papers AI Learner
The Github is limit! Click to go to the new site.

Transfer Learning with Neural AutoML

2019-01-28
Catherine Wong, Neil Houlsby, Yifeng Lu, Andrea Gesmundo

Abstract

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.

Abstract (translated by Google)
URL

https://arxiv.org/abs/1803.02780

PDF

https://arxiv.org/pdf/1803.02780


Similar Posts

Comments