Research map / Label-efficient and robust learning / Recognition and long-tail
An Additive Instance-Wise Approach to Multi-class Model Interpretation
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 |
|---|---|---|---|
| Instance-wise multi-class additive explainer (AIM) vs. L2X/LIME/VIBI/FI baselines | word-level text explanations, K=10 · IMDB, HateXplain, AG News | Faithfulness (%): AIM highest on all three datasets (e.g., 99.62 vs 84.80 for best baseline L2X on IMDB) | Helped |
| Inference from explainer weight matrix W instead of selector probabilities pi (top-10 by pi) (ablation) | AIM · IMDB | Faithfulness (%): nearly 20% drop in faithfulness when inferring from selector | Helped |
| Joint training of explainer with selector vs. selector-only training (ablation) | AIM · IMDB | Faithfulness (%) | Helped |
| Learnable selector-based local sampling vs. heuristic (LIME-style uniform) sampling (ablation) | AIM explainer training · IMDB | Faithfulness (%) | Helped |
| Instance-wise amortized training (AIM) vs. per-instance LIME/FI optimization | training time · IMDB | Training time (minutes): 11.19 vs 551.42 for LIME and 311.42 for FI | Helped |
| Instance-wise AIM vs. LIME for stability | explanation stability · IMDB | IoU (%): 6.11 vs 1.44 (LIME) | Helped |
| Multi-class explainer producing class-specific explanations from one W matrix vs. LIME | class-specific explanations · HateXplain and AG News | Pairwise IoU (%): 3.69 vs 66.91 (HateXplain); 9.24 vs 51.09 (AG News) | Helped |
| Multi-class explainer AIM vs. LIME on class-specific faithfulness | IMDB class-specific explanations · IMDB | Class-specific Faithfulness (%): 98.09 vs 50.02 for class 0 | Helped |
Closest papers by meaning
- Local vs. Global Interpretability: A Computational Complexity Perspective (ICML 2024 · similarity 0.87)
- Selective Explanations (NeurIPS 2024 · similarity 0.87)
- Explainable Neural Networks with Guarantee: A Sparse Estimation Approach (AAAI 2025 · similarity 0.87)
- MCPNet: An Interpretable Classifier via Multi-Level Concept Prototypes (CVPR 2024 · similarity 0.87)
- Additive Models Explained: A Computational Complexity Approach (NeurIPS 2025 · similarity 0.86)
- Provably Better Explanations with Optimized Aggregation of Feature Attributions (ICML 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.