Federated Learning (FL) reshapes the AI model training paradigm by enabling privacy-preserving collaborative learning, where models are trained across distributed clients without sharing raw data, but only model parameters or updates. This learning can be centralized or distributed. Centralized FL (CFL) may suffer from latency and lack of robustness due to the reliance on a coordinating server for model convergence. On the other hand, Decentralized Federated Learning (DFL) enables direct collaboration among participating devices without relying on a central server. Each device can independently connect to other devices and share model parameters. In such collaborative training paradigm, model convergence in the presence of various deployment topologies, AI model types, Non-IID data distribution, and training strategies demands systematic analysis to realize their practical deployment in critical applications such as intelligent transportation, smart factories, and real-time surveillance. Some works have attempted to conduct only partial analysis and completely neglected incorporating Non-IID data distribution, a critical factor in practical deployment of DFL in mentioned applications. This work conducts a systematic analysis on the convergence of DFL considering a wide range of AI models (e.g., classical, deep neural networks, and Large Language Models), network topologies (e.g., linear, ring, star, and mesh), training strategies (e.g., continuous and aggregate), and degree of Non-IID data distributions. The analysis includes both mathematical formulations and their implementation and evaluation using real-world data. The results confirm that the convergence rate of the models is inversely proportional to the degree of Non-IID data distribution. Moreover, judicial selection of network topologies and training strategies can aid in this convergence process for the practical edge deployment of DFL.

A content delivery network (CDN) aims to reduce the content delivery latency to end-users by using distributed cache servers. Nevertheless, deploying and maintaining cache servers on a large scale is very expensive. To solve this problem, CDN providers have developed a new content delivery strategy: allowing end-users’s IoT edge devices to share their storage/bandwidth resources. This new edge CDN platform must address two core questions: (1) how can we incentivize end users to share IoT devices? (2) how can we facilitate a safe and transparent content transaction environment for end users? This paper introduces SmartSharing, a new content delivery network solution to address these questions. In smartSharing, the over-the-top (OTT) IoT devices belonging to end-users are used as mini-cache servers. To motivate end users to share the idle devices and storage/bandwidth resources, SmartSharing designs the content delivery schedule and the pricing scheme based on game theory and machine learning algorithms (specifically, a tailored Expectation-Maximization (EM) algorithm). To facilitate content trading among end users, SmartSharing creates a secure and transparent transaction platform based on smart contracts in Ethereum. In addition, SmartSharing’s performance evaluation is through trace-driven simulations in the real world and a prototype using content metadata and the achieved pricing schemes. The evaluation results show that CDN providers, end users and content providers can all benefit from our SmartSharing framework.

Root cause analysis is the process of identifying the underlying factors responsible for triggering anomaly detection alarms. In the context of anomaly detection for Internet of Things (IoT) traffic, these alarms can be triggered by various factors, not all of which are malicious attacks. It is crucial to determine whether a malicious attack or benign operations cause an alarm. To address this challenge, we propose an innovative root cause analysis system called score-variational autoencoder (VAE), designed to complement existing IoT anomaly detection systems based on the federated learning (FL) framework. Score-VAE harnesses the full potential of the VAE network by integrating its training and testing schemes strategically. This integration enables Score-VAE to effectively utilize the generation and reconstruction capabilities of the VAE network. As a result, it exhibits excellent generalization, lifelong learning, collaboration, and privacy protection capabilities, all of which are essential for performing root cause analysis on IoT systems. We evaluate Score-VAE using real-world IoT trace data collected from various scenarios. The evaluation results demonstrate that Score-VAE accurately identifies the root causes behind alarms triggered by IoT anomaly detection systems. Furthermore, Score-VAE outperforms the baseline methods, providing superior performance in discovering root causes and delivering more accurate results.

Clustered federated learning (CFL) is a promising solution to address the non-IID problem in the spatial domain for federated learning (FL). However, existing CFL solutions overlook the non-IID issue in the temporal domain and lack consideration of time efficiency. In this work, we propose a novel approach, calledClusterFLADS, which takes advantage of the false predictions of the inappropriate global models, together with knowledge of temperature scaling and catastrophic forgetting to reveal distributional similarities between the training data (of different clusters) and the test data. Additionally, we design an efficient feature extraction scheme by exploiting the role of each layer in a neural network's learning process. By strategically selecting model parameters and using PCA for dimensionality reduction,ClusterFLADSeffectively improves clustering speed. We evaluateClusterFLADSusing real-world IoT trace data in various scenarios. Our results show thatClusterFLADSaccurately and efficiently clusters clients, achieving a$100\%$true positive rate and low false positives across various data distributions in both the spatial and temporal domains.

It is often needed to update deep learning-based detection models in traffic anomaly detection systems for the Internet of Things (IoT) because of mislabeled samples or device firmware upgrades. Machine unlearning, a technique that quickly updates the anomaly detection model without retraining the model from scratch, has recently attracted much research attention. We propose a novel machine unlearning method, called virtual federated learning approach (ViFLa), which groups training data based on estimated unlearning probability and treats each group as a virtual client in the federated learning framework. Since the virtual clients are physically in the same machine, ViFLa only leverages the concept of data/local models isolation in federated learning without incurring any network communication. ViFLa adopts an attention-based aggregation method called enhanced class distribution weighted sum (ECDWS) to tackle the nonindependent and identically distributed (non-iid) data problem caused by the data grouping strategy. It also introduces a new state transition ring mechanism into the statistical query (SQ) learning framework to update the local model of each virtual client quickly. Using real-world IoT traffic data, we showcase the benefit of ViFLa regarding its efficiency and completeness for model updates in the context of IoT traffic anomaly detection.

Nowadays, more and more content providers (CPs) use multiple content delivery networks (CDNs) to deliver their content (a.k.a. content multihoming). Since the decisions on which CDN to use are made by the CP or by a CDN broker based on their local view of network conditions, content multihoming still has much room to improve for a better content delivery performance. In addition, content multihoming may negatively impact CDN vendors since in the price competition they are enforced to lower content delivery price to attract CPs to use their CDNs. To build a better CDN ecosystem, multi-CDN federation has been proposed to interconnect standalone CDNs. The real-world implementation of CDN interconnection (CDNI), however, poses significant technical obstacles not easy to solve in the short term. In order to improve the content delivery performance under current multi-CDN strategies, in this paper, we propose a feasible and efficient solution to multi-CDN, termed as CDN semi-federation, which can better schedule and utilize the resources from multiple CDNs without requiring full CDNI. The benefit of our solution comes from an effective optimization algorithm which reshapes the patterns of traffic from multiple CPs delivered over multipe CDN Points of Presence (PoPs). Experiments across North American and European ISP PoP networks demonstrate that, compared with current multi-CDN solutions, CDN semi-federation can reduce the content delivery latency by around 20% during peak traffic hours.