Concrete dropout

Conference item

Dropout is used as a practical tool to obtain uncertainty estimates in large vision models and reinforcement learning (RL) tasks. But to obtain well-calibrated uncertainty estimates, a grid-search over the dropout probabilities is necessary— a prohibitive operation with large models, and an impossible one with RL. We propose a new dropout variant which gives improved performance and better calibrated uncertainties. Relying on recent developments in Bayesian deep learning, we use a continuous relaxation of dropout’s discrete masks. Together with a principled optimisation objective, this allows for automatic tuning of the dropout probability in large models, and as a result faster experimentation cycles. In RL this allows the agent to adapt its uncertainty dynamically as more data is observed. We analyse the proposed variant extensively on a range of tasks, and give insights into common practice in the field where larger dropout probabilities are often used in deeper model layers.

Authors: Gal, Y; Hron, J; Kendall, A

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: NIPS Foundation
• Host title: Advances in Neural Information Processing Systems 31 (NIPS 2017)
• Journal: Advances in Neural Information Processing Systems 31 (NIPS 2017)
• Publication date: 2018-07-01
• Acceptance date: 2017-09-04