Research map / Architectures and efficiency / Pruning and efficient tuning
Recover Fair Deep Classification Models via Altering Pre-trained Structure
Yanfu Zhang, Shangqian Gao, Heng Huang
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 |
|---|---|---|---|
| prune corrupted weights then adversarially fine-tune instead of fine-tuning the full model (ablation) | Adv.Ft · CIFAR-10S | accuracy: 0.917 -> 0.920 | Improved |
| prune corrupted weights then adversarially fine-tune instead of fine-tuning the full model (ablation) | Adv.Ft · CIFAR-10S | bias: 0.051 -> 0.033 | Improved |
| graft external filters into corrupted weights instead of noise (ablation) | Grafting · CIFAR-10S | accuracy: 0.909 -> 0.927 | Improved |
| graft internal filters instead of noise (ablation) | Grafting · CIFAR-10S | bias: 0.045 -> 0.013 | Improved |
| use network augmentation instead of external grafting | ExtGraft · CIFAR-10S | accuracy: 0.927 -> 0.931 | Improved |
| use network augmentation instead of external grafting | ExtGraft · CIFAR-10S | bias: 0.028 -> 0.014 | Improved |
| use grafting or NetAug instead of prune + adversarial fine-tune on tabular data | Prune + AdvFt · Adult/COMPAS | accuracy and bias | No clear effect |
| use NetAug at bias level 99% instead of Adv.Ft | Adv.Ft · CIFAR-10S (bias level 99%) | accuracy: 0.916 -> 0.915 | No clear effect |
| use ExtGraft instead of Adv.Ft at bias level 90% | Adv.Ft · CIFAR-10S (bias level 90%) | accuracy: 0.937 -> 0.945 | Improved |
| use NetAug on CelebA instead of Adv.Ft | Adv.Ft · CelebA | accuracy: 0.91 -> 0.94 | Improved |
| prune unfair model guided by ExtGraft teacher at 99% imbalance (ablation) | Unfair full model · CIFAR-10S | bias: 0.115 -> 0.082 | Improved |
Closest papers by meaning
- Benchmarking Bias Mitigation Algorithms in Representation Learning through Fairness Metrics (NeurIPS 2021 · similarity 0.89)
- FairNet: Dynamic Fairness Correction without Performance Loss via Contrastive Conditional LoRA (NeurIPS 2025 · similarity 0.88)
- Classes Are Not Equal: An Empirical Study on Image Recognition Fairness (CVPR 2024 · similarity 0.87)
- Fair Bilevel Neural Network (FairBiNN): On Balancing fairness and accuracy via Stackelberg Equilibrium (NeurIPS 2024 · similarity 0.86)
- FairSIN: Achieving Fairness in Graph Neural Networks through Sensitive Information Neutralization (AAAI 2024 · similarity 0.86)
- Adversarial Fairness Network (AAAI 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.
