Machine unlearning (MU) aims to remove the influence of specific ''forget'' data from a trained model while preserving its knowledge of the remaining ''retain'' data. Existing MU methods based on label manipulation or model weight perturbations often achieve limited unlearning effectiveness. To address this, we introduce CoUn, a novel MU framework inspired by the observation that a model retrained from scratch using only retain data classifies forget data based on their semantic similarity to the retain data. CoUn emulates this behavior by adjusting learned data representations through contrastive learning (CL) and supervised learning, applied exclusively to retain data. Specifically, CoUn (1) leverages semantic similarity between data samples to indirectly adjust forget representations using CL, and (2) maintains retain representations within their respective clusters through supervised learning. Extensive experiments across various datasets and model architectures show that CoUn consistently outperforms state-of-the-art MU baselines in unlearning effectiveness. Additionally, integrating our CL module into existing baselines empowers their unlearning effectiveness.

Decentralized federated learning (DFL) enables clients to train a neural network model in a device-to-device (D2D) manner without central coordination. In practical systems, DFL faces challenges due to dynamic topology changes, timevarying channel conditions, and limited computational capability of the clients. These factors can affect the learning performance and efficiency of DFL. To address the aforementioned challenges, in this paper, we propose a graph neural network (GNN)–based algorithm to minimize the total delay and energy consumption on training and improve the learning performance of DFL in D2D wireless networks. In our proposed GNN, a multihead graph attention mechanism is used to capture different features of clients and wireless channels. We design a neighbor selection module which enables each client to select a subset of its neighbors for the participation of model aggregation. We develop a decoder that enables each client to determine its transmit power and computational resource. Experimental results show that our proposed algorithm achieves a lower total delay and energy consumption on training when compared with five baseline schemes. Furthermore, by properly selecting a subset of neighbors for each client, our proposed algorithm achieves similar testing accuracy to the full participation scheme.

360°videos require significant bandwidth to provide an immersive viewing experience. Wireless systems using terahertz (THz) frequency band can meet this high data rate demand. However, self-blockage is a challenge in such systems. To ensure reliable transmission, this paper explores THz-enabled 360° video streaming through multiple multi-antenna access points (APs). Guaranteeing users’ quality of experience (QoE) requires accurate viewport prediction to determine which video tiles to send, followed by asynchronous bitrate selection for those tiles and beamforming design at the APs. To address users’ privacy and data heterogeneity, we propose a content-based viewport prediction framework, wherein users’ head movement prediction models are trained using a personalized federated learning (PFL) algorithm. To address asynchronous decision-making for tile bitrates and dynamic THz link connections, we formulate the optimization of bitrate selection and beamforming as a macro-action decentralized partially observable Markov decision process (MacDec-POMDP) problem. To efficiently tackle this problem for multiple users, we develop two deep reinforcement learning (DRL) algorithms based on multi-agent actor-critic methods and propose a hierarchical learning framework to train the actor and critic networks. Experimental results show that our proposed approach provides a higher QoE when compared with three benchmark algorithms.

360° videos demand substantial bandwidth to deliver an immersive viewing experience to users. In wireless networks, this high data rate demand can be accommodated by utilizing the terahertz (THz) frequency band. However, THz band communications are susceptible to self-blockage. To ensure reliable transmission, this paper studies the streaming of$360^{\circ}$videos over THz wireless systems using multiple multi-antenna access points (APs). Users' requests for video tiles give rise to an optimization problem that involves asynchronous bitrate selection for those tiles and beamforming design for the APs. We formulate this problem as a macro-action decentralized partially observable Markov decision process (MacDec-POMDP). To efficiently tackle this problem for multiple users, we propose an asynchronous deep reinforcement learning (DRL) algorithm using a multi-agent actor-critic method to determine the bitrate selection policy. The APs' beamforming is determined by solving an optimization problem using the weighted minimum mean square error (WMMSE) algorithm. Results show that our proposed approach provides a higher average quality of experience (QoE) for the users when compared with two benchmark algorithms.

Decentralized federated learning (DFL) enables clients to train a neural network model in a device-to-device (D2D) manner without central coordination. In practical systems, DFL faces challenges due to the dynamic topology changes, time-varying channel conditions, and limited computational capability of devices. These factors can affect the performance of DFL. To address the aforementioned challenges, in this paper, we propose a graph neural network (GNN)-based approach to minimize the total delay on training and improve the learning performance of DFL in D2D wireless networks. In our proposed approach, a multi-head graph attention mechanism is used to capture different features of clients and channels. We design a neighbor selection module which enables each client to select a subset of its neighbors for the participation of model aggregation. We develop a decoder which enables each client to determine its transmit power and CPU frequency. Experimental results show that our proposed algorithm can achieve a lower total delay on training when compared with three baseline schemes. Furthermore, the proposed algorithm achieves similar performance on the testing accuracy when compared with the full participation scheme.

Viewport prediction is a key enabler for 360° video streaming over wireless networks. To improve the prediction accuracy, a common approach is to use a content-based viewport prediction model. Saliency detection based on traditional convolutional neural networks (CNNs) suffers from distortion due to equirectangular projection. Also, the viewers may have their own viewing behavior and are not willing to share their historical head movement with others. To address the aforementioned issues, in this paper, we first develop a saliency detection model using a spherical CNN (SPCNN). Then, we train the viewers’ head movement prediction model using personalized federated learning (PFL). Finally, we propose a content-based viewport prediction framework by integrating the video saliency map and the head orientation map of each viewer using fusion techniques. The experimental results show that our proposed framework provides higher average accuracy and precision when compared with three state-of-the-art algorithms from the literature.

Network slicing is a promising technique for wireless service providers to support enhanced mobile broadband (eMBB) and ultra-reliable low-latency communication (URLLC) services in a shared radio access network (RAN) infrastructure. In this paper, we apply numerology, mini-slot based transmission, and punctured scheduling techniques to support eMBB and URLLC network slices. For efficient allocation of radio resources (e.g., physical resource blocks, transmit power) to the users, we formulate RAN slicing problem as a multi-timescale problem. To solve this problem and address the dynamics of the traffic, we propose a hierarchical deep learning framework. Specifically, in each long time slot, the service provider employs a deep reinforcement learning (DRL) algorithm to determine the slice configuration parameters. The eMBB and URLLC schedulers use their own attention-based deep neural network (DNN) algorithm to allocate radio resources to their corresponding users in each short and mini time slot, respectively. Simulation results show that the proposed framework can achieve a higher aggregate throughput and a higher service level agreement (SLA) satisfaction ratio compared to some other RAN slicing approaches, including the resource proportional placement algorithm, decomposition and relaxation based resource allocation algorithm, and distributed bandwidth optimization algorithm.

Software-defined networking (SDN) and network functions virtualization (NFV) enable service providers to accommodate diversified service requests in the fifth generation (5G) core networks. Given the time-varying traffic demand of the service requests, it is crucial for service providers to embed the service function chains (SFCs) of the service requests in the network to support load balancing, and to minimize the reconfiguration overhead due to virtual network functions (VNFs) migration while satisfying their quality of service (QoS) requirements. In this paper, we study a delay-aware VNF migration problem for embedding SFCs in a network with limited processing resource capacity for NFV-enabled nodes. We formulate it as a mixed-integer nonlinear optimization problem. We decompose this problem into two subproblems for stateful VNF mapping and allocation of processing resources, where the second subproblem is a convex optimization problem. To solve the first subproblem, we propose an algorithm based on deep neural network (DNN) with attention mechanism for learning the stochastic policy of a near-optimal VNF mapping. Simulation results show that our proposed algorithm provides a solution which is very close to the optimal solution obtained by solving a mixed-integer quadratically constrained programming problem.

Efficient allocation of resources (i.e., physical resource blocks (PRBs), transmit power) for remote radio heads (RRHs) in the fifth generation (5G) cloud radio access network (C-RAN) is crucial for the mobile network operators (MNOs) to support different use cases with diverse quality of service (QoS) requirements. In this paper, we study the resource allocation of enhanced mobile broadband (eMBB) and ultra-reliable lowlatency communications (URLLC) network slices in a 5G C-RAN. We formulate the resource allocation problem as a mixedinteger nonlinear program. We address the isolation between eMBB and URLLC network slices and the uncertainty in the traffic load by using the chance constraint. We consider short packet transmission to enable URLLC data transmission with low latency and high reliability. We propose an algorithm based on penalized successive convex approximation to determine a suboptimal solution of the formulated problem. The proposed algorithm has a polynomial time complexity. Simulation results show that the proposed algorithm on average achieves 30% higher throughput when compared with a baseline scheme that only optimizes the transmit power of users.