Research map / Label-efficient and robust learning / Recognition and long-tail
OccamNets: Mitigating Dataset Bias by Favoring Simpler Hypotheses
Robik Shrestha, Kushal Kafle, Christopher Kanan
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 |
|---|---|---|---|
| add multi-exit setup (vs single exit at end) (ablation) | OccamResNet-18 · Biased MNIST | accuracy: 35.9 -> 65.0 | Improved |
| add multi-exit setup (vs single exit at end) (ablation) | OccamResNet-18 · COCO-on-Places | accuracy: 35.0 -> 43.4 | Improved |
| add weighted output losses (vs Wj = 1) (ablation) | OccamResNet-18 · COCO-on-Places | accuracy: 40.8 -> 43.4 | Improved |
| add weighted output losses (vs Wj = 1) (ablation) | OccamResNet-18 · Biased MNIST | accuracy: 66.3 -> 65.0 | Hurt |
| add CAM suppression loss (ablation) | OccamResNet-18 · Biased MNIST | accuracy: 48.2 -> 65.0 | Improved |
| add CAM suppression loss (ablation) | OccamResNet-18 · COCO-on-Places | accuracy: 37.1 -> 43.4 | Improved |
| replace ResNet-18 with OccamResNet-18 | ERM-style training · Biased MNIST | accuracy: 36.8 -> 65.0 | Improved |
| replace ResNet-18 with OccamResNet-18 for gDRO | gDRO · BAR | accuracy: 38.7 -> 52.9 | Improved |
| run PGI on OccamResNet instead of ResNet | PGI · Biased MNIST | accuracy: 48.6 -> 69.6 | Improved |
| run PGI on OccamResNet instead of ResNet | PGI · COCO-on-Places | accuracy: 42.7 -> 43.6 | No clear effect |
| use OccamNet modifications on EfficientNet-B2 | EfficientNet-B2 · Biased MNIST | accuracy: 34.4 -> 59.2 | Improved |
| use OccamNet modifications on MobileNet-v3 | MobileNet-v3 · COCO-on-Places | accuracy: 34.9 -> 40.1 | Improved |
Closest papers by meaning
- AracNet: Revealing Debiasing Signals across Layers with Shallow Monitors (ECCV 2026 · similarity 0.88)
- DeNetDM: Debiasing by Network Depth Modulation (NeurIPS 2024 · similarity 0.88)
- Learning Not to Learn: Training Deep Neural Networks With Biased Data (CVPR 2019 · similarity 0.88)
- Simplicity Bias in Overparameterized Machine Learning (AAAI 2024 · similarity 0.87)
- Initial Guessing Bias: How Untrained Networks Favor Some Classes (ICML 2024 · similarity 0.86)
- A Simple Remedy for Dataset Bias via Self-Influence: A Mislabeled Sample Perspective (NeurIPS 2024 · similarity 0.86)
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.
