Researchers have introduced EAGLE, a novel post-hoc model-agnostic explanation framework designed to enhance the trustworthiness of opaque machine learning models. EAGLE formulates perturbation selection as an information-theoretic active learning problem, adaptively sampling perturbations to maximize expected information gain. This approach efficiently learns a linear surrogate model, providing feature importance scores along with confidence estimates. Empirical results demonstrate EAGLE's superiority over existing methods like Tilia, US-LIME, GLIME, and BayesLIME in terms of explanation reproducibility, neighborhood stability, and perturbation sample quality. AI
IMPACT Enhances trust in ML models by providing more reliable and reproducible explanations with uncertainty estimates.
RANK_REASON Research paper introducing a new methodology for ML model explanations. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
