Overview

• AI Challenges (presented fall 2020)

AI Data Readiness

• Guide (2 pages)
• MIT Open Course Ware Video (45 minutes)

Data Sharing Agreements

• Example Air Force Signable Agreement (Author: David Jacobs)
• Example Air Force Click-Thru Agreement (Author: David Jacobs)

Additional Resources

• AI Acquisition Handbook

Example Challenges from other Domains

• GraphChallenge, YOHO, MNIST, HPC Challenge, ImageNet, VAST, Moments-in-Time
• Neural Information Processing Systems (NeurIPS) Competition Track

Tools

• rAI-Toolbox: The DAF-MIT AI Accelerator (AIA) and MIT Lincoln Laboratory have released an open-source toolbox, called the rAI-Toolbox. This PyTorch-centric library provides ML researchers and developers with tools to help evaluate and enhance the robustness of AI models, i.e., improve a model’s ability to maintain performance in the presence of adversarially-perturbed or naturally-corrupted data inputs.This toolbox is a great example of a collaboration between an AIA-funded effort (RAIDEN) and an MIT Lincoln Laboratory project (ASERT) to provide a capability for the DoD and broader community.  As Olivia Brown, co-lead of the RAIDEN effort, describes, “developing robust AI presents a multifaceted challenge that often stretches the limits of existing ML tooling and frameworks.” The rAI-toolbox reduces this complexity by providing a set of core building blocks for generating perturbed data that are scalable and compose naturally with other popular ML model training frameworks. While the toolbox is currently focused on supporting robust AI R&D, the team has plans to add additional capabilities to address a broader range of responsible AI principles in the future.
• Intelipedia (for government readers only, CAC-required): https://intellipedia.intelink.gov