Training_Data

Founded in 2015 as part of IQT Labs, CosmiQ Works launched to focus on the geospatial analytics market and provide technical insights, targeted research, reports, and more. Over the past six years CosmiQ has produced many projects and insights that have helped the intelligence and academic communities better understand how geospatial can help tackle hard problems. Our final Training_Data podcast brings together CosmiQ's current colleagues and alum in a special episode celebrating the team's key accomplishments, milestones, and lessons learned through the years.

Satellite imagery analytics have numerous human development and disaster response applications, particularly when time series methods are involved. The Multi-Temporal Urban Development SpaceNet 7 Challenge focuses on developing novel computer vision methods for non-video time series data, asking participants to identify and track buildings in satellite imagery time series collected over rapidly urbanizing areas. In this episode, CosmiQ’s Ryan Lewis, Adam Van Etten, and Daniel Hogan are joined by Planet’s Jesus Martinez Manzo and AWS Disaster Response’s Grace Kitzmiller to explore this new challenge.

Learn more at www.spacenet.ai, and at the DownLinQ (https://medium.com/the-downlinq)

CosmiQ’s Jake Shermeyer and Daniel Hogan are joined by Capella Space’s Jason Brown and IEEE Geoscience and Remote Sensing’s (GRSS) Ronny Hänsch to once again discuss the SpaceNet 6 Dataset and post-challenge experiments. Learn more about data fusion and deep learning approaches that work to blend synthetic aperture radar (SAR) and optical imagery. Additionally, the podcast also explores the value of frequent SAR revisits that can be beneficial for foundational mapping applications. Finally, the group discusses CosmiQ’s new addition to the Solaris python package: a multi-modal pre-processing library and a new API called ‘PipeSegment’ for seamlessly stringing together different operations.

SpaceNet is a nonprofit made possible by co-founder and managing partner, CosmiQ Works; co-founder and co-chair, Maxar Technologies; and all the other Partners: Amazon Web Services (AWS), Capella Space, Topcoder, IEEE Geoscience and Remote Sensing (GRSS), the National Geospatial-Intelligence Agency, and Planet.  

SpaceNet is a non-profit dedicated to accelerating open source, applied research in geospatial machine learning. In this episode, CosmiQ’s Ryan Lewis, Jake Shermeyer, and Daniel Hogan discuss the SpaceNet 6 Challenge where participants were asked to automatically extract building footprints with computer vision and AI algorithms using a combination of synthetic aperture radar (SAR) and electro-optical imagery. Hear about the challenge’s winning artificial intelligence models and the tradeoff between inference speed and model performance.

SpaceNet is made possible by co-founder and managing partner, CosmiQ Works; co-founder and co-chair, Maxar Technologies; and all the other Partners: Amazon Web Services (AWS), Capella Space, Topcoder, IEEE Geoscience and Remote Sensing (GRSS), the National Geospatial-Intelligence Agency, and Planet.

Learn more at www.spacenet.ai, and at the DownLinQ (https://medium.com/the-downlinq)

Despite its application to myriad humanitarian and civil use cases, automated road network extraction from overhead satellite imagery remains quite challenging. However, the SpaceNet 5 challenge made significant progress in this field with top participants being able to extract both road networks and speed/travel time estimates for each roadway. On today’s pod, CosmiQ’s Ryan Lewis and Dr. Adam Van Etten explore the challenge’s unique dataset and geographic diversity over time, the winning models, and the tradeoff between inference speed and model performance.

SpaceNet is a non-profit LLC co-founded and managed by In-Q-Tel's CosmiQ Works in collaboration Maxar Technologies, a co-founder, and the other SpaceNet Partners including AWS, Intel AI, Topcoder, Capella Space, IEEE GRSS, The National Geospatial-Intelligence Agency (NGA), and Planet.

How can lessons from geospatial computer vision applications impact bio image analysis? CosmiQ’s Dr. Nick Weir and B.Next’s Dr. Dylan George explore the intersection of these two fields and why artificial intelligence (AI) has struggled to gain traction with both satellite imagery and medicine. Hear about their project that researched similarities and differences between satellite imagery, microscopy, and “normal” photographs, and why researchers developing AI methods for microscopy might want to leverage the work done for geospatial applications.

Read more about their project at The DownLinQ and BioQuest:

· Viewing the World Through a Straw, Part 1

· Viewing the World Through a Straw, Part 2

How does emerging technology like artificial intelligence (AI) fit into the broader geopolitical landscape, particularly one that continues to rapidly evolve? CSIS’s Melissa Dalton and Lindsey Sheppard join the podcast to discuss their recent research project, the Gray Zone Project (https://www.csis.org/grayzone), on current challenges faced by the U.S. and its allies, and how emerging technologies could be leveraged to help resolve or mitigate some of the analytic tasks facing national security organizations.

CSIS is an independent, bipartisan think tank in Washington, D.C. focused on producing objective analytic analysis that is targeted primarily toward U.S policy makers, but also an increasingly broader audience to include the tech community, allies and partners, and private sector. CSIS’s International Security Program analyzes how the U.S. can best deter, campaign in, and respond to “gray zone” approaches.

Learn more about CosmiQ at www.cosmiqworks.org, and CSIS at www.csis.org.

How much labeled data does a machine learning model need in order to achieve sufficient performance? Most analyses simply use all available data and focus on model architecture, with scant attention given to whether the dataset size is appropriate for the task and architecture’s complexity. But is that the right answer?

In this episode, IQT CosmiQ Works’ Ryan Lewis, Daniel Hogan, and Adam Van Etten dive into the importance of deep quantitative analysis and the impact of training data size on model performances. Learn how they used CosmiQ’s Machine Learning Robustness Study to help address dependence on geography and dataset size at the leading edge of geospatial machine learning.

Learn more about CosmiQ at www.cosmiqworks.org.

Spanning three years of featuring five unique datasets and challenges, SpaceNet® continues to apply various aspects of machine learning to solve difficult foundational mapping problems. Looking ahead, Payam Banazadeh, Founder & CEO of Capella Space, joins IQT CosmiQ’s Ryan Lewis and Jake Shermeyer to discuss the SpaceNet 6 Challenge which explores an under-explored modality of data: Synthetic Aperture Radar (SAR). The trio provide an overview on the importance of SAR data and its value in the upcoming challenge.

SpaceNet is a collaboration between CosmiQ Works, Maxar Technologies, Amazon Web Services (AWS), Intel AI, Capella Space, Topcoder, and IEEE GRSS.

Space Club Rule 42: When In Doubt, Use SAR.

Thank you to everyone who made the inaugural season of Training_Data possible! We will be back with new content for Season 2 in January 2020. Have a great holiday season and happy new year.