Dr. Shari Feth

Director, Innovation, Science and Technology, Missile Defense Agency

Supernova Award Category: 
Data to Decisions
The Organization: 

The Missile Defense Agency (MDA) is a research, development, and acquisition agency within the US Department of Defense. The Agency develops, tests, and fields an integrated Missile Defense System, working closely with the Combatant Commands (e.g. Pacific Command, Northern Command, etc.) who will rely on the system to protect the United States, its forward deployed forces, and its friends and allies from hostile missile attack. MDA works with the Combatant Commanders to ensure the development of a robust missile defense system technology and development program to address the challenges of an evolving threat. MDA is committed to maximizing the mission assurance and cost effectiveness of our management and operations through continuous process improvement.

The Problem: 

Ballistic Missile Trends: Countries invest in ballistic missiles and other similar technologies because they are a means to project power in regional and strategic contexts, and a capability to launch an attack from a distance. According to the Intelligence Community, ballistic missile proliferation continues to grow as countries acquire a greater number of ballistic missiles, increasing their capabilities, incorporating BMD countermeasures and making them more complex, survivable, reliable and accurate. Maneuvering threats continue to be developed and fielded in threat countries. The proliferation of ballistic missiles is increasing the number of anti-access weapons available to potential regional adversaries. These weapons could be used to reduce military options for Combatant Commanders and decrease the survivability of regional military assets. Technological advances are now making hypersonic glide vehicles and missiles flying non-ballistic trajectories practicable. Through its capabilities for defending critical nodes, military assets, and seats of government, missile defense enhances existing non-proliferation activities. Missile defenses can provide a permanent presence in a region and discourage adversaries from believing they can use ballistic missiles to coerce or intimidate the U.S. or its allies. 

The Solution: 

MDA must concurrently address evolving threats while also trying to decrease development costs. Existing agency systems are challenged by sparse data, and the time from intelligence collection to model development is often prohibitively long due to the current physics-based modeling approach. The potential to rapidly generate potential missile trajectories that are representative of the performance of physics-based models, while also enabling enterprise artificial intelligence (AI) across the agency, has transformative cost and speed benefits. Using a generative modeling application, agency personnel can now rapidly generate large datasets in data starved environments for a broad spectrum of missile defense technology initiatives. By leveraging a common enterprise AI toolset across development teams, MDA has accelerated the development, deployment, system integration, and life cycle management of missile defense systems for which machine learning (ML) is an integral part.

The Results: 

Existing physics-based agency solutions to generate trajectories that are representative of potential missile threats were computationally expensive, time consuming, and resource intensive. Producing large volumes of physics-based threat data packages representative of different enemy missile profiles required substantial effort due to the wide array of physically viable missile paths. The agency wanted an ML-based approach to develop generative models representative of high-fidelity physics-based model performance while capable of more rapidly generating data. These ML models could be used by MDA partners to generate thousands of potential threat data packages that are necessary to enabling a range of defensive strategies. 

By scaling out the data to meet the demand, MDA can provide third party contractors a dynamic model capable of generating enough trajectories to satisfy the entire test envelope. Doing so improves comparison insights for MDA and accelerates systems development.

When connected with intel data, generative models can update directly from new data in minutes via a method known as online learning. Comparatively, the baseline process for updating a physics model to incorporate new trajectory data is a months-long, multi-step process involving feature inference, control algorithm redesign, and physics model assembly. Generative models decrease the time to go from intelligence collection to model deployment by as much as 100x over current agency systems.


Across several engagements, MDA & C3 AI worked to: Deploy C3 AI Platform with the appropriate GPU configuration; Ingest 500 missile trajectories and 2.9+ million point measurements into unified data image along with IR reading across each trajectory; Extend unified data image to incorporate image datasets and binary telemetry files requiring custom de-code routines; Complete 400 deep learning experiments, iterating with agency experts through 3 modeling approaches to develop a successful end-to-end generative model pipeline; Leverage the generative model pipeline to create, in minutes, tens of thousands of physically realistic AI generated trajectories as evaluated across 5 categories of physics-based constraints; Leverage the generative model pipeline to create IR readings across any trajectory within minutes. Model backtesting demonstrated that all client communicated KPIs were achieved (Greater than 90% of all predictions fell within 20% of high fidelity generated results).; Collaboratively design a user interface to generate, analyze, and export new threat data packages through comprehensive trajectory and IR analysis pages and immersive 3D visualization tool. By leveraging C3 AI’s end-to-end machine learning capabilities from data ingestion through model training and output, generative models decrease the time to go from intelligence collection to model deployment by as much as 100x over current agency systems.

The Technology: 

Using the C3 AI Platform and the C3 AI Generative Modeling application, MDA personnel can now rapidly generate large datasets in data starved environments for a broad spectrum of missile defense technology initiatives. By leveraging the C3 AI Platform as a common enterprise AI toolset across development teams, the agency can greatly accelerate the development, deployment, system integration, and life cycle management of missile defense systems for which machine learning is an integral part.

Disruptive Factor: 

Global threats are constantly changing and securing the homeland is heavily dependent upon the ability to test defensive capabilities against novel threat scenarios. The Defense Innovation Unit (DIU), and DoD organization that accelerates commercial technology into the Department, was a catalyst for the use of this technology after a competitive prototype, and the success lead to an expansion to a $500M procurement vehicle and additional use cases.

These tests require generating data packages describing how threats traverse through space and their IR readings across the trajectory. To date the generation of this data has been done solely through physics-based methods. Complimenting high-fidelity physics-based data generation with mid-fidelity rapid AI-generated data will increase the responsiveness of the MDA to a quickly evolving threat-space.

Across the MDA, the generation of threat data packages has solely been completed using high-fidelity physics-based models. These models provide very rich detail describing threats, but routinely take weeks or months to produce. MDA collaborated with DIU and C3.ai to shift this paradigm and realize value in the rapid generation of mid-fidelity threat data packages (trained on high fidelity data) using AI methods. This allows consumers of threat data to quickly understand where high fidelity data may truly be needed so that the future use of time-consuming methods can be better focused on areas where it adds immediate value.

Shining Moment: 

In addition to modeling incoming trajectories, MDA used the C3 AI Platform to apply deep learning models to predict infrared emissions across 4 critical bands. As the bands approach the wavelength of visible light, achieving the MDA’s accuracy thresholds became increasingly difficult. MDA and C3 AI experts were able to refine the model architecture and features to distinguish the threat emissions relative to the sun to exceed all thresholds, increasing the overall mission value.

About Missile Defense Agency

The Missile Defense Agency's (MDA) mission is to develop and deploy a layered Missile Defense System to defend the United States, its deployed forces, allies, and friends from missile attacks in all phases of flight. The MDA designs, develops, and delivers capabilities while also sustaining advanced missile defense systems as part of the fully integrated and layered missile defense system.