ProToMEx: Rapid, Interpretable Explanations via Structured Representations
ProToMEx introduces a new explainability paradigm using Probabilistic Topic Modeling to identify complex, combinatorial patterns in classifier decisions.
Traditional post-hoc explainers often rely on simple feature attribution, which fails to capture the nuanced logic behind model predictions. ProToMEx addresses this by leveraging structured representations to articulate how specific combinations of features drive decision-making, offering a more interpretable alternative for complex machine learning models.