Research map / Label-efficient and robust learning / Recognition and long-tail
GRADIEND: Feature Learning within Neural Networks Exemplified through Biases
Jonathan Drechsel, Steffen Herbold
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 |
|---|---|---|---|
| combine GRADIEND Female/Male with INLP | GRADIEND Female/Male · Gender debiasing | mean proportional rank: 0.46 -> 0.88 | Improved |
| add GRADIEND Female/Male weight modification to INLP | INLP · Gender debiasing | mean proportional rank: 0.67 -> 0.88 | Improved |
| add GRADIEND Female/Male to SentDebias | SentDebias · Gender debiasing | mean proportional rank: 0.60 -> 0.64 | Improved |
| add INLP to CDA | CDA · Gender debiasing | mean proportional rank: 0.44 -> 0.75 | Improved |
| add INLP to Dropout | Dropout · Gender debiasing | mean proportional rank: 0.30 -> 0.71 | Improved |
| apply GRADIEND Female/Male to the base model | base models · Gender debiasing | mean proportional rank: 0.17 -> 0.46 | Improved |
| apply GRADIEND Asian/White to base model | base models · Race debiasing | mean proportional rank: 0.44 -> 0.58 | Improved |
| apply GRADIEND Black/White to base model | base models · Race debiasing | mean proportional rank: 0.44 -> 0.36 | Hurt |
| apply GRADIEND Asian/White to base model | base models · Race debiasing | LMS (%): 87.46 -> 82.00 | Hurt |
| apply GRADIEND Black/White to base model | base models · Race debiasing | LMS (%): 87.46 -> 87.37 | No clear effect |
| apply GRADIEND Christian/Muslim to base model | base models · Religion debiasing | LMS (%): 87.46 -> 84.76 | Hurt |
| apply GRADIEND Female/Male to base model | base models · Gender debiasing | LMS (%): 87.46 -> 86.72 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LEACE — LEACE: Perfect linear concept erasure in closed form (NeurIPS 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Interpretable Debiasing of Vision-Language Models for Social Fairness (CVPR 2026 · similarity 0.89)
- MABR: Multilayer Adversarial Bias Removal Without Prior Bias Knowledge (AAAI 2025 · similarity 0.87)
- The Devil is in the Neurons: Interpreting and Mitigating Social Biases in Language Models (ICLR 2024 · similarity 0.87)
- Diverse Yet Biased: Towards Mitigating Biases in Generative AI (Student Abstract) (AAAI 2024 · similarity 0.84)
- SANER: Annotation-free Societal Attribute Neutralizer for Debiasing CLIP (ICLR 2025 · similarity 0.84)
- Debiasing Algorithm through Model Adaptation (ICLR 2024 · similarity 0.84)
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.
