Research map / Architectures and efficiency / RNNs and normalization
Bias Also Matters: Bias Attribution for Deep Neural Network Explanation
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 bias terms to training (ablation) | VGG-11 · CIFAR10 | test accuracy (%): 87.0 -> 90.9 | Helped |
| add bias terms to training (ablation) | VGG-11 · CIFAR100 | test accuracy (%): 62.8 -> 66.8 | Helped |
| add bias terms to training (ablation) | VGG-11 · FMNIST | test accuracy (%): 94.1 -> 94.7 | Helped |
| keep only wx term at test instead of full output (ablation) | VGG-11 · CIFAR10 | test accuracy (%): 90.9 -> 71.5 | Hurt |
| keep only b term at test instead of full output (ablation) | VGG-11 · CIFAR10 | test accuracy (%): 90.9 -> 62.2 | Hurt |
| keep only wx term at test (ablation) | VGG-11 · CIFAR100 | test accuracy (%): 66.8 -> 40.3 | Hurt |
| keep only b term at test (ablation) | VGG-11 · CIFAR100 | test accuracy (%): 66.8 -> 36.5 | Hurt |
| keep only wx term at test (ablation) | VGG-11 · FMNIST | test accuracy (%): 94.7 -> 76.1 | Hurt |
| keep only b term at test (ablation) | VGG-11 · FMNIST | test accuracy (%): 94.7 -> 24.6 | Hurt |
| add bias attribution on top of gradient attribution | BBp · ImageNet | share of top-10% pixels not in gradient attribution: 59.4%, 56.3% and 55.2% | Helped |
| add bias attribution on top of gradient attribution | BBp · STL-10 | share of top-10% pixels not in gradient attribution: 50.52%, 48.29%, and 43.94% | Helped |
| exclude bias of early layers from BBp attribution (ablation) | BBp options 1-3 · ImageNet VGG-11 | attribution quality (qualitative) | Mixed |
Closest papers by meaning
- Interpreting Deep Neural Networks with Relative Sectional Propagation by Analyzing Comparative Gradients and Hostile Activations (AAAI 2021 · similarity 0.88)
- Learning Not to Learn: Training Deep Neural Networks With Biased Data (CVPR 2019 · similarity 0.88)
- Initial Guessing Bias: How Untrained Networks Favor Some Classes (ICML 2024 · similarity 0.88)
- Towards better understanding of gradient-based attribution methods for Deep Neural Networks (ICLR 2018 · similarity 0.88)
- NeuronTune: Towards Self-Guided Spurious Bias Mitigation (ICML 2025 · similarity 0.87)
- How to Probe: Simple Yet Effective Techniques for Improving Post-hoc Explanations (ICLR 2025 · 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.