Research map / Large language models / Language, safety and interpretability
ADEPT: A DEbiasing PrompT Framework
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 |
|---|---|---|---|
| use continuous prompt tuning with the manifold-inspired debiasing criterion | BERT-LARGE-UNCASED · SEAT C6, C7, and C8; CrowS-Pairs | bias benchmark score | Helped |
| use ADEPT debiasing before downstream fine-tuning | BERT-LARGE-UNCASED · GLUE SST-2, MRPC, and RTE | accuracy | Helped |
| use ADEPT debiasing before downstream fine-tuning | BERT-LARGE-UNCASED · StereoSet filtered overall | ICAT: 70.288 -> 74.759 | Helped |
| replace prompt tuning with full-model fine-tuning using ADEPT's criterion (ablation) | BERT-LARGE-UNCASED · SEAT and CrowS-Pairs | bias benchmark score | Hurt |
| add the ADEPT prompt during downstream fine-tuning, rather than fine-tuning only the base model (ablation) | ADEPT · GLUE MRPC | accuracy: 84.6 -> 85.0 | Helped |
| add the ADEPT prompt during downstream fine-tuning, rather than fine-tuning only the base model (ablation) | ADEPT · GLUE SST-2 | accuracy: 93.3 -> 92.7 | Hurt |
| filter attribute-word sentence lists with fewer than 30 examples (ablation) | ADEPT-finetuning · StereoSet-Intrasentence | ICAT: 65.282 -> 65.605 | Helped |
| equalize sentence counts for pairwise attribute words (ablation) | ADEPT-finetuning · StereoSet-Intrasentence | ICAT: 65.282 -> 65.343 | Helped |
| use quantity-100 attribute corpus instead of the raw corpus (ablation) | ADEPT-finetuning · StereoSet-Intrasentence | ICAT: 65.282 -> 67.712 | Helped |
| use quantity-1000 attribute corpus instead of the raw corpus (ablation) | ADEPT-finetuning · StereoSet-Intrasentence | ICAT: 65.282 -> 66.920 | Helped |
| use quantity-10000 attribute corpus instead of the raw corpus (ablation) | ADEPT-finetuning · StereoSet-Intrasentence | ICAT: 65.282 -> 66.713 | Helped |
Closest papers by meaning
- A General Framework for Implicit and Explicit Debiasing of Distributional Word Vector Spaces (AAAI 2020 · similarity 0.88)
- Biases Mitigation and Expressiveness Preservation in Language Models: A Comprehensive Pipeline (Student Abstract) (AAAI 2024 · similarity 0.87)
- Collapsed Language Models Promote Fairness (ICLR 2025 · similarity 0.87)
- Evaluating the Efficacy of Prompting Techniques for Debiasing Language Model Outputs (Student Abstract) (AAAI 2024 · similarity 0.86)
- ADePT: Adaptive Decomposed Prompt Tuning for Parameter-Efficient Fine-tuning (ICLR 2025 · similarity 0.86)
- BayesPrompt: Prompting Large-Scale Pre-Trained Language Models on Few-shot Inference via Debiased Domain Abstraction (ICLR 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.