PulseAugur
LIVE 03:33:05
research · [2 sources] ·
0
research

GRALIS framework unifies linear attribution methods for deep neural networks

Researchers have introduced GRALIS, a novel mathematical framework designed to unify various linear attribution methods used in Explainable AI (XAI). This framework establishes a canonical representation for attribution functionals, encompassing methods like SHAP, Integrated Gradients, and LIME, while excluding nonlinear ones. GRALIS offers simultaneous guarantees across multiple axiomatic properties, including completeness, sensitivity, and multi-scale aggregation, outperforming individual methods in theoretical validation. AI

Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →

IMPACT Provides a unified theoretical basis for comparing and improving XAI methods, potentially leading to more reliable model explanations.

RANK_REASON The cluster contains an arXiv preprint detailing a new theoretical framework for XAI methods.

Read on arXiv stat.ML →

COVERAGE [2]

  1. arXiv cs.LG TIER_1 · Raimondo Fanale ·

    GRALIS: A Unified Canonical Framework for Linear Attribution Methods via Riesz Representation

    arXiv:2605.05480v1 Announce Type: new Abstract: The main XAI attribution methods for deep neural networks -- GradCAM, SHAP, LIME, Integrated Gradients -- operate on separate theoretical foundations and are not formally comparable. We present GRALIS (Gradient-Riesz Averaged Locall…

  2. arXiv stat.ML TIER_1 · Raimondo Fanale ·

    GRALIS: A Unified Canonical Framework for Linear Attribution Methods via Riesz Representation

    The main XAI attribution methods for deep neural networks -- GradCAM, SHAP, LIME, Integrated Gradients -- operate on separate theoretical foundations and are not formally comparable. We present GRALIS (Gradient-Riesz Averaged Locally-Integrated Shapley), a mathematical framework …