Transfer Learning with Neural AutoML
This addresses the problem of computational efficiency for researchers and practitioners using AutoML, though it is incremental as it builds on existing RL-based methods.
The paper tackles the high computational cost of Neural AutoML by proposing Transfer Neural AutoML, which uses knowledge from prior tasks to speed up network design, reducing convergence time by over an order of magnitude on language and image classification tasks.
We reduce the computational cost of Neural AutoML with transfer learning. AutoML relieves human effort by automating the design of ML algorithms. Neural AutoML has become popular for the design of deep learning architectures, however, this method has a high computation cost. To address this we propose Transfer Neural AutoML that uses knowledge from prior tasks to speed up network design. We extend RL-based architecture search methods to support parallel training on multiple tasks and then transfer the search strategy to new tasks. On language and image classification tasks, Transfer Neural AutoML reduces convergence time over single-task training by over an order of magnitude on many tasks.