Research map / Learning theory and algorithms / Fairness
Towards Fairness-aware Adversarial Network Pruning
Lei Zhang, Zhibo Wang, Xiaowei Dong, Yunhe Feng, Xiaoyi Pang, Zhifei Zhang, Kui Ren
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 |
|---|---|---|---|
| optimize fairness jointly during pruning via adversarial discriminator instead of pruning only | ResNet-18 on CelebA Attractive, 80% pruned · CelebA Attractive | DEO: 0.5352 -> 0.2149 | Improved |
| joint adversarial pruning instead of debiasing first then pruning (Fair-SGDP) | SGDP pruning · CelebA Attractive 80% pruned | DEO: 0.3261 -> 0.2149 | Improved |
| prune from the dense pretrained network instead of training a sparse network from scratch (ablation) | ResNet-18 80% sparse · CelebA Attractive | DEO: 0.5071 -> 0.2149 | Improved |
| increase pruning ratio from 80% to 90% (ablation) | FARPrune · CelebA Attractive | DEO: 0.2149 -> 0.3143 | Hurt |
| increase pruning ratio from 80% to 95% (ablation) | FARPrune · CelebA Attractive | DI: 0.4849 -> 0.3891 | Hurt |
| apply method to ShuffleNet v2 instead of ResNet-18 (ablation) | FARPrune · CelebA Attractive 80% pruned | DEO: 0.5086 -> 0.3582 | Improved |
| apply method to MobileNet v2 instead of ResNet-18 (ablation) | FARPrune · CelebA Attractive 80% pruned | DEO: 0.5229 -> 0.3390 | Improved |
| search the mask on a randomly initialized network without training (ablation) | FARPrune · CelebA Attractive 80% pruned | DEO: 0.4418 (SGDP) -> 0.1959 | Improved |
| apply method on CelebA Blond Hair target | FARPrune vs Fair-LMW · CelebA Blond Hair 80% pruned | DI: 0.1181 vs 0.1188 | No clear effect |
| use fairness-aware pruning rather than SGDP, accuracy cost | FARPrune · CelebA Attractive random init 90% pruned | ACC: ACC 78.27% (SGDP) vs 76.18%, DEO 0.4675 vs 0.3444 | Mixed |
Closest papers by meaning
- On the Fairness ROAD: Robust Optimization for Adversarial Debiasing (ICLR 2024 · similarity 0.87)
- Device-Wise Federated Network Pruning (CVPR 2024 · similarity 0.86)
- GDeR: Safeguarding Efficiency, Balancing, and Robustness via Prototypical Graph Pruning (NeurIPS 2024 · similarity 0.86)
- DAFA: Distance-Aware Fair Adversarial Training (ICLR 2024 · similarity 0.86)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.85)
- Chasing Fairness in Graphs: A GNN Architecture Perspective (AAAI 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.
