Verifiable Split Learning via zk-SNARKs

TL;DR

The paper tackles the lack of verifiability in split learning by introducing a verifiable split learning framework that inserts two-sided Groth16 zk-SNARK proofs into both forward and backward passes, enabling correctness verification without exposing private data. It designs a decentralized architecture with a Prover (PE) and Verifier (VE) and develops a simple arithmetic circuit alongside a quantization strategy to support proof construction. A comparative evaluation against a blockchain-enabled SL baseline demonstrates that zk-SNARKs yield strong verifiability and integrity but incur substantial computational overhead, while blockchain-only approaches are faster to process updates but do not verify computations. The work demonstrates the feasibility of auditable, privacy-preserving collaborative learning in high-stakes domains, and highlights the trade-offs between verifiability and efficiency, outlining directions for lighter proofs and optimization. Overall, the proposed approach advances trustworthy collaborative learning by providing rigorous computation verification alongside data privacy, with practical implications for regulated environments.

Abstract

Split learning is an approach to collaborative learning in which a deep neural network is divided into two parts: client-side and server-side at a cut layer. The client side executes its model using its raw input data and sends the intermediate activation to the server side. This configuration architecture is very useful for enabling collaborative training when data or resources are separated between devices. However, split learning lacks the ability to verify the correctness and honesty of the computations that are performed and exchanged between the parties. To this purpose, this paper proposes a verifiable split learning framework that integrates a zk-SNARK proof to ensure correctness and verifiability. The zk-SNARK proof and verification are generated for both sides in forward propagation and backward propagation on the server side, guaranteeing verifiability on both sides. The verifiable split learning architecture is compared to a blockchain-enabled system for the same deep learning network, one that records updates but without generating the zero-knowledge proof. From the comparison, it can be deduced that applying the zk-SNARK test achieves verifiability and correctness, while blockchains are lightweight but unverifiable.

Figures (3)

• Figure 1: Proposed split learning with zk-SNARK system architecture
• Figure 2: Blockchain overview architecture
• Figure 3: Performance results and comparison to blockchain support.