NoEsis: Differentially Private Knowledge Transfer in Modular LLM Adaptation
This work addresses privacy concerns in modular LLM adaptation for domains like code generation, offering a solution that balances privacy and generalization, though it is incremental as it builds on existing techniques like differential privacy and parameter-efficient fine-tuning.
The paper tackles the problem of privacy leakage in modular large language models (LLMs) during adaptation by proposing NoEsis, a framework that integrates differential privacy with parameter-efficient fine-tuning to enable knowledge transfer across domains while providing provable privacy guarantees and protection against membership inference attacks, achieving at least 77% of the accuracy gap between non-shared and non-private baselines on code completion tasks.
Large Language Models (LLM) are typically trained on vast amounts of data from various sources. Even when designed modularly (e.g., Mixture-of-Experts), LLMs can leak privacy on their sources. Conversely, training such models in isolation arguably prohibits generalization. To this end, we propose a framework, NoEsis, which builds upon the desired properties of modularity, privacy, and knowledge transfer. NoEsis integrates differential privacy with a hybrid two-staged parameter-efficient fine-tuning that combines domain-specific low-rank adapters, acting as experts, with common prompt tokens, acting as a knowledge-sharing backbone. Results from our evaluation on CodeXGLUE showcase that NoEsis can achieve provable privacy guarantees with tangible knowledge transfer across domains, and empirically show protection against Membership Inference Attacks. Finally, on code completion tasks, NoEsis bridges at least 77% of the accuracy gap between the non-shared and the non-private baseline.