Explainability and Interpretability in Robust and Secure AI Algorithms

EasyChair Preprint 13460

Abstract

Explainability and interpretability are crucial aspects in developing robust and secure AI algorithms. In this abstract, we explore the significance of these concepts and their implications for ensuring the reliability and trustworthiness of AI systems.

Explainability refers to the ability of an AI algorithm to provide clear and understandable explanations for its decisions and predictions. It enables users to comprehend the reasoning behind AI outputs, leading to improved transparency and accountability. Interpretability, on the other hand, focuses on understanding the internal workings of AI models, uncovering the relationships between input features and model outputs.

In the context of robustness and security, explainability plays a vital role in identifying potential vulnerabilities or biases in AI algorithms. By providing explanations, it becomes easier to detect and rectify issues related to fairness, bias, or incorrect decision-making. Interpretability assists in understanding how AI models process information, which aids in identifying potential weaknesses or areas where attacks or adversarial manipulations can occur.

However, achieving explainability and interpretability in AI algorithms is not without challenges. Complex deep learning models, trade-offs between explainability and performance, and the black box nature of certain algorithms pose obstacles. Nevertheless, various techniques have emerged to address these challenges, such as rule-based models, local interpretable model-agnostic explanations (LIME), SHapley Additive exPlanation (SHAP), and model distillation.

Keyphrases: Data Quality, Explainability, Security, adversarial attacks, interpretability, robustness

@booklet{EasyChair:13460,
  author    = {Edwin Frank},
  title     = {Explainability and Interpretability in Robust and Secure AI Algorithms},
  howpublished = {EasyChair Preprint 13460},
  year      = {EasyChair, 2024}}