A Survey on XAI for 5G and Beyond Security: Technical Aspects, Challenges and Research Directions

TL;DR

The goal of using XAI in the security domain of 5G and beyond is to allow the decision-making processes of ML-based security systems to be transparent and comprehensible to 5G and beyond stakeholders, making the systems accountable for automated actions.

Abstract

With the advent of 5G commercialization, the need for more reliable, faster, and intelligent telecommunication systems is envisaged for the next generation beyond 5G (B5G) radio access technologies. Artificial Intelligence (AI) and Machine Learning (ML) are immensely popular in service layer applications and have been proposed as essential enablers in many aspects of 5G and beyond networks, from IoT devices and edge computing to cloud-based infrastructures. However, existing 5G ML-based security surveys tend to emphasize AI/ML model performance and accuracy more than the models' accountability and trustworthiness. In contrast, this paper explores the potential of Explainable AI (XAI) methods, which would allow stakeholders in 5G and beyond to inspect intelligent black-box systems used to secure next-generation networks. The goal of using XAI in the security domain of 5G and beyond is to allow the decision-making processes of ML-based security systems to be transparent and comprehensible to 5G and beyond stakeholders, making the systems accountable for automated actions. In every facet of the forthcoming B5G era, including B5G technologies such as ORAN, zero-touch network management, and end-to-end slicing, this survey emphasizes the role of XAI in them that the general users would ultimately enjoy. Furthermore, we presented the lessons from recent efforts and future research directions on top of the currently conducted projects involving XAI.

Figures (11)

• Figure 1: XAI is applicable in many facets of each layer in the 3-layered 5G and beyond architecture. XAI methods deployed around security algorithms in the perception layer would enhance the interpretability of the devices and contain additional information as they are the closest accessible points for the general users. The ubiquitous use of ML applications in the network layer requires quantifiable approaches to interpretability. In the network layer, XAI will become an essential component in the interactions between operators and the ML model. Interpretations generated in the first and second layers approach the users through the service layer. The comprehensiveness and relevance of the explanations will determine the attraction of new clients and the retention of existing clients for service providers.
• Figure 2: This figure outlines the paper structure. We lay down the context with motivation for XAI in the B5G security, our contributions, and the outline for the paper. Stemming from the theme set in the introduction, we answer the questions of whats and whys for XAI in the background section. The rest of the paper extends the minutiae of the XAI's potential in B5G security aspects, current standardizations, and projects. Finally, the Lessons learned and future research directions conclude the main takeaways of the survey.
• Figure 3: XAI Taxonomy. Pre-model XAI explains the training data used to build AI models (e.g.,Principal component analysis (PCA), and t-Distributed Stochastic Neighbor Embedding (t-SNE) ). In-model XAI refers to transparent AI models that are self-explanatory (e.g., decision trees, random forests). Post-hoc XAI models explain the results of the trained AI models (e.g., LIME, SHAP).
• Figure 4: XAI Stakeholders: Different levels of influence that each stakeholder has on the systems and their respective explainability requirements.
• Figure 6: Depiction of the ORAN architecture with an additional layer of XAI-based security. The left-hand block diagram shows the AI/ML workflow in ORAN, and the right-hand figure shows the overall architecture for ORAN. Input data obtained through the RAN nodes via the E2 interface are vulnerable to poisoning and manipulation by adversaries. With pre-model XAI, some of the problematic data can be filtered out. Furthermore, any undetected threats can be detected through post-hoc/in-model XAI methods (in Non-RT RIC) during the validation phase before deploying them to the inference host. Verification of x/rApps that use black-box (for either proprietary reasons or complexity) models can be aided with the XAI methods that unveil the reasons for the model's adversarial behaviours. This would also avoid subsequent attacks, such as DoS attacks, which can cause control and policy conflicts in RICs.oranalliance2021wg2AIMLpolese2022understanding