Browse Teach Curate

Rule-Guided LLM Translation of Python to MP-SPDZ Privacy Code

A rule-based framework (SPDZCoder) that pairs large language models with expert-derived transformation rules to translate ordinary Python into data-oblivious MP-SPDZ secure-multi-party-computation code without extra training, applying the rules in a three-stage pipeline and evaluated on the purpose-built SPDZEval benchmark.

Data Text

2501.00363

This paper introduces SPDZCoder, a rule-based framework that combines large language models with expert knowledge to generate privacy-computing code, specifically translating ordinary Python into MP-…