecture, please complete this online form.

Request a Tour

To request a tour with this speaker, please complete this online form.

All requests will be sent to ACM headquarters for review.