A CONCEPTUAL MODEL OF A CRYPTOGRAPHIC PROTOCOL FOR SECURE PARAMETER EXCHANGE IN FEDERATED LEARNING BASED ON HOMOMORPHIC ENCRYPTION

Abstract

Federated learning enables multiple participants to jointly train a machine learning model without centrally transferring their original training data. However, transmitted parameters and gradients may disclose information about participants’ local data. Homomorphic encryption enables certain computations on encrypted values while preserving the confidentiality of the underlying data. Nevertheless, encryption alone does not specify how messages should be exchanged, validated, or handled when protocol rules are violated.
This paper aims to develop a conceptual model of a cryptographic protocol for secure parameter exchange in federated learning. The model defines system participants, a threat model, parameter states, transition rules, and an admissibility function for cryptographic operations. It also specifies the algorithmic sequence of parameter exchange, message verification, and conditions for accepting an aggregated result. The work presents a protocol design rather than a completed implementation or experimental evaluation.