Practical Trustworthiness Model for DNN in Dedicated 6G Application

TL;DR

This paper proposes a practical trustworthiness model for DNNs in 6G applications, focusing on Automatic Modulation Recognition in Terahertz communications, emphasizing attributes like robustness, sensitivity, and security.

Contribution

It introduces a novel trustworthiness evaluation framework tailored for DNNs in 6G, addressing key attributes relevant to future wireless communication applications.

Findings

Trustworthiness attributes are crucial for DNNs in 6G.

The model evaluates robustness, sensitivity, and security.

Results highlight the importance of trustworthiness in AI for 6G.

Abstract

Artificial intelligence (AI) is considered an efficient response to several challenges facing 6G technology. However, AI still suffers from a huge trust issue due to its ambiguous way of making predictions. Therefore, there is a need for a method to evaluate the AI's trustworthiness in practice for future 6G applications. This paper presents a practical model to analyze the trustworthiness of AI in a dedicated 6G application. In particular, we present two customized Deep Neural Networks (DNNs) to solve the Automatic Modulation Recognition (AMR) problem in Terahertz communications-based 6G technology. Then, a specific trustworthiness model and its attributes, namely data robustness, parameter sensitivity, and security covering adversarial examples, are introduced. The evaluation results indicate that the proposed trustworthiness attributes are crucial to evaluate the trustworthiness of DNN for this 6G application.