Scalable Transfer Learning with Expert Models
This work addresses the need for scalable and efficient transfer learning in vision tasks, offering a practical solution with significant computational savings.
The paper tackles the problem of generic pre-trained representations not being tailored to specific downstream tasks by using expert representations selected via cheap performance proxies, resulting in a 2-3 orders of magnitude speed-up and outperforming baselines on over 20 diverse vision tasks.
Transfer of pre-trained representations can improve sample efficiency and reduce computational requirements for new tasks. However, representations used for transfer are usually generic, and are not tailored to a particular distribution of downstream tasks. We explore the use of expert representations for transfer with a simple, yet effective, strategy. We train a diverse set of experts by exploiting existing label structures, and use cheap-to-compute performance proxies to select the relevant expert for each target task. This strategy scales the process of transferring to new tasks, since it does not revisit the pre-training data during transfer. Accordingly, it requires little extra compute per target task, and results in a speed-up of 2-3 orders of magnitude compared to competing approaches. Further, we provide an adapter-based architecture able to compress many experts into a single model. We evaluate our approach on two different data sources and demonstrate that it outperforms baselines on over 20 diverse vision tasks in both cases.