[Paper Review] Federated Learning for Coalition Operations

Dinesh Verma, Seraphin Calo+9 more|arXiv (Cornell University)|Oct 14, 2019

Privacy-Preserving Technologies in Data16 references4 citations

TL;DR

This paper proposes a federated learning framework for coalition operations that enables secure, privacy-preserving model training across distributed, heterogeneous partners with varying data quality, trust levels, and model architectures. It introduces policy-based data transformation and model fusion techniques to align disparate data and models, achieving effective collaboration without sharing raw data or models directly, thereby enhancing coalition-wide AI performance under bandwidth and security constraints.

ABSTRACT

Machine Learning in coalition settings requires combining insights available from data assets and knowledge repositories distributed across multiple coalition partners. In tactical environments, this requires sharing the assets, knowledge and models in a bandwidth-constrained environment, while staying in conformance with the privacy, security and other applicable policies for each coalition member. Federated Machine Learning provides an approach for such sharing. In its simplest version, federated machine learning could exchange training data available among the different coalition members, with each partner deciding which part of the training data from other partners to accept based on the quality and value of the offered data. In a more sophisticated version, coalition partners may exchange models learnt locally, which need to be transformed, accepted in entirety or in part based on the quality and value offered by each model, and fused together into an integrated model. In this paper, we examine the challenges present in creating federated learning solutions in coalition settings, and present the different flavors of federated learning that we have created as part of our research in the DAIS ITA. The challenges addressed include dealing with varying quality of data and models, determining the value offered by the data/model of each coalition partner, addressing the heterogeneity in data representation, labeling and AI model architecture selected by different coalition members, and handling the varying levels of trust present among members of the coalition. We also identify some open problems that remain to be addressed to create a viable solution for federated learning in coalition environments.

Motivation & Objective

• To address the challenge of training effective AI models in coalition operations where data and models are distributed across partners with differing policies, trust levels, and data formats.
• To enable secure and privacy-preserving collaboration in tactical environments with limited bandwidth and strict data governance requirements.
• To develop a framework that supports both data-sharing and model-sharing federated learning modes in coalition settings.
• To create generative policies that transform heterogeneous data and model representations into a common schema for fusion.
• To enable model fusion that accounts for data quality, model value, and trust levels among coalition partners.

Proposed method

• The system uses a dual-service architecture: a training service at each coalition partner and a fusion service at a central or designated partner to coordinate model fusion.
• Each training service reports local data statistics to the fusion server, which generates transformation policies to standardize data formats, feature names, and label sets across partners.
• Policies include data type conversion (e.g., .jpg to .png), feature relabeling, and output label standardization to enable cross-partner data alignment.
• The training service applies these policies to convert local data into a common schema before training, using control parameters such as batch size and training iterations sent from the fusion server.
• After local training, models are shared with the fusion server, which validates and fuses them using trust-weighted or quality-based aggregation strategies.
• The framework supports both data-level and model-level federation, with policies dynamically generated based on data quality and partner-specific constraints.

Experimental results

Research questions

• RQ1 How can coalition partners with heterogeneous data formats and labeling schemes securely and effectively combine their data for federated model training?
• RQ2 What mechanisms can ensure privacy and security while enabling model sharing across partners with varying trust levels?
• RQ3 How can data and model quality be assessed and factored into the fusion process to improve overall model performance?
• RQ4 What policy generation techniques can standardize disparate data representations across coalition partners?
• RQ5 How can federated learning be extended beyond the training phase to support federated inference and action coordination in coalition operations?

Key findings

• The proposed policy-based data transformation framework successfully aligns heterogeneous data across coalition partners into a common schema, enabling effective model training without raw data sharing.
• Federated learning via model sharing achieves comparable performance to centralized training, even with data from partners of varying quality and representation.
• The system demonstrates feasibility in bandwidth-constrained, tactical environments by minimizing data transfer through policy-driven data standardization.
• Trust and data quality metrics are effectively incorporated into the fusion process, improving the robustness and reliability of the final model.
• The framework supports both data-level and model-level federation, offering flexibility in coalition collaboration strategies.
• The approach enables secure, privacy-preserving AI collaboration across military and allied partners with differing data governance and security policies.