À-la-carte Prompt Tuning (APT): Combining Distinct Data Via Composable Prompting
This enables constructing bespoke models for individual users based on access rights and preferences without retraining from scratch, addressing privacy and customization in machine learning.
The paper tackles the problem of combining distinct data sources for inference by introducing À-la-carte Prompt Tuning (APT), which allows prompts trained on different data to be arbitrarily composed at inference time, achieving accuracy within 5% of models trained on the union of data sources and state-of-the-art performance on continual learning benchmarks.
We introduce À-la-carte Prompt Tuning (APT), a transformer-based scheme to tune prompts on distinct data so that they can be arbitrarily composed at inference time. The individual prompts can be trained in isolation, possibly on different devices, at different times, and on different distributions or domains. Furthermore each prompt only contains information about the subset of data it was exposed to during training. During inference, models can be assembled based on arbitrary selections of data sources, which we call "à-la-carte learning". À-la-carte learning enables constructing bespoke models specific to each user's individual access rights and preferences. We can add or remove information from the model by simply adding or removing the corresponding prompts without retraining from scratch. We demonstrate that à-la-carte built models achieve accuracy within $5\%$ of models trained on the union of the respective sources, with comparable cost in terms of training and inference time. For the continual learning benchmarks Split CIFAR-100 and CORe50, we achieve state-of-the-art performance.