Logo: to the web site of Uppsala University

Federated learning is a distributed and privacy-preserving approach to train a statistical model collaboratively from decentralized data held by different parties. However, when the datasets are not independent and identically distributed, models trained by naive federated algorithms may be biased towards certain participants, and model performance across participants is non-uniform. This is known as the fairness problem in federated learning. In this paper, we formulate fairness-controlled federated learning as a dynamical multi-objective optimization problem to ensure the fairness and convergence with theoretical guarantee. To solve the problem efficiently, we study the convergence and bias of Adam as the server optimizer in federated learning, and propose Adaptive Federated Adam ( AdaFedAdam ) to accelerate fair federated learning with alleviated bias. We validated the effectiveness, Pareto optimality and robustness of AdaFedAdam with numerical experiments and show that AdaFedAdam outperforms existing algorithms, providing better convergence and fairness properties of the federated scheme.

In many areas of data-driven science, large datasets are generated where the individual data objects are images, matrices, or otherwise have a clear structure. However, these objects can be information-sparse, and a challenge is to efficiently find and work with the most interesting data as early as possible in an analysis pipeline. We have recently proposed a new model for big data management where the internal structure and information of the data are associated with each data object (as opposed to simple metadata). There is then an opportunity for comprehensive data management solutions to account for data-specific internal structure as well as access patterns. In this article, we explore this idea together with our recently proposed hierarchical storage management framework that uses reinforcement learning (RL) for autonomous and dynamic data placement in different tiers in a storage hierarchy. Our case-study is based on four scientific datasets: Protein translocation microscopy images, Airfoil angle of attack meshes, 1000 Genomes sequences, and Phenotypic screening images. The presented results highlight that our framework is optimal and can quickly adapt to new data access requirements. It overall reduces the data processing time, and the proposed autonomous data placement is superior compared to any static or semi-static data placement policies.

Intrusion detection systems (IDSs) are widely used to identify anomalies in computer networks and raise alarms on intrusive behaviors. ML-based IDSs generally take network traces or host logs as input to extract patterns from individual samples, whereas the inter-dependencies of network are often not captured and learned, which may result in large amounts of uncertain predictions, false positives, and false negatives. To tackle the challenges in intrusion detection, we propose a graph neural network-based intrusion detection system (GNN-IDS), which is data-driven and machine learning-empowered. In our proposed GNN-IDS, the attack graph and real-time measurements that represent static and dynamic attributes of computer networks, respectively, are incorporated and associated to represent complex computer networks. Graph neural networks are employed as the inference engine for intrusion detection. By learning network connectivity, graph neural networks can quantify the importance of neighboring nodes and node features to make more reliable predictions. Furthermore, by incorporating an attack graph, GNN-IDS could not only detect anomalies but also identify the malicious actions causing the anomalies. The experimental results on a use case network with two synthetic datasets (one generated from public IDS data) show that the proposed GNN-IDS achieves good performance. The results are analyzed from the aspects of uncertainty, explainability, and robustness.

Federated learning (FL) enables edge nodes to collaboratively contribute to constructing a global model without sharing their data. This is accomplished by devices computing local, private model updates that are then aggregated by a server. However, computational resource constraints and network communication can become a severe bottleneck for larger model sizes typical for deep learning (DL) applications. Edge nodes tend to have limited hardware resources (RAM, CPU), and the network bandwidth and reliability at the edge is a concern for scaling federated fleet applications. In this paper, we propose and evaluate a FL strategy inspired by transfer learning in order to reduce resource utilization on devices, as well as the load on the server and network in each global training round. For each local model update, we randomly select layers to train, freezing the remaining part of the model. In doing so, we can reduce both server load and communication costs per round by excluding all untrained layer weights from being transferred to the server. The goal of this study is to empirically explore the potential trade-off between resource utilization on devices and global model convergence under the proposed strategy. We implement the approach using the FL framework FEDn. A number of experiments were carried out over different datasets (CIFAR-10, CASA, and IMDB), performing different tasks using different DL model architectures. Our results show that training the model partially can accelerate the training process, efficiently utilizes resources on-device, and reduce the data transmission by around 75% and 53% when we train 25%, and 50% of the model layers, respectively, without harming the resulting global model accuracy. Furthermore, our results demonstrate a negative correlation between the number of participating clients in the training process and the number of layers that need to be trained on each client's side. As the number of clients increases, there is a decrease in the required number of layers. This observation highlights the potential of the approach, particularly in cross-device use cases.

With the rapid development of big data and cloud computing, data management has become increasingly challenging. Over the years, a number of frameworks for data management have become available. Most of them are highly efficient, but ultimately create data silos. It becomes difficult to move and work coherently with data as new requirements emerge. A possible solution is to use an intelligent hierarchical (multi-tier) storage system (HSS). A HSS is a meta solution that consists of different storage frameworks organized as a jointly constructed storage pool. A built-in data migration policy that determines the optimal placement of the datasets in the hierarchy is essential. Placement decisions is a non-trivial task since it should be made according to the characteristics of the dataset, the tier status in a hierarchy, and access patterns. This paper presents an open-source hierarchical storage framework with a dynamic migration policy based on reinforcement learning (RL). We present a mathematical model, a software architecture, and implementations based on both simulations and a live cloud-based environment. We compare the proposed RL-based strategy to a baseline of three rule-based policies, showing that the RL-based policy achieves significantly higher efficiency and optimal data distribution in different scenarios.

With the rapid development of big data and cloud computing, data management has become increasingly challenging. A possible solution is to use an intelligent hierarchical (multi-tier) storage system (HSS). An HSS is a meta solution that consists of different storage frameworks organized as a jointly constructed storage pool. A built-in data migration policy that determines the optimal placement of the datasets in the hierarchy is essential. Placement decisions are a non-trivial task since they should be made according to the characteristics of the dataset, the tier status in a hierarchy, and access patterns. This paper presents an open-source hierarchical storage framework with a dynamic migration policy based on reinforcement learning (RL).

The information explosion drives enterprises and individuals to rent cloud computing infrastructure for their applications in the cloud. However, the agreements between cloud computing providers and clients are often inefficient. We propose an agent-based auto-negotiation system for resource scheduling using fuzzy logic. Our method completes a one-to-one auto-negotiation process and generates optimal offers for providers and clients. We compare the impact of different member functions, fuzzy rule sets, and negotiation scenarios on the offers to optimize the system. Our proposed method efficiently utilizes resources and offers interpretability, high flexibility, and customization. We successfully train machine learning models to replace the fuzzy negotiation system, improving processing speed. The article also highlights potential future improvements to the proposed system and machine learning models. All codes and data are available as an open source repository.

The proliferation of the Internet of Things (IoT) has incentivised extending cloud resources to the edge in what is deemed fog computing. The latter is manifesting as an ecosystem of connected clouds, geo-dispersed and of diverse capacities. In such ecosystem, workload allocation to fog services becomes a non-trivial challenge. Users' demand at the edge is diverse, which does not lend to straightforward resource planning. Conversely, running services at the edge may leverage proximity, but it comes at higher operational cost let alone increasing risk of resource straining. Consequently, there is a need for intelligent yet scalable allocation solutions that counter the adversity of demand, while efficiently distributing load between the edge and farther clouds. Machine learning is increasingly adopted in resource planning. This paper proposes a federated deep reinforcement learning system, based on deep Q-learning network (DQN), for workload distribution in a fog ecosystem. The proposed solution adapts a DQN to optimize local workload allocations, made by single gateways. Federated learning is incorporated to allow multiple gateways in a network to collaboratively build knowledge of users' demand. This is leveraged to establish consensus on the fraction of workload allocated to different fog nodes, using lower data supply and computation resources. System performance is evaluated using realistic demand from Google Cluster Workload Traces 2019. Evaluation results show over 50% reduction in failed allocations when spreading users over larger number of gateways, given fixed number of fog nodes. The results further illustrate the trade-offs between performance and cost under different conditions.

With Digital transformation, the diversity of services and infrastructure in backhaul fog network(s) is rising to unprecedented levels. This is causing a rising threat of a wider range of cyber attacks coupled with a growing integration of constrained range of infrastructure, particularly seen at the network edge. Deep reinforcement-based learning is an attractive approach to detecting attacks, as it allows less dependency on labeled data with better ability to classify different attacks. However, current approaches to learning are known to be computationally expensive (cost) and the learning experience can be negatively impacted by the presence of outliers and noise (quality). This work tackles both the cost and quality challenges with a novel service-based federated deep reinforcement learning solution, enabling anomaly detection and attack classification at a reduced data cost and with better quality. The federated settings in the proposed approach enable multiple edge units to create clusters that follow a bottom-up learning approach. The proposed solution adapts deep Q-learning Network (DQN) for service-tunable flow classification, and introduces a novel federated DQN (FDQN) for federated learning. Through such targeted training and validation, variation in data patterns and noise is reduced. This leads to improved performance per service with lower training cost. Performance and cost of the solution, along with sensitivity to exploration parameters are evaluated using an example publicly available dataset (UNSW-NB15). Evaluation results show the proposed solution to maintain detection accuracy with lower data supply, while improving the classification rate by a factor of ≈ 2.