Machine Learning-based Resource Allocation and Management of Communication Networks
One of the significant challenges in the next generation AI-driven wireless communication network that still needs more investigation is the spectrum management among the various heterogeneous RATs. Due to their robustness in making optimal decisions in dynamic and stochastic environments, machine learning techniques and particularly Deep Reinforcement Learning-based algorithms have attracted considerable research interest recently in the context of resource allocation and network management. My objective is to expand my current research in this area by discovering new frontiers where complexity is increased with multiple radio access technologies working together heterogeneously and resources distributed amongst multiple agents.
Optimization of Communication Networks for Distributed Edge Learning
Another critical aspect of the use of AI is its performance concerning network parameters. Federated Learning is a nascent edge learning framework for data privacy preservation. Federated Learning allows training machine learning models by aggregating local learning models at the edge servers instead of users’ raw data to preserve users’ privacy. Federated learning requires multiple training iterations. During each iteration round, smart devices locally train their local models using their local data and send these models to the edge server for aggregation to build a global model. Then, the global model is sent back to the end devices, and this whole process is repeated until convergence. With thousands of end devices continuously updating and sending their local models to the edge, network constraints, such as bandwidth underutilization, latency, and spectrum scarcity, can severely affect and degrade Federated Learning-based applications’ performance. Hence, this part of my research plan aims to solve network optimization problems for distributed edge learning by using machine learning. In a way, AI-based algorithms will help to improve AI-based applications by improving communication network parameters and participant selection. I have already started working on this topic by using federated learning in various settings.
Privacy-preservation and Security with Machine Learning
In AI-based algorithms, by knowing the model weights, learning parameters, or attacks can be launched, such as model inversion attacks, which may reveal information on the user’s sensitive data. I propose using privacy-preserving data aggregation and evaluating the models using encrypted input data to solve inversion attacks. Until now, I have used ML-based algorithms with multiple layers of privacy-preservation for different applications, such as indoor positioning and network intrusion detection. I plan to measure privacy-preservation and increase the security of privacy by applying various new methods without hindering the performance of AI-based algorithms.
As the second branch of this topic, I am planning to apply different intrusion detection algorithms along with AI-based intrusion prevention algorithms to secure the distributed edge learning. AI-based intrusion prevention mechanisms will work proactively to detect, learn about attacker profiles, and respond to threats before they become attacks.
Explainable AI (XAI) for Signal Processing and Wireless Communication
As future work and a new frontier for my research, I am planning to work on explainable AI. One of the most essential and distinctive research subjects is explainable AI research as the fourth research pillar. The deep learning-based methodologies boom has embraced the research world. However, many researchers have distanced themselves from ML-based techniques, as most of the papers approach DL as a magic black-box, and assume ML-based approach is missing the hikmah (حكمة), wisdom due to low interpretability. In reality, it is an automated statistical application on either data or action-reward pairs depending on the type of system. Thus we should reveal what is happening inside this black-box by visualization and similar techniques to interpret its results. Still, especially in signal processing and wireless communications, AI has still got a lot to offer. One of these gift boxes to be opened is explainable AI. Applying proper procedures and using explainable AI tools, I am planning to provide insights about signal processing and wireless communication systems where complexity hides the beauty.