Federated learning (FL) is a distributed framework for machine learning that enables collaborative training of a shared model across data silos while preserving data privacy. However, the FL aggregation server faces a challenge in waiting for a large volume of model parameters from selected nodes before generating a global model, which leads to inefficient communication and aggregation. Although transmitting only the signs of stochastic gradient descent (SignSGD) reduces the transmission load, it decreases model accuracy, and the time waiting for local model collection remains substantial. Moreover, the security of FL is severely compromised by prevalent poisoning, backdoor, and DDoS attacks, causing ineffective and inaccurate model training. To overcome these challenges, this paper proposes aBlockchain-basedAsynchronousSignSGD (BASS) architecture for efficient and secure federated learning. By integrating a blockchain-based semi-asynchronous aggregation scheme with sign-based gradient compression, BASS considerably improves communication and aggregation efficiency, while providing resistance against attacks. Besides, a novel node-summarized sign aggregation algorithm is developed for the blockchain leaders to ensure the convergence and accuracy of the global model. An open-source prototype is developed, on top of which extensive experiments are conducted. The results validate the superiority of BASS in terms of efficiency, model accuracy, and security.

Federated learning (FL) enables collaborative training of a shared model on edge devices while maintaining data privacy. FL is effective when dealing with independent and identically distributed (iid) datasets, but struggles with non-iid datasets. Various personalized approaches have been proposed, but such approaches fail to handle underlying shifts in data distribution, such as data distribution skew commonly observed in real-world scenarios (e.g., driver behavior in smart transportation systems changing across time and location). Additionally, trust concerns among unacquainted devices and security concerns with the centralized aggregator pose additional challenges. To address these challenges, this paper presents a dynamically optimized personal deep learning scheme based on blockchain and federated learning. Specifically, the innovative smart contract implemented in the blockchain allows distributed edge devices to reach a consensus on the optimal weights of personalized models. Experimental evaluations using multiple models and real-world datasets demonstrate that the proposed scheme achieves higher accuracy and faster convergence compared to traditional federated and personalized learning approaches.

Device-to-Device (D2D) communication has attracted extensive researches because of its ability to reduce latency and improve the spectrum resource utilization. This paper studies a D2D Mobile Edge Computing (MEC) system which contains multiple busy smart devices (SDs) and multiple idle smart devices. To minimize the total energy consumption of the MEC system and satisfy the latency constraints of SDs, the computation intensive task of each busy SD can be partially offloaded to one or more idle SDs as helpers. Therefore, a joint optimization problem of helpers selection and communication and computation resources allocation is proposed. The problem is formulated as an integer-mixed non-convex optimization problem which is a NP-hard problem. We thus propose a two-phase iterative approach by jointly optimizing helpers selection and communication and computation resources allocation. In the first phase, we obtain the suboptimal helpers selection policy with convex optimization techniques and block coordinate descent method. In the second phase, the resource allocation strategy is achieved by applying block coordinate descent after obtaining the suboptimal helpers selection policy. The simulation results demonstrate that not only the proposed algorithm achieves fast convergence in both phases, but also the overall energy consumption is less than other benchmarks.

Agent based models (ABMs) simulate actions and interactions of autonomous agents/groups and their effect on systems as a whole, accounting for learning without assuming perfect rationality or complete knowledge. ABMs are an increasingly popular approach to studying complex, spatially distributed socio-environmental systems, but have still to become an established approach in the sense of being one that is expected by those wanting to explore scenarios in such systems. Partly, this is an issue of awareness – ABM is still new enough that many people have not heard of it; partly, it is an issue of confidence – ABM has more to do to prove itself if it is to become a preferred method. This paper will identify advances in the craft and deployment of ABM needed if ABM is to become an accepted part of mainstream science for policy or stakeholders. The conduct of ABM has, over the last decade, seen a transition from using abstracted representations of systems (supporting theory-led thought experiments) to more accessible representations derived empirically (to deliver more applied analysis). This has enhanced the perception of potential users of ABM outputs that the latter are salient and credible. Empirical ABM is not, however, a panacea, as it demands more computing and data resources, limiting applications to domains where data exist along with suitable environmental models where these are required. Further, empirical ABM is still facing serious questions of validation and the ontology used to describe the system in the first place. Using Geoffrey A. Moore’s Crossing the Chasm as a lens, we argue that the way ahead for ABM lies in identifying the niches in which it can best demonstrate its advantages, working with collaborators to demonstrate that it can deliver on its promises. This leads us to identify several areas where work is needed.

This paper proposes a novel naïve Bayesian classifier in categorical uncertain data streams. Uncertainty in categorical data is usually represented by vector valued discrete pdf, which has to be carefully handled to guarantee the underlying performance in data mining applications. In this paper, we map the probabilistic attribute to deterministic points in the Euclidean space and design a distance based and a density based algorithms to measure the correlations between feature vectors and class labels. We also devise a new pre-binning approach to guarantee bounded computation and memory cost in uncertain data streams classification. Experimental results in real uncertain data streams prove that our density-based naive classifier is efficient, accurate, and robust to data uncertainty.