Research map / Label-efficient and robust learning / Recognition and long-tail
Semi-factual Explanations in AI
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Doctoral consortium proposal on semi-factual explanations for XAI, summarizing a survey and a proposed baseline (Most Distant Neighbor) and outlining planned methods and user studies.
Problem
Semi-factual explanations (what input changes do not alter the outcome) have received little attention in XAI compared with counterfactuals, and lack established computation methods and benchmarks.
Approach
Proposes the Most Distant Neighbor (MDN) method, a naive benchmark that retrieves the furthest same-class dataset instance from the query along a feature dimension, as a baseline for semi-factual computation; plans regularized MDNs, a conditional VAE generator, user studies, and image extension.
Key results
- MDN performed best in four of seven distance-based metrics, including distance from the query, compared with four historical semi-factual methods.
- MDN scored less well on sparsity than the historical methods.
- Proposed planned work includes a CVAE to generate MDN-like semi-factuals and a user study on comprehension.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the Most Distant Neighbor (MDN) method instead of the historical methods | Semi-factual generation · Semi-factual benchmark | distance from query | Helped |
| Use the Most Distant Neighbor (MDN) method instead of the historical methods | Semi-factual generation · Semi-factual benchmark | sparsity | Hurt |
Surprising findings
- The naive MDN baseline outperformed established semi-factual methods on most metrics, though it was weaker on sparsity.
Limitations
- MDN returns existing dataset instances and may not be an optimal semi-factual solution.
- MDN is weaker on sparsity (number of feature differences).
- Existing semi-factual methods are mostly tailored to tabular data; image extension is unexplored.
Experimental setup
Metrics: distance from query sparsity
Baselines: NUN LIME-based semi-factuals Kenny & Keane generative method nearest neighbor
Builds on: counterfactual explanations nearest unlike neighbor case-based reasoning
Closest papers by meaning
- Counterfactual Explanations for Misclassified Images: How Human and Machine Explanations Differ (Abstract Reprint) (AAAI 2024 · similarity 0.86)
- Counterfactual Explanations of Time Varying Rankings (Student Abstract) (AAAI 2025 · similarity 0.85)
- Even-if Explanations: Formal Foundations, Priorities and Complexity (AAAI 2025 · similarity 0.84)
- Generating Likely Counterfactuals Using Sum-Product Networks (ICLR 2025 · similarity 0.84)
- Position: Explain to Question not to Justify (ICML 2024 · similarity 0.84)
- Delivering Inflated Explanations (AAAI 2024 · similarity 0.83)
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.