S8941 SYNTHETIC LABEL DATA FOR TRAINING DEEP LEARNING ISR - PowerPoint PPT Presentation

Place image here (13.33” x 3.5”) S8941 – SYNTHETIC LABEL DATA FOR TRAINING DEEP LEARNING ISR ALGORITHMS WILL RORRER, PROGRAM MANAGER NVIDIA GTC San Jose 29 March 2018 NON-EXPORT CONTROLLED THESE ITEM(S) / DATA HAVE BEEN REVIEWED IN ACCORDANCE WITH THE INTERNATIONAL TRAFFIC IN ARMS REGULATIONS (ITAR), 22 CFR PART 120.11, AND THE EXPORT ADMINISTRATION REGULATIONS (EAR), 15 CFR 734(3)(b)(3), AND MAY BE RELEASED WITHOUT EXPORT RESTRICTIONS. HARRIS.COM | #HARRISCORP

Agenda Harris Corporation Introduction A Call to Action: The Urgency Behind the DoD’s Adoption of AI Review: Applications of Deep Learning at Harris Review: Harris’ Work to Scale Deep Learning Harris’ Approach for Handling the Label Data Burden Q&A S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 2

Harris Corporation introduction – segment overviews Communication Systems Tactical and airborne radios, night vision technology, and defense and public safety networks Electronic Systems Electronic warfare, avionics, robotics, advanced communications and maritime systems for the defense industry, as well as air traffic management solutions for the civil aviation industry Space and Intelligence Systems Complete solutions encompassing advanced sensors and payloads, processing systems, and analytics for global situational awareness, space superiority missions, and Earth insights S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 3

A call to action: the urgency behind the DoD’s adoption of AI “We’re going to find ourselves in the not too “The GEOINT discipline has grown beyond the distant future swimming in sensors and drowning limits of human interpretation and in data” explanation . The explosion of available data Lt. Gen. David A Deptula, diminishes the comparative advantage collection 2010 USAF Dep Chief of Staff for ISR provides. Instead, automated processing, "The skies will ‘darken’ with the hundreds of small advancing tradecraft, human-machine satellites to be launched by U.S. companies and collaboration, and the ability to anticipate as procedures are developed to allow safe behaviors will provide us a new advantage.” operation of unmanned aerial vehicles in civil Robert Cardillo, airspace," Director of NGA Robert Cardillo, 2015 Director – NGA “So just how big is this rising tide? If we were to attempt to manually exploit the commercial satellite imagery we expect to have over the next 20 years, we would need eight million imagery analysts. Even now, every day in just one combat theater with a single sensor, we collect the data equivalent of three NFL seasons – every game. In high definition! Imagine a coach trying to understand the strategy of his opponents by watching every play made by every team in every game for three seasons – all in one single day. Because three more seasons will be coming tomorrow. That’s what we ask our analysts to do – when we don’t augment them with automation. But with all this data – and dramatic improvements in computing power – we have a phenomenal opportunity to do and achieve even more.” Robert Cardillo, 2017 Director – NGA S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 4

A call to action: the urgency behind the DoD’s adoption of AI 2017 ImageNet Challenge Object Classification Winner: WMW, Momenta.ai 2.25% Error Rate Graph from https://www.dsiac.org 14M training images 1,000 object categories S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 5

A call to action: the urgency behind the DoD’s adoption of AI S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 6

Agenda Harris Corporation Introduction A Call to Action: The Urgency Behind the DoD’s Adoption of AI Review: Applications of Deep Learning at Harris Review: Harris’ Work to Scale Deep Learning Harris’ Approach for Handling the Label Data Burden Q&A S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 7

Harris deep learning research and development Harris Corp. has been internally investing in taking state-of-the-art deep learning technologies and applying them to remote sensing and geospatial intelligence customer problems Focus Areas: Reducing the cost of training How long? • Reduce dollar cost, human cost, and • Harris has been working on deep computer cost of building new models learning for over five years Extensibility How much? • Ability to quickly redeploy and repackage • Multimillion dollar internal research and tools to support new problem sets development investment in the last three years Support multiple types of data • Additional commercialization investment • New sensors and data fusion Investment approach Automation • Research • Ability for processes to interface to tools, • Pilot Projects removing human from the loop • Software tool development • Commercialization S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 8

Harris’ deep learning R&D and applications S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 9

Harris’ deep learning R&D and applications S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 10

Harris deep learning R&D motion imagery applications – high revisit rate still imagery Problem: Clandestine airfields in South American countries used for illegal narcotic trafficking Goal: Detect new airfields and determine activity levels with high temporal resolution data (Planet) S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 11

Harris deep learning R&D motion imagery applications – high revisit rate still imagery Sources: • Planet imagery • DigitalGlobe EGD imagery • Ecuadorean geoportal shapefile of known remote landing strips • Google Earth imagery • Wikimapia • OpenStreetMap S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 12

Applications: Lessons Learned and Takeaways High accuracy achievable with Perform ATR with multiple data types AI enabled by powerful GPUs appropriate NN architecture (minutes instead of days) and labeled data MSI PAN SAR LIDAR Dramatic performance improvement p(Wet) AUC = 98.1% S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 13

Agenda Harris Corporation Introduction A Call to Action: The Urgency Behind the DoD’s Adoption of AI Review: Applications of Deep Learning at Harris Review: Harris’ Work to Scale Deep Learning Harris’ Approach for Handling the Label Data Burden Q&A S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 14

The art of scaling machine learning Problem Machine Learning Answers! Mobile Access Present solution Understand Data ingest PhD problem Hardware Training Data Performance Retrain model monitoring Lots of hardware Classification Curation Export Controls Sensor Labels knowledge Scheduling Notifications Orthorectification Prioritization Geographic Refinement Metadata S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 15

Breaking it down Accessibility Infrastructure Automation Learning All Source / Multi-INT Source Information S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 16

Rinse and repeat [ Automated Activity Based Intelligence ] Higher Order Sense Making Cognitive Ecosystem S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 17

Harris’ work to scale deep learning for defense source information S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 18

Harris’ work to scale deep learning for defense all source/multi-INT S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 19

Harris’ work to scale deep learning for defense learning S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 20

Harris’ work to scale deep learning for defense infrastructure S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 21

Harris’ work to scale deep learning for defense accessibility and automation S8941 – Synthetic Label Data for Training Deep Learning ISR Algorithms NON-Export Controlled Information | 22