RSS feed LinkedIn Watch us on Youtube

Activity title

Machine Learning Ecosystem for the Rapid Research, Development, and Deployment of Artificial Intelligence and Machine Learning Capabilities

Activity Reference

IST-ET-112 (AI2S)

Panel

Information Systems Technology

Security Classification

NATO UNCLASSIFIED

Status

Planning

Activity type

ET

Start date

2020

End date

2020

Keywords

Artificial Intelligence, Computing Infrastructure, Machine Learning

Background

Artificial Intelligence (AI) and Machine learning (ML) are foundational to realizing future intelligent systems. AI/ML has the potential to transform how the Alliance collects, processes, shares, and analyzes diverse data types as well as how it learns, adapts to, and reasons about the environment. In just the past few years, the global pace of AI/ML advancements and application domains has undergone exponential growth. AI/ML has mastered tasks previously considered too complex for a machine - ImageNet, AlphaGo, Libratus - while doing so at a breakneck pace. The potential for AI/ML to impact every aspect of future coalition missions is immense, spanning next generation large-scale autonomous multi-domain systems, to complex autonomous decision making. A key challenge towards realizing this potential is the ability to both rapidly leverage global AI/ML advances for military data and problem regimes as well as accelerate AI/ML research to address unique operational challenges. However, the current stove-piped approach to AI/ML is ill-equipped to take advantage of these advancements in a timely and cost efficient manner nor is it able to address the unique learning challenges that NATO must solve. As a result, the pace of development and application of AI/ML to new problems and domains across individual nations and NATO is severely limited - even in the face of large interest and need.

Objectives

While NATO has a growing volume of valuable data obtained from operations, exercises, and other sources, currently, the Alliance lacks a common AI/ML environment to realize and capitalize on the emerging learning landscape. As a result, the Alliance is not positioned to exploit joint AI/ML advancements and overcome inherent learning challenges due to limited data access, model reuse, and the duplication of efforts across the nations and NATO bodies. Further, a common AI/ML computing and collaboration environment to foster research & development and to unite AI/ML and application domain experts is required. To address these challenges and to achieve the goal of a NATO ecosystem for bringing together data, algorithms, models, and experts, this effort can be decomposed into the following thrusts: • Develop and establish a globally accessible machine learning framework to reduce the time and cost to realize NATO ML capabilities. Through the developed framework, this activity will establish a NATO ML ecosystem that will persist and grow through contributions from all nations. Such an ecosystem will enable an accelerated means for ML research, development, and evaluation of new capabilities and has been shown to be successful in academic and commercial settings, driving contributed data repositories, collaboration, and competitions (ImageNet, FlickR, Kaggle). • Address the unique learning challenges inherent in NATO missions and environments. Specifically, this activity will identify real, operational learning challenges in data-efficiency, robustness, security, and interactive learning for seamless human-machine-teaming in a coalition environment. • Demonstrate a core set of ML applications that have game-changing potential across the Alliance. • Investigate techniques that allow the training of AI systems without releasing the training data (“bring SW to immobile data”). Make assets available without the necessity to release data.

Topics

Specific technologies and problems of interest will include but are not limited to robust and data-efficient learning, interactive learning, supervised / deep learning methods, data representation challenges for ML, continuous evaluation across data, algorithms, and models, transfer learning and domain adaptation, hyper-parameter selection, and AI/ML collaborative systems and learning systems.

Contact Panel Office