Selective Pre-training for Private Fine-tuning
This addresses privacy and efficiency constraints for domain-specific text prediction, though it appears incremental in combining existing techniques.
The paper tackles the problem of training small, private language models for applications like email clients by showing that selective pre-training on a public subset guided by private data is crucial for achieving state-of-the-art performance with differential privacy, with smaller models matching larger non-private ones.
Text prediction models, when used in applications like email clients or word processors, must protect user data privacy and adhere to model size constraints. These constraints are crucial to meet memory and inference time requirements, as well as to reduce inference costs. Building small, fast, and private domain-specific language models is a thriving area of research. In this work, we show that a careful pre-training on a \emph{subset} of the public dataset that is guided by the private dataset is crucial to train small language models with differential privacy. On standard benchmarks, small models trained with our new framework achieve state-of-the-art performance. In addition to performance improvements, our results demonstrate that smaller models, through careful pre-training and private fine-tuning, can match the performance of much larger models that do not have access to private data. This underscores the potential of private learning for model compression and enhanced efficiency.