Work

CEFC: Calibrated Explanation Fault Containment for Dependable Explainable AI

Year: 2026

Type: software

Abstract: This capsule provides the source code, configurations, reproducibility workflow, and experimental artifacts for CEFC (Calibrated Explanation Fault Containment), a dependable XAI framework designed to reduce user-visible explanation failures through calibrated risk estimation and selective EXPLAIN/ABSTAIN decisions. CEFC treats explanation delivery as a fault-containment problem: an independent operational criterion identifies explanation failures, multiple reliability signals estimate instance-level failure risk, isotonic calibration maps risk scores to probabilities, and a disjoint certification set determines whether an explanation can be released under a prescribed risk budget. The capsule includes a lightweight CPU DEMO for reproducibility under constrained resources and the configuration, checkpoints, aggregate results, statistical analyses, and publication artifacts corresponding to the full 30-seed experimental protocol.

Source: Code Ocean

DOI: 10.24433/co.4994814.v1

Authors: Danilo M. Souza, Sérgio M. M. Fernandes, Alexssandra K. S. Correia

Institutions: Universidade de Pernambuco, Universidade Federal de Pernambuco

Language: English

Topic: Explainable Artificial Intelligence (XAI)

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7208820445

Cites:

Cited by:

Related to: 10