Research map / Optimization / Implicit bias and generalization
Beyond Gradient Descent for Regularized Segmentation Losses
Check 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 ADM with α-expansion to optimize the grid CRF loss instead of using GD (ablation) | Weakly supervised segmentation with Deeplab-largeFOV · PASCAL VOC 2012 val | mIOU: 60.4 -> 61.7 | Helped |
| Add ADM with α-expansion to optimize the grid CRF loss instead of using GD (ablation) | Weakly supervised segmentation with Deeplab-MSc-largeFOV · PASCAL VOC 2012 val | mIOU: 61.2 -> 62.9 | Helped |
| Add ADM with α-expansion to optimize the grid CRF loss instead of using GD (ablation) | Weakly supervised segmentation with Deeplab-VGG16 · PASCAL VOC 2012 val | mIOU: 63.3 -> 65.2 | Helped |
| Add ADM with α-expansion to optimize the grid CRF loss instead of using GD (ablation) | Weakly supervised segmentation with ResNet-101 · PASCAL VOC 2012 val | mIOU: 71.7 -> 72.8 | Helped |
| Add ADM with α-expansion to optimize the grid CRF loss instead of using GD (ablation) | Grid CRF loss minimization with Deeplab-LargeFOV · PASCAL VOC 2012 training set | grid CRF loss: 2.52 -> 2.41 | Helped |
| Add ADM with α-expansion to optimize the grid CRF loss instead of using GD (ablation) | Grid CRF loss minimization with Deeplab-LargeFOV · PASCAL VOC 2012 validation set | grid CRF loss: 2.51 -> 2.33 | Helped |
| Use grid CRF + ADM instead of dense CRF + GD | Weakly supervised segmentation with Deeplab-VGG16 · PASCAL VOC 2012 val | mIOU: 64.4 -> 65.2 | Helped |
| Use grid CRF + ADM instead of dense CRF + GD | Weakly supervised segmentation with ResNet-101 · PASCAL VOC 2012 val | mIOU: 72.9 -> 72.8 | Hurt |
| Use ADM with grid CRF rather than GD with grid CRF (ablation) | Weakly supervised segmentation with Deeplab-MSc-largeFOV · PASCAL VOC 2012 val, pixels within 16 pixels of boundaries | pixel-wise accuracy: 74.8 -> 76.7 | Helped |
| Use ADM with grid CRF rather than GD with grid CRF (ablation) | Weakly supervised segmentation with Deeplab-MSc-largeFOV · PASCAL VOC 2012 val, pixels within 8 pixels of boundaries | pixel-wise accuracy: 68.4 -> 71.4 | Helped |
Closest papers by meaning
- Normalized Cut Loss for Weakly-Supervised CNN Segmentation (CVPR 2018 · similarity 0.89)
- Penalizing Gradient Norm for Efficiently Improving Generalization in Deep Learning (ICML 2022 · similarity 0.88)
- Newton Losses: Using Curvature Information for Learning with Differentiable Algorithms (NeurIPS 2024 · similarity 0.85)
- Loss Landscape Characterization of Neural Networks without Over-Parametrization (NeurIPS 2024 · similarity 0.85)
- Generalized Gradient Norm Clipping & Non-Euclidean $(L_0,L_1)$-Smoothness (NeurIPS 2025 · similarity 0.85)
- Implicit bias of SGD in $L_2$-regularized linear DNNs: One-way jumps from high to low rank (ICLR 2024 · 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.