Research map / Learning theory and algorithms / Prediction and decision losses
Interpreting and Boosting Dropout from a Game-Theoretic View
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 |
|---|---|---|---|
| Add interaction loss with a tuned weight | AlexNet and VGG networks · CIFAR-10 | accuracy: 67.5 -> 70.0 | Helped |
| Add interaction loss with a tuned weight | RN-18, RN-34, VGG-16, and VGG-19 · Tiny ImageNet | accuracy: 47.4 -> 52.2; 46.0 -> 49.6; 36.8 -> 42.8; 32.6 -> 45.2 | Helped |
| Add interaction loss with a tuned weight | BERT · SST-2 | accuracy: 77.1 -> 78.0 | Helped |
| Add interaction loss with a too-large weight (ablation) | AlexNet · CIFAR-10 | accuracy: 70.0 -> 64.3 | Hurt |
| Increase interaction-loss weight from 500.0 to 1000.0 (ablation) | VGG-13 · CIFAR-10 | accuracy: 65.9 -> 66.2 | Helped |
| Increase interaction-loss weight beyond its best setting (ablation) | BERT · SST-2 | accuracy: 78.0 -> 77.0 -> 76.5 | Hurt |
| Use dropout instead of no regularization (ablation) | RN-18 · Tiny ImageNet | accuracy: 48.8 -> 47.4 | Hurt |
| Use dropout instead of no regularization (ablation) | RN-18 · CelebA gender estimation | accuracy: 92.7 -> 92.1 | Hurt |
| Increase dropout rate (ablation) | AlexNet · CIFAR-10 | interaction strength: 21.73 -> 3.93 | Helped |
| Increase dropout rate to 0.9 (ablation) | AlexNet · CIFAR-10 | accuracy: 67.5 -> 57.0 | Hurt |
| Apply interaction loss to a low convolutional layer rather than a high fully connected layer (ablation) | AlexNet, VGG-13, VGG-16, VGG-19 · CIFAR-10, Tiny ImageNet, CelebA | accuracy | Mixed |
| Apply dropout at different network positions (ablation) | AlexNet and VGG-11 · CIFAR-10 | accuracy: AlexNet: 67.5, 69.4, 68.4, 67.6; VGG-11: 60.9, 58.1, 57.6, 60.9 | Mixed |
Closest papers by meaning
- Demystifying Dropout (ICML 2019 · similarity 0.90)
- Guided Dropout (AAAI 2019 · similarity 0.88)
- On the Implicit Bias of Dropout (ICML 2018 · similarity 0.88)
- Activation by Interval-wise Dropout: A Simple Way to Prevent Neural Networks from Plasticity Loss (ICML 2025 · similarity 0.85)
- Stochastic Modified Equations and Dynamics of Dropout Algorithm (ICLR 2024 · similarity 0.85)
- Effects of Dropout on Performance in Long-range Graph Learning Tasks (NeurIPS 2025 · similarity 0.84)
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.