Put AI to Work
April 15-18, 2019
New York, NY
Please log in

ImageNet for satellite imagery: Opportunities and risks

Ryan Mukherjee (JHU/APL), Neil Fendley (JHU/APL)
2:40pm–3:20pm Thursday, April 18, 2019
Implementing AI
 Location: Trianon Ballroom
Secondary topics:  Computer Vision, Data and Data Networks, Ethics, Privacy, and Security, Reliability and Safety
Average rating: *****
(5.00, 1 rating)

Who is this presentation for?

  • Anyone interested in satellite image analysis and deep and adversarial machine learning

Level

Beginner

Prerequisite knowledge

  • A basic understanding of deep learning (useful but not required)

What you'll learn

  • Gain an overview of the functional Map of the World dataset and challenge, the cutting-edge satellite image processing models that were and can be created, and the security implications of these models

Description

While deep learning has led to many advancements in computer vision, most models can only be used with ground-based imagery and do not apply to overhead or satellite imagery. Ryan Mukherjee and Neil Fendley offer an overview of functional Map of the World (fMoW), an ImageNet for satellite imagery built to address this issue, as part of an Intelligence Advanced Research Projects Activity (IARPA) challenge focused on fostering machine learning research for satellite image analysis.

Join in to explore this novel dataset; deep learning models to classify facilities, buildings, and land use; and results from a public prize challenge that sought to develop the best performing solutions. And because accuracy and reliability are critical for government, commercial, and humanitarian applications of this technology, Ryan and Neil also present their work investigating the unique aspects of how deep remote sensing models may be fooled and defended.

Photo of Ryan Mukherjee

Ryan Mukherjee

JHU/APL

Ryan Mukherjee is a senior research engineer at JHU/APL. Ryan has been involved in machine learning, computer vision, and remote sensing projects for over nine years within JHU/APL’s Research and Exploratory Development Department. Most recently, Ryan led JHU/APL’s support of IARPA’s functional Map of the World effort and is focused on providing easy and free access to associated tools and data.

Photo of Neil Fendley

Neil Fendley

JHU/APL

Neil Fendley is a computer vision researcher at JHU/APL, where he works on machine perception and reasoning, focusing on deep learning. Over the last two years, Neil has worked on automated retinal disease diagnosis and satellite imagery classification through JHU/APL’s support of IARPA’s functional Map of the World effort. Recently, Neil has focused on adversarial machine learning and hosted a JHU/APL internal challenge to design and defend against adversarial perturbations.