From Interpretability to Explainability

March 1, 2026

From Interpretability to Explainability

Explaining the model training procedure that transforms interpretation into explanation.

Deep image classifiers are accurate but opaque. This project asks a simple question, “why
does the model look there?”, and answers it with structured explanations that localize and
describe the evidence behind a prediction.

Beyond saliency heatmaps

A single saliency map tells you roughly where a model attends, but not how those regions
combine into a decision. We produce structured explanations that name the relevant regions
and relate them to the predicted class, making the reasoning easier to inspect and to
trust:

Why Does It Look There? Structured Explanations for Image Classification

(arXiv 2026) arXiv Preprint

Jiarui Li, Zixiang Yin, Samuel J. Landry, Zhengming Ding, Ramgopal R. Mettu

arXiv

The full reference is listed below.