Federated AI for building AI Solutions across Multiple Agencies
This tackles data privacy and regulatory compliance issues for government agencies, but it is incremental as it applies an existing federated learning approach to a specific domain.
The paper addresses the challenge of training AI models across government agencies due to regulatory restrictions on data sharing, and it demonstrates that federated AI enables model creation without moving data, achieving results comparable to centralized training with minimal accuracy loss (e.g., within 2%).
The different sets of regulations existing for differ-ent agencies within the government make the task of creating AI enabled solutions in government dif-ficult. Regulatory restrictions inhibit sharing of da-ta across different agencies, which could be a significant impediment to training AI models. We discuss the challenges that exist in environments where data cannot be freely shared and assess tech-nologies which can be used to work around these challenges. We present results on building AI models using the concept of federated AI, which al-lows creation of models without moving the training data around.