Research map / Reinforcement learning / Policy gradient and actor-critic
Concrete Dropout
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace manually tuned dropout with learned Concrete Dropout | DenseNet semantic segmentation · CamVid | IoU: 67.2 -> 67.4 | Helped |
| Add MC sampling at evaluation (ablation) | Concrete Dropout DenseNet · CamVid | IoU: 67.2 -> 67.4 | Helped |
| Use Concrete Dropout instead of standard dropout | 2-layer NN · UCI concrete | Test RMSE: 4.50 -> 4.46 | Helped |
| Use Concrete Dropout instead of standard dropout | 3-hidden-layer MLP · MNIST | Accuracy | No clear effect |
| Adapt dropout probability as data grows instead of holding it fixed | Model-based RL dynamics model · Pendulum balancing | Cumulative reward | No clear effect |
| Optimize the dropout probability using the variational objective | Concrete Dropout · UCI datasets | Test log likelihood/nats | Mixed |
| Add the final M-step for precision (ablation) | CDropout · UCI boston | Test log likelihood/nats: -2.72 -> -2.35 | Helped |
| Add the final M-step for precision (ablation) | CDropout · UCI energy | Test log likelihood/nats: -2.30 -> -0.84 | Helped |
| Use a continuous Concrete relaxation of Bernoulli masks | Dropout probability optimization · Model training | Gradient estimator variance | Helped |
| Allow dropout probabilities to adapt rather than fixing them | Dropout neural networks · Synthetic data with increasing dataset size | Epistemic uncertainty | Helped |
| Learn separate dropout probabilities by layer | Semantic segmentation DenseNet · CamVid | Uncertainty calibration RMSE | Helped |
Closest papers by meaning
- Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning (ICML 2016 · similarity 0.92)
- Contextual Dropout: An Efficient Sample-Dependent Dropout Module (ICLR 2021 · similarity 0.88)
- Dropout Inference in Bayesian Neural Networks with Alpha-divergences (ICML 2017 · similarity 0.88)
- Activation by Interval-wise Dropout: A Simple Way to Prevent Neural Networks from Plasticity Loss (ICML 2025 · similarity 0.86)
- Enhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding (NeurIPS 2025 · similarity 0.85)
- Dynamic Uncertainty Estimation for Offline Reinforcement Learning (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.