Distributed Learning Approach Research Articles

RTHop: Real-Time Hop-by-hop Mobile Network Routing by Decentralized Learning with Semantic Attention

Multi-access Edge Computing and ubiquitous smart devices help serve end-users efficiently by providing emerging edge-deployed services. On the other hand, more heavy and time-varying traffic loads are generated in mobile edge networks, so that an efficient traffic forwarding mechanism is highly required to handle the routing problem in complex and highly dynamic edge environments. Thus, Deep Reinforcement Learning (DRL) is introduced since it can work in a model-free approach. However, previous centralized DRL-based methods work in a turn-based way that mismatches the real-time property of routing. In this paper, we propose a real-time and distributed learning approach, RTHop, to adapt to the volatile environment and realize a hop-by-hop routing. The Multi-Agent Deep Reinforcement Learning (MADRL) and the Real-Time Markov Decision Process (RTMDP) are used to alleviate network congestion and maximize the utilization of network resources. By joining with the self-attention mechanism, RTHop obtains the semantics from elements of the network state to help agents learn the importance of each element on routing. Experiment results show that RTHop not only overcomes the weakness of conventional turn-based DRL methods but also achieves the increase of delivered packet ratios and effective throughput compared with other routing methods.

FogFL: Fog-Assisted Federated Learning for Resource-Constrained IoT Devices

In this article, we propose a fog-enabled federated learning framework—FogFL—to facilitate distributed learning for delay-sensitive applications in resource-constrained IoT environments. While federated learning (FL) is a popular distributed learning approach, it suffers from communication overheads and high computational requirements. Moreover, global aggregation in FL relies on a centralized server, prone to malicious attacks, resulting in inefficient training models. We address these issues by introducing geospatially placed fog nodes into the FL framework as local aggregators. These fog nodes are responsible for defined demographics, which help share location-based information for applications with similar environments. Furthermore, we formulate a greedy heuristic approach for selecting an optimal fog node for assuming a global aggregator’s role at each round of communication between the edge and cloud, thereby reducing the dependence on the execution at the centralized server. Fog nodes in the FogFL framework reduce communication latency and energy consumption of resource-constrained edge devices without affecting the global model’s convergence rate, thereby increasing the system’s reliability. Extensive deployment and experimental results corroborate that, in addition to a decrease in global aggregation rounds, FogFL reduces energy consumption and communication latency by 92% and 85%, respectively, as compared to state of the art.

Random walk based distributed representation learning and prediction on Social Networking Services

Social Networking Services (SNSs) provide online platforms for users with two kinds of behavior: user-user social behavior (e.g., following a user, making friends with others) and user-item consumption behavior (e.g., rating, showing likeness, clicking, giving thumbs up to items). With the increasing popularity of SNSs and demand for SNS features, predicting potential social links and recommending preferable items to users have become two hot research lines. However, previous works either modeled just one of these two kinds of behaviors in isolation or only considered the observed user behavior data. In fact, social scientists have long recognized that the user-user and user-item behaviors have a mutual reinforcement effect. On the one hand, the two behaviors have correlations, and they can influence each other. On the other hand, due to the sparsity of the observed user behavior data, the user behavior prediction performance is far from satisfactory, although, using two types of behavioral data at the same time can mitigate the sparsity problem. These two problems remains open: how to better model the correlation of user-user social and user-item consumption activities and how to mitigate the data sparsity issue. In this paper, we propose a random walk based distributed representation learning model to jointly predict these behaviors on SNSs. Specifically, we first construct a joint behavior graph that combines the two behaviors, with the edges denoting the sparse observed user behavior data. Then, we adopt a random walk to capture higher-order relationships between users and items. After that, we utilize a distributed learning approach to embed both users and items into a latent space. In this way, the behavior prediction tasks are transformed into similarity calculations in the latent space. Finally, extensive experimental results using two real-world datasets demonstrate the effectiveness of our proposed approach on the two behavior prediction tasks.

An Adaptive Wireless Virtual Reality Framework in Future Wireless Networks: A Distributed Learning Approach

Wireless virtual reality (VR) is predicted to become a killer application in 5G and beyond, which provides an immersive experience and revolutionizes the way people communicate. It is well-known that rendering is the key performance bottleneck in wireless VR systems, especially for VR games. However, real-time rendering and data correlation are ignored by most researchers. In this paper, we propose an adaptive VR framework that enables high-quality wireless VR in future mmWave-enabled wireless networks with mobile edge computing (MEC), where real-time VR rendering tasks can be offloaded to MEC servers adaptively and the caching capability of MEC servers enables further performance improvement. First, we formulate the addressed problem to maximize the quality of experience (QoE) of the users, where association, adaptive offloading mode selection, and caching policy are jointly optimized. Considering the high complexity of the addressed problem, we then propose a distributed learning approach consisting of an offline training phase and an online running phase, which maintains scalability and adaptation capability. The offline phase is based on deep reinforcement learning (DRL) while the latter utilizes game theory. At last, simulation results show the superiority of the proposed algorithm over the other baseline algorithms in terms of QoE utility values, latency, and convergence time.

Distributed Online Learning of Cooperative Caching in Edge Cloud

Cooperative caching can unify storage across edge clouds and provide efficient delivery of popular contents under effective content placement. However, the placement and delivery are non-trivial in cooperative caching due to the decentralized property of edge clouds, as well as the temporal and spatial correlation of the placement. We propose a new distributed online learning approach to jointly optimize content placement and delivery without the a-priori knowledge on file popularity and link availability. Content placement and delivery can be asymptotically optimized in real-time by running distributed online learning at individual edge servers by exploiting stochastic gradient descent (SGD). The proposed approach can allow operations at different timescales by integrating mini-batch learning for farsighted content placement. The optimality loss, stemming from the different timescales, can asymptotically reduce, as the SGD stepsize declines. Simulations confirm that the proposed approach outperforms existing techniques in terms of cache hit ratio and cost effectiveness. Insights are shed on the optimal placement of popular contents.

Gaussian processes modifier adaptation with uncertain inputs for distributed learning and optimization of wind farms

A modifier adaptation scheme based on Gaussian processes is presented to optimize the control inputs of a wind farm. Often an approximate model of the wind farm is available, however due to the high complexity of the process plant-model mismatch is prevalent. For example the mechanics of wakes is not well-understood, which may have a profound impact on the power production of wind farms. Therefore, Gaussian process (GP) regression is exploited to account for this deviation. A distributed learning approach is used to learn the plant-model mismatch of each individual turbine considering explicitly the uncertainty of the uncontrolled inputs, like the wind direction. Afterwards, a distributed optimization scheme using alternating direction method of multipliers is applied to iteratively attain the wind farm optimum despite the presence of plant-model mismatch.

External Validation of Radiation-Induced Dyspnea Models on Esophageal Cancer Radiotherapy Patients.

Purpose: Radiation-induced lung disease (RILD), defined as dyspnea in this study, is a risk for patients receiving high-dose thoracic irradiation. This study is a TRIPOD (Transparent Reporting of A Multivariable Prediction Model for Individual Prognosis or Diagnosis) Type 4 validation of previously-published dyspnea models via secondary analysis of esophageal cancer SCOPE1 trial data. We quantify the predictive performance of these two models for predicting the maximal dyspnea grade ≥ 2 within 6 months after the end of high-dose chemo-radiotherapy for primary esophageal cancer.Materials and methods: We tested the performance of two previously published dyspnea risk models using baseline, treatment and follow-up data on 258 esophageal cancer patients in the UK enrolled into the SCOPE1 multi-center trial. The tested models were developed from lung cancer patients treated at MAASTRO Clinic (The Netherlands) from the period 2002 to 2011. The adverse event of interest was dyspnea ≥ Grade 2 (CTCAE v3) within 6 months after the end of radiotherapy. As some variables were missing randomly and cannot be imputed, 212 patients in the SCOPE1 were used for validation of model 1 and 255 patients were used for validation of model 2. The model parameter Forced Expiratory Volume in 1 s (FEV1), as a predictor to both validated models, was imputed using the WHO performance status. External validation was performed using an automated, decentralized approach, without exchange of individual patient data.Results: Out of 258 patients with esophageal cancer in SCOPE1 trial data, 38 patients (14.7%) developed radiation-induced dyspnea (≥ Grade 2) within 6 months after chemo-radiotherapy. The discrimination performance of the models in esophageal cancer patients treated with high-dose external beam radiotherapy was moderate, area under curve (AUC) of 0.68 (95% CI 0.55–0.76) and 0.70 (95% CI 0.58–0.77), respectively. The curves and AUCs derived by distributed learning were identical to the results from validation on a local host.Conclusion: We have externally validated previously published dyspnea models using an esophageal cancer dataset. FEV1 that is not routinely measured for esophageal cancer was imputed using WHO performance status. Prediction performance was not statistically different from previous training and validation sets. Risk estimates were dominated by WHO score in Model 1 and baseline dyspnea in Model 2. The distributed learning approach gave the same answer as local processing, and could be performed without accessing a validation site's individual patients-level data.

Demonstration of distributed collaborative learning with end-to-end QoT estimation in multi-domain elastic optical networks.

This paper proposes a distributed collaborative learning approach for cognitive and autonomous multi-domain elastic optical networking (EON). The proposed approach exploits a knowledge-defined networking framework which leverages a broker plane to coordinate the operations of multiple EON domains and applies machine learning (ML) to support autonomous and cognitive inter-domain service provisioning. By employing multiple distributed ML blocks learning domain-level features and working with broker plane aggregation ML blocks (through the chain rule-based training), the proposed approach enables to develop cognitive networking applications that can fully exploit the multi-domain EON states while obviating the need for the raw and confidential intra-domain data. In particular, we investigate end-to-end quality-of-transmission estimation application using the distributed learning approach and propose three estimator designs incorporating the concepts of multi-task learning (MTL) and transfer learning (TL). Evaluations with experimental data demonstrate that the proposed designs can achieve estimation accuracies very close to (with differences less than 0.5%) or even higher than (with MTL/TL) those of the baseline models assuming full domain visibility.

Motivating Workers in Federated Learning: A Stackelberg Game Perspective

Due to the large size of the training data, distributed learning approaches such as federated learning have gained attention recently. However, the convergence rate of distributed learning suffers from heterogeneous worker performance. In this letter, we consider an incentive mechanism for workers to mitigate the delays in completion of each batch. We analytically obtained equilibrium solution of a Stackelberg game. Our numerical results indicate that with a limited budget, the model owner should judiciously decide on the number of workers due to trade off between the diversity provided by the number of workers and the latency of completing the training.

Distributed learning with indefinite kernels

We investigate the distributed learning with coefficient-based regularization scheme under the framework of kernel regression methods. Compared with the classical kernel ridge regression (KRR), the algorithm under consideration does not require the kernel function to be positive semi-definite and hence provides a simple paradigm for designing indefinite kernel methods. The distributed learning approach partitions a massive data set into several disjoint data subsets, and then produces a global estimator by taking an average of the local estimator on each data subset. Easy exercisable partitions and performing algorithm on each subset in parallel lead to a substantial reduction in computation time versus the standard approach of performing the original algorithm on the entire samples. We establish the first mini-max optimal rates of convergence for distributed coefficient-based regularization scheme with indefinite kernels. We thus demonstrate that compared with distributed KRR, the concerned algorithm is more flexible and effective in regression problem for large-scale data sets.

Privacy-Preserving Database Assisted Spectrum Access for Industrial Internet of Things: A Distributed Learning Approach

Industrial Internet of Things (IIoT) has been shown to be of great value to the deployment of smart industrial environment. With the immense growth of Internet of Things (IoT) devices, dynamic spectrum sharing is introduced, envisaged as a promising solution to the spectrum shortage in IIoT. Meanwhile, cyber–physical safety issue remains to be a great concern for the reliable operation of IIoT system. In this article, we consider the dynamic spectrum access in IIoT under a received signal strength-based adversarial localization attack. We employ a practical and effective power perturbation approach to mitigate the localization threat on the IoT devices and cast the privacy-preserving spectrum sharing problem as a stochastic channel selection game. To address the randomness induced by the power perturbation approach, we develop a two-timescale distributed learning algorithm that converges almost surely to the set of correlated equilibria of the game. The numerical results show the convergence of the algorithm and corroborate that the design of two-timescale learning process effectively alleviates the network throughput degradation brought by the power perturbation procedure.

Head pose estimation using improved label distribution learning with fewer annotations

Head pose estimation in unconstrained environment remains a challenging task due to background clutter, illumination changes, and appearance variabilities. Multivariate label distribution has been successfully applied to head pose estimation. However, it is not applicable to unconstrained environments where assigning reasonable label distributions for images is difficult, and its performance significantly degrades when accurate grid information is unavailable (e.g., only yaw angles are known). To alleviate these problems, we propose an improved label distribution learning approach with fewer annotations. A data-driven weak learning strategy is first developed to construct label distributions to alleviate the problem of unreasonable label distributions. Regularization terms (e.g., L1,2 norm) are then introduced into the loss function induced by weighted Jeffreys divergence to avoid over-fitting. To further ameliorate the performance, positive correlation and negative competition are also introduced into the loss function to fine-tune the parameters of the corresponding model. Extensive experiments have been conducted on public databases: LFW and Pointing04. The proposed method achieves comparable performance over the state-of-art and possesses good generalization ability, but uses only fewer annotations, which suggests that it has strong potential for head pose estimation in unconstrained environments where sufficient annotations are routinely unavailable.

Comprehensive image quality assessment via predicting the distribution of opinion score

Image quality assessment is a challenge problem in image processing area. Previous works usually predict the mean opinion score (MOS) to evaluate image quality. However, it is found that the distribution of opinion scores provides richer and more precise semantics information. Therefore, in this work, we focus on the distribution of opinion scores (DOS) and aims to comprehensively evaluate image quality via automatically predicting DOS. Specifically, we first extract image features via convolutional neural network and then adopt the label distribution support vector regressor (LDSVR) algorithm to predict score distribution. To the best of our knowledge, we are the first to introduce label distribution learning approach for image quality assessment. Extensive experiments have been carried out and validate that the proposed algorithm can well predict the DOS and provide a comprehensive assessment to image quality.

Multi-task distribution learning approach to anomaly detection of operational states of wind turbines

The detection of abnormal operation modes is of fundamental importance for both operational management and predictive maintenance of wind turbines. Anomaly detection approaches in this context should consider the additional information content that probabilistic models can provide. Instead of binary anomaly classification, the probabilistic information is necessary for proper decision making and risk assessment. Common models, such as quantile and distribution regression can provide probabilistic information. While they are appropriate in predicting the cumulative distribution function, they struggle to accurately describe the probability of an event to occur. In this article we present a new, multi-task learning based approach for a continuous distribution regression with deep neural networks. Using real-world data from an offshore wind turbine, we show that with this model we can better reflect the probability of observed events than with conventional methods. While the predicted cumulative distribution function has a similar quality and no significant differences are visible in the continuous ranked probability score, the probability density function will be substantially smoother. This is also reflected in a significantly lower ignorance score.

Dynamic Event Detection Using a Distributed Feature Selection Based Machine Learning Approach in a Self-Healing Microgrid

A self-healing function is an attractive feature in any modern microgrid. Once a fault occurs, it is imperative for a grid to monitor its status, take action based on the level of severity, and after the contingency has been cleared, restore the system. With an increasing number of microgrids and distributed generation stations, deploying a centralized control is no longer a cost-effective option, and therefore distributed control is a likely solution. In an interconnected network, it is important to detect the underlying events taking place in each of the distributed stations, otherwise operational decisions become noncoherent. This paper proposes a novel, feature selection-based distributed machine learning approach to detect the dynamic signatures of different power system events. The purpose is to facilitate a postfault decision-making process in order to restore a stand-alone microgrid without the intervention of a central station. The proposed method detects meaningful features from the generator data and then applies a multiclass classification algorithm to the feature data. Each class represents one dynamic event taking place. The methodology is demonstrated in an interconnected two-area-based microgrid with multiple types of energy generation schemes.

On distributed non‐linear active noise control using diffusion collaborative learning strategy

Active noise control in a non-linear spatial region is a challenging problem, especially for low-frequency noise control applications. The investigation of the existing literature reveals that this problem is tackled by systems with multiple sensors and loudspeakers with a centralised processor. However, the use of centralised techniques is bulky, computationally expensive and lacks scalability. Therefore, the authors propose a distributed learning approach for noise cancellation using a diffusion collaborative strategy. The proposed Legendre-functional link network diffusion filtered s least mean square (FsLMS) algorithm is compared with the standard multi-channel FsLMS algorithm. For different non-linear scenarios, the performance of the proposed method is evaluated in terms of noise reduction performance and computational complexity. It is demonstrated that the proposed method offers significant improvement in noise mitigation performance and computational load as compared with its centralised multi-channel counterpart.

Examining Science Teachers’ Argumentation in a Teacher Workshop on Earthquake Engineering

The purpose of this study was to examine changes in the quality of science teachers’ argumentation as a result of their engagement in a teacher workshop on earthquake engineering emphasizing distributed learning approaches, which included concept mapping, collaborative game playing, and group lesson planning. The participants were ten high school science teachers from US high schools who elected to attend the workshop. To begin and end the teacher workshop, teachers in small groups engaged in concept mapping exercises with other teachers. Researchers audio-recorded individual teachers’ argumentative statements about the inclusion of earthquake engineering concepts in their concept maps, which were then analyzed to reveal the quality of teachers’ argumentation. Toulmin’s argumentation model formed the framework for designing a classification schema to analyze the quality of participants’ argumentative statements. While the analysis of differences in pre- and post-workshop concept mapping exercises revealed that the number of argumentative statements did not change significantly, the quality of participants’ argumentation did increase significantly. As these differences occurred concurrently with distributed learning approaches used throughout the workshop, these results provide evidence to support distributed learning approaches in professional development workshop activities to increase the quality of science teachers’ argumentation. Additionally, these results support the use of concept mapping as a cognitive scaffold to organize participants’ knowledge, facilitate the presentation of argumentation, and as a research tool for providing evidence of teachers’ argumentation skills.

A communication efficient distributed learning framework for smart environments

Due to the pervasive diffusion of personal mobile and IoT devices, many “smart environments” (e.g., smart cities and smart factories) will be, among others, generators of huge amounts of data. To provide value-add services in these environments, data will have to be analysed to extract knowledge. Currently, this is typically achieved through centralised cloud-based data analytics services. However, according to many studies, this approach may present significant issues from the standpoint of data ownership, and even wireless network capacity. One possibility to cope with these shortcomings is to move data analytics closer to where data is generated. In this paper we tackle this issue by proposing and analysing a distributed learning framework, whereby data analytics are performed at the edge of the network, i.e., on locations very close to where data is generated. Specifically, in our framework, partial data analytics are performed directly on the nodes that generate the data, or on nodes close by (e.g., some of the data generators can take this role on behalf of subsets of other nodes nearby). Then, nodes exchange partial models and refine them accordingly. Our framework is general enough to host different analytics services. In the specific case analysed in the paper we focus on a learning task, considering two distributed learning algorithms. Using an activity recognition and a pattern recognition task, both on reference datasets, we compare the two learning algorithms between each other and with a central cloud solution (i.e., one that has access to the complete datasets). Our results show that using distributed machine learning techniques, it is possible to drastically reduce the network overhead, while obtaining performance comparable to the cloud solution in terms of learning accuracy. The analysis also shows when each distributed learning approach is preferable, based on the specific distribution of the data on the nodes.

Wildfire: Approximate synchronization of parameters in distributed deep learning

In distributed deep learning approaches, contributions to changes in the parameter values from multiple learners are gathered at periodic intervals and collectively used to update weights associated with the learning network. Gathering these contributions at a centralized location, as in common synchronous parameter server models, causes a bottleneck in two ways. First, the parameter server needs to wait until gradients from all learners have been received, and second, the traffic pattern of the gradients between the learners and the parameter server causes an imbalance in bandwidth utilization in most common networks. In this paper, we introduce a scheme called Wildfire, which communicates weights among subsets of parallel learners, each of which updates its model using only the information received from other learners in the subset. Different subsets of learners communicate at different times, allowing the learning to diffuse through the system. Wildfire reduces the communication overhead in deep learning by allowing learners to communicate directly among themselves rather than through a parameter server, and by limiting the time that learners need to wait before updating their models. We demonstrate the effectiveness of Wildfire on common deep learning benchmarks, using the IBM Rudra deep learning framework.

Learning from decades of online distance education: MOOCs and the Community of Inquiry framework

Despite their growing popularity, there are many contradictory arguments between supporters and detractors of MOOCs. Nevertheless, the advent of mass-scale online courses is increasingly credited to have the potential to reshape higher education significantly over time, and recent research analyses how and in which ways such a potential can be leveraged. Aim of this conceptual study is to incorporate the Community of Inquiry (CoI) framework into learning design practice to overcome current MOOCs pedagogical limitations. In order to be applied to a large numbers of participants, the three presences of the CoI framework (social, cognitive and metacognitive, and teaching) need to be adjusted and combined with the distributed learning approach. In this way, fostering distributed learning among participants would lead learners to take responsibility for their learning experience through the exploitation of roles and tasks traditionally assigned to distinct and separate roles. These adjustments could address issues of participation, motivation, and enhance successful learning experience.