MCMC-LIME: enhancing stability of LIME using Markov chain Monte Carlo approach
Bibliographic record
Abstract
This paper deeply investigates the stability issues of the visual explainability technique, called the local interpretable model-agnostic explanations (LIME)1 and presents an enhanced technique and our findings. Explainability is one of the important tasks in explainable artificial intelligence (XAI) systems where LIME tool has been widely used as an explainer model. However, LIME’s perturbation-based sampling introduces randomness, negatively affecting the stability of LIME’s explanations. In this paper, we utilize the unique sampling mechanism of the well-known Markov Chain Monte Carlo (MCMC) model to address this issue. However, a question arises: Will MCMC alone provide a semantically meaningful segmentation for visual explainability? To ensure this, we modify 40% of perturbation using MCMC while preserving 60% of the original perturbation of LIME and introduce additional stability-enhancing mechanisms. Adaptive Segment Merge (ASM) is employed to refine segmentation boundaries before applying MCMC, ensuring that segments align with meaningful visual structures. Additionally, Sign Entropy-Based Feature Elimination (SEFE) suppresses the effects of the remaining unstable segments by filtering them based on their sign entropy, further improving the robustness of the explanations. To evaluate the performance of the proposed technique, we adopt the average rank similarity (ARS), which quantifies the consistency of feature rankings across multiple runs. Our simulations, conducted on images of five dog breeds (Samoyed, Pomeranian, Newfoundland, Pug, and Cocker Spaniel) along with a Coyote, and a Robin image, demonstrate that MCMC-LIME achieves ARS scores of 0.893, 0.887, 0.851, 0.874, 0.906, 0.850, and 0.903 compared to 0.803, 0.862, 0.596, 0.676, 0.718, 0.771, and 0.846 for LIME. These results indicate that MCMC-LIME significantly enhances the stability and reliability of visual explanations, making it a more robust alternative to the original LIME.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".