Hierarchical composition in multi-level and multi-agent
scenarios11:
Hierarchical composition deals with privacy guarantees in settings
involving multiple levels of data analysts or agents, each with varying
degrees of access and analysis capabilities. In these scenarios, privacy
needs to be preserved across multiple levels, ensuring that lower-level
agents cannot extract sensitive information about individuals.
This concept is particularly relevant in applications such as
collaborative machine learning, where different parties contribute data
or models. Hierarchical composition ensures that even if certain agents
have access to aggregated or partially processed data, they cannot
compromise the privacy of individual contributors.