Research map / Label-efficient and robust learning / Recognition and long-tail
DropMax: Adaptive Variational Softmax
Hae Beom Lee, Juho Lee, Saehoon Kim, Eunho Yang, Sung Ju Hwang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Use input-adaptive stochastic class dropout with variational inference instead of Base Softmax | CNN · MNIST-1K | test error (%): 7.09 -> 5.32 | Improved |
| Use input-adaptive stochastic class dropout with variational inference instead of Base Softmax | CNN · MNIST-5K | test error (%): 2.13 -> 1.64 | Improved |
| Use input-adaptive stochastic class dropout with variational inference instead of Base Softmax | CNN · MNIST-55K | test error (%): 0.65 -> 0.59 | Improved |
| Use input-adaptive stochastic class dropout with variational inference instead of Base Softmax | ResNet-34 · CIFAR-10 | test error (%): 7.90 -> 7.67 | Improved |
| Use input-adaptive stochastic class dropout with variational inference instead of Base Softmax | ResNet-34 · CIFAR-100 | test error (%): 30.60 -> 29.87 | Improved |
| Use input-adaptive stochastic class dropout with variational inference instead of Base Softmax | ResNet-18 · AWA | test error (%): 30.29 -> 26.91 | Improved |
| Use input-adaptive stochastic class dropout with variational inference instead of Base Softmax | ResNet-18 · CUB-200-2011 | test error (%): 48.84 -> 41.07 | Improved |
| Replace label-informed approximate posterior with q=p (ablation) | DropMax · CUB-200-2011 | test error (%): 41.07 -> 42.08 | Hurt |
| Replace stochastic adaptive class dropout with supervised deterministic attention (ablation) | DropMax · CUB-200-2011 | test error (%): 41.07 -> 47.35 | Hurt |
| Use DropMax instead of Base Softmax on fine-grained datasets | ResNet-18 · AWA and CUB-200-2011 | test error (%): 30.29 -> 26.91; 48.84 -> 41.07 | Improved |
Closest papers by meaning
- One-vs-Each Approximation to Softmax for Scalable Estimation of Probabilities (NeurIPS 2016 · similarity 0.86)
- Variational Dropout Sparsifies Deep Neural Networks (ICML 2017 · similarity 0.85)
- Softmax is not Enough (for Adaptive Conformal Classification) (ICLR 2026 · similarity 0.85)
- Adaptive Robust Learning using Latent Bernoulli Variables (ICML 2024 · similarity 0.85)
- DRoP: Distributionally Robust Data Pruning (ICLR 2025 · similarity 0.84)
- Stochastic Modified Equations and Dynamics of Dropout Algorithm (ICLR 2024 · 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.
