Research map / Learning theory and algorithms / Prediction and decision losses
Controlling Neural Level Sets
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 |
|---|---|---|---|
| Use sampled decision-boundary distances for adversarial training | ConvNet-4a · MNIST, PGD^40 (ε_attack=0.3) | Rob. Acc. Xent: 99.23\% vs 96.78\% | Helped |
| Use sampled decision-boundary distances for adversarial training | ConvNet-4b · CIFAR10, PGD^20 (ε_attack=0.031) | Rob. Acc. Xent: 38.45\% vs 42.04\% | Hurt |
| Use sampled decision-boundary distances for adversarial training | ResNet-18 · CIFAR10, PGD^20 (ε_attack=0.031) | Rob. Acc. Xent: 79.74\% vs 53.49\% | Helped |
| Use sampled decision-boundary distances for adversarial training | ResNet-18 · CIFAR10, PGD^20 (ε_attack=0.031) | Rob. Acc. Margin: 43.17\% vs 51.22\% | Hurt |
| Reconstruct surfaces with neural level-set losses from raw point clouds | Ours · FAUST surface reconstruction | Chamfer L1: 11.54 -> 10.71 | Helped |
| Reconstruct surfaces with neural level-set losses from raw point clouds | Ours · FAUST surface reconstruction | Chamfer L2: 7.89 -> 7.32 | Helped |
| Evaluate robustness using margin-loss attacks in addition to Xent attacks | Ours · MNIST, ConvNet-4a | Rob. Acc.: 99.23\% -> 97.35\% | Hurt |
Closest papers by meaning
- Neural Set Function Extensions: Learning with Discrete Functions in High Dimensions (NeurIPS 2022 · similarity 0.77)
- Defining Neural Network Architecture through Polytope Structures of Datasets (ICML 2024 · similarity 0.75)
- Towards Understanding Hierarchical Learning: Benefits of Neural Representations (NeurIPS 2020 · similarity 0.75)
- Set-Valued Sensitivity Analysis of Deep Neural Networks (AAAI 2025 · similarity 0.75)
- Task structure and nonlinearity jointly determine learned representational geometry (ICLR 2024 · similarity 0.74)
- RNNs perform task computations by dynamically warping neural representations (NeurIPS 2025 · similarity 0.74)
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.