Massive access or communication, emerging as one of six usage scenarios in 6G, has attracted considerable recent attention due to its potential to empower next-generation industrial cyber-physical systems such as smart grids, factory automation, industrial internet-of-things (IIoT), etc. However, to guarantee its QoS, the associated power control becomes computationally intractable with a huge number of users. In this paper, we present a tractable algorithm for power control in massive access, based on mean-field approximations. In particular, our aim is to maximize the overall throughput in each scheduling period, at the beginning of which each user has a finite number of backlogged bits. To achieve this goal and overcome the curse of dimensionality, a mean-field game (MFG) is formulated. Unfortunately, the formulated MFG is still a non-convex optimization problem. Enlightened by MAPEL, an efficient solver for non-convex power control problem, we leverage multiplicative linear fractional programming (MLFP) to tackle the non-convexity in our formulated MFG. Furthermore, the mean-field approximation assisted power control strategy requires low signaling overhead consumed for estimation and feedback of channel state information (CSI). Simulation results demonstrate that the proposed tractable power control attains substantial performance gains in both the overall throughput and computational complexity.

Previous studies on federated learning (FL) often encounter performance degradation due to data heterogeneity among different clients. In light of the recent advances in multimodal large language models (MLLMs), such as GPT-4v and LLaVA, which demonstrate their exceptional proficiency in multimodal tasks, such as image captioning and multimodal question answering. We introduce a novel federated learning framework, named Multimodal Large Language Model Assisted Federated Learning (MLLM-LLaVA-FL), which employs powerful MLLMs at the server end to address the heterogeneous and long-tailed challenges. Owing to the advanced cross-modality representation capabilities and the extensive open-vocabulary prior knowledge of MLLMs, our framework is adept at harnessing the extensive, yet previously underexploited, open-source data accessible from websites and powerful server-side computational resources. Hence, the MLLM-LLaVA-FL not only enhances the performance but also avoids increasing the risk of privacy leakage and the computational burden on local devices, distinguishing it from prior methodologies. Our framework has three key stages. Initially, we conduct global visual-text pretraining of the model. This pretraining is facilitated by utilizing the extensive open-source data available online, with the assistance of MLLMs. Subsequently, the pretrained model is distributed among various clients for local training. Finally, once the locally trained models are transmitted back to the server, a global alignment is carried out under the supervision of MLLMs to further enhance the performance. Experimental evaluations on established benchmarks, show that our framework delivers promising performance in the typical scenarios with data heterogeneity and long-tail distribution across different clients in FL.

Federated learning (FL) has attracted increasing attention as a promising technique to drive a vast number of edge devices with artificial intelligence. However, it is very challenging to guarantee the efficiency of a FL system in practice due to the heterogeneous computation resources on different devices. To improve the efficiency of FL systems in the real world, asynchronous FL (AFL) and semi-asynchronous FL (SAFL) methods are proposed such that the server does not need to wait for stragglers. However, existing AFL and SAFL systems suffer from poor accuracy and low efficiency in realistic settings where the data is non-IID distributed across devices and the on-device resources are extremely heterogeneous. In this work, we propose FedSEA - a semi-asynchronous FL framework for extremely heterogeneous devices. We theoretically disclose that the unbalanced aggregation frequency is a root cause of accuracy drop in SAFL. Based on this analysis, we design a training configuration scheduler to balance the aggregation frequency of devices such that the accuracy can be improved. To improve the efficiency of the system in realistic settings where the devices have dynamic on-device resource availability, we design a scheduler that can efficiently predict the arriving time of local updates from devices and adjust the synchronization time point according to the devices' predicted arriving time. We also consider the extremely heterogeneous settings where there exist extremely lagging devices that take hundreds of times as long as the training time of the other devices. In the real world, there might be even some extreme stragglers which are not capable of training the global model. To enable these devices to join in training without impairing the systematic efficiency, Fed-SEA enables these extreme stragglers to conduct local training on much smaller models. Our experiments show that compared with status quo approaches, FedSEA improves the inference accuracy by 44.34% and reduces the systematic time cost and local training time cost by 87.02× and 792.9×. FedSEA also reduces the energy consumption of the devices with extremely limited resources by 752.9×.

Federated learning aims to protect users' privacy while performing data analysis from different participants. However, it is challenging to guarantee the training efficiency on heterogeneous systems due to the various computational capabilities and communication bottlenecks. In this work, we propose FedSkel to enable computation-efficient and communication-efficient federated learning on edge devices by only updating the model's essential parts, named skeleton networks. FedSkel is evaluated on real edge devices with imbalanced datasets. Experimental results show that it could achieve up to 5.52x speedups for CONV layers' back-propagation, 1.82x speedups for the whole training process, and reduce 64.8% communication cost, with negligible accuracy loss.

Wireless federated learning (FL) holds the potential of preserving data privacy and reducing network traffic congestion, thereby attracting much recent attention. Due to the fading nature of wireless channels, wireless FL suffers from the random delay in each uplink and downlink transmission. As a result, how to analyze the overall random delay of a FL task over wireless fading channels remains open. To solve this challenging problem, we present a saddle point approximation based approach to obtain the distribution of the delay caused by communication in wireless FL systems. In particular, we obtain the uplink delay distribution and the downlink delay distribution by Lugannani-Rice formula. The overall delay distribution is then obtained through the convolution of those two distributions and the generating function. Simulation results demonstrate that the theoretical results provide accurate characterizations for the empirical results, which corroborates the validity of the analysis in this paper.

Federated learning (FL) is a collaborative machine learning paradigm, which enables deep learning model training over a large volume of decentralized data residing in mobile devices without accessing clients’ private data. Driven by the ever increasing demand for model training of mobile applications or devices, a vast majority of FL tasks are implemented over wireless fading channels. Due to the time-varying nature of wireless channels, however, random delay occurs in both the uplink and downlink transmissions of FL. How to analyze the overall time consumption of a wireless FL task, or more specifically, a FL’s delay distribution, becomes a challenging but important open problem, especially for delay-sensitive model training. In this paper, we present a unified framework to calculate the approximate delay distributions of FL over arbitrary fading channels. Specifically, saddle point approximation, extreme value theory (EVT), and large deviation theory (LDT) are jointly exploited to find the approximate delay distribution along with its tail distribution, which characterizes the quality-of-service of a wireless FL system. Simulation results will demonstrate that our approximation method achieves a small approximation error, which vanishes with the increase of training accuracy.

Predicting the popularity of online contents especially videos has drawn a lot of attention recently, since successful prediction of popularity can benefit many practical applications such as recommender systems and proactive caching, and help optimize the advertisement strategies or balance the throughput in the network. In this paper, we formulate a popularity prediction problem and present an Attention-based Long Short Term Memory (LSTM) with Feature Embedding method (ALFE) to tackle the popularity prediction problem. Several features including publish time, follower count, and type of the video are considered and compared. Experiments on a real world dataset show that our method outperforms other competitive baselines from existing works in terms of prediction accuracy. The attention mechanism and feature embedding contribute to the improvement of accuracy. Among all the features, timestamp of popularity and video duration are shown to be the most informative ones, due to the regularity and periodicity of human daily activities.

The coronavirus disease 2019 (COVID-19) has recently attracted extensive attention due to its serious impact on public health worldwide. In this paper, we study and verify that the popularity of virus-related content has a negative correlation with the epidemic spread by means of statistical analysis. Inspired by this result, a practical solution of recommender system is proposed for pushing virus-related content, aiming to gain insight about the newly discovered virus for people and thus reduce the epidemic spread to the utmost extent. First, we formulate the optimization of recommendation policy subject to quality of experience (QoE) loss constraints as a finite-horizon Constrained Markov Decision Problem (CMDP). To solve this problem, then, we present both enumeration and heuristic methods, from perspectives of achieving optimal recommendation policy and reducing computational complexity, respectively. Finally, our simulations validate the benefit of our solution by showing that to recommend virus-related content following our strategy does help slow down the spread of the epidemic.

Ultra-dense network (UDN) is recognized as a key technology for the fifth generation (5G) network to deliver high capacity. Due to cell densification, UDN suffers from heavy intercell interference (ICI). To mitigate ICI, an accurate interference modeling is essential. Compared to distributed mechanism, centralized interference management is more adept at addressing server ICI problem, especially in dense deployments. Hence, we adapt centralized radio access network (C-RAN) architecture to obtain global channel state data. In this paper, a novel regression-based uplink interference identification and signal-to-interference ratio (SINR) prediction algorithm is proposed for C-RAN based 5G UDN. By leveraging big data technology with in-depth analysis of the cause of ICI, this algorithm can precisely model the interference relationship and thus achieve very accurate SINR prediction. In addition, the proposed algorithm is highly efficient in computing, as demonstrated by complexity analysis.

Sparse code multiple access (SCMA) is being considered as a promising multiple access solution for 5G systems. A distinguishing feature of SCMA is that it combines the procedures of bit to constellation symbol mapping and subsequent spreading using multidimensional codebooks differentiated by users. Such codebooks dominate the system implementation as a main source of not only performance gain but also design complexity. This letter presents a joint constellation with mapping matrix design for SCMA codebooks, which formulates the constellations optimization as a nonconvex quadratically constrained quadratic programming problem based on a set of well-constructed mapping matrices. We elaborately solve the problem to achieve outperformance over existing SCMA design in terms of bit error rate (BER). For improving practicality, an approximate approach is further proposed to reduce the complexity significantly with a limited BER loss.

With the development of next generation mobile communications, the underlay coexistence problem of the OFDMA based Secondary System(SS) with LTE-A systems becomes more and more important, which yet has not been studied in a systematic way. In contrast to other Primary Systems(PS), the LTE-A system puts high demands on the low complexity of the coexistence strategies. This paper focuses on the resource allocation and interference mitigation issues in the aforementioned scenario, whose objective is to protect the spectrum utilization priority of PS as well as utilize secondary resource efficiently. The difficulty lies in the fact that even the subproblem, or power allocation with interference, is NP-Hard. Therefore, this paper will propose a two-phase resource allocation algorithm using maximum weighted Matching in the subcarrier allocation phase and interference Quantizing in the power allocation phase, referred to as the MQ algorithm. As presented in this paper, the MQ algorithm enjoys the advantage of polynomial complexity of O(KJ3+ LKJ), where K, J and L denote the number of SSs, subcarriers and quantizing steps, respectively. The simulation results will show that the proposed MQ algorithm is capable of achieving near optimal system and user throughputs, which are close to the exhaustive searching algorithm.