Research map / Label-efficient and robust learning / Recognition and long-tail
Probing Classifiers are Unreliable for Concept Removal and Detection
Abhinav Kumar, Chenhao Tan, Amit Sharma
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 |
|---|---|---|---|
| run INLP iterations on a clean classifier | INLP null-space removal · Synthetic-Text | main-task accuracy: 100\% -> 50\% | Hurt |
| higher predictive correlation kappa | INLP null-space removal · Synthetic-Text | speed of main-task feature removal | Hurt |
| early stopping INLP when main performance drops | INLP null-space removal · Synthetic-Text kappa=0.8 | DeltaProb: 25\% | Hurt |
| apply INLP to clean classifier on MultiNLI | INLP null-space removal · MultiNLI | main-task accuracy | Hurt |
| adversarial removal instead of ERM | RoBERTa classifier · MultiNLI | spuriousness score | Mixed |
| adversarial removal instead of ERM | RoBERTa classifier · Twitter-PAN16, Twitter-AAE | spuriousness score | No clear effect |
| adversarial removal instead of ERM | RoBERTa classifier · Synthetic-Text | spuriousness score | Hurt |
| feed clean classifier to adversarial removal | AR · MultiNLI | spuriousness score | Hurt |
Closest papers by meaning
- When Are Concepts Erased From Diffusion Models? (NeurIPS 2025 · similarity 0.86)
- Removing Spurious Concepts from Neural Network Representations via Joint Subspace Estimation (ICML 2024 · similarity 0.85)
- Erasing Undesirable Concepts in Diffusion Models with Adversarial Preservation (NeurIPS 2024 · similarity 0.84)
- Erasing Conceptual Knowledge from Language Models (NeurIPS 2025 · similarity 0.83)
- Circumventing Concept Erasure Methods For Text-To-Image Generative Models (ICLR 2024 · similarity 0.82)
- SAeUron: Interpretable Concept Unlearning in Diffusion Models with Sparse Autoencoders (ICML 2025 · similarity 0.82)
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.
