Interpretable Machine Learning for COVID-19: An Empirical Study on Severity Prediction Task.

The black-box nature of machine learning models hinders the deployment of some high-accuracy medical diagnosis algorithms. It is risky to put one's life in the hands of models that medical researchers do not fully understand or trust. However, through model interpretation, black-box models can promptly reveal significant biomarkers that medical practitioners may have overlooked due to the surge of infected patients in the COVID-19 pandemic. This research leverages a database of 92 patients with confirmed SARS-CoV-2 laboratory tests between 18th January 2020 and 5th March 2020, in Zhuhai, China, to identify biomarkers indicative of infection severity prediction. Through the interpretation of four machine learning models, decision tree, random forests, gradient boosted trees, and neural networks using permutation feature importance, partial dependence plot, individual conditional expectation, accumulated local effects, local interpretable model-agnostic explanations, and Shapley additive explanation, we identify an increase in N-terminal pro-brain natriuretic peptide, C-reaction protein, and lactic dehydrogenase, a decrease in lymphocyte is associated with severe infection and an increased risk of death, which is consistent with recent medical research on COVID-19 and other research using dedicated models. We further validate our methods on a large open dataset with 5644 confirmed patients from the Hospital Israelita Albert Einstein, at São Paulo, Brazil from Kaggle, and unveil leukocytes, eosinophils, and platelets as three indicative biomarkers for COVID-19.

DFML: Dynamic Federated Meta-Learning for Rare Disease Prediction.

Millions of patients suffer from rare diseases around the world. However, the samples of rare diseases are much smaller than those of common diseases. Hospitals are usually reluctant to share patient information for data fusion due to the sensitivity of medical data. These challenges make it difficult for traditional AI models to extract rare disease features for disease prediction. In this paper, we propose a Dynamic Federated Meta-Learning (DFML) approach to improve rare disease prediction. We design an Inaccuracy-Focused Meta-Learning (IFML) approach that dynamically adjusts the attention to different tasks according to the accuracy of base learners. Additionally, a dynamic weight-based fusion strategy is proposed to further improve federated learning, which dynamically selects clients based on the accuracy of each local model. Experiments on two public datasets show that our approach outperforms the original federated meta-learning algorithm in accuracy and speed with as few as five shots. The average prediction accuracy of the proposed model is improved by 13.28% compared with each hospital's local model.

Flexico: An efficient dual-mode consensus protocol for blockchain networks.

Blockchain is a Byzantine fault tolerant (BFT) system wherein decentralized nodes execute consensus protocols to drive the agreement process on new blocks added to a distributed ledger. Generally, two-round communications among [Formula: see text] nodes are required to tolerate up to [Formula: see text] faults in BFT-based consensus networks. This communication pattern corresponds to the worse-case scenario of consensus achievement, even under asynchronous network conditions. Nevertheless, it is not uncommon for a network to operate under better conditions, where a consensus can be reached with a lower communication cost. Hence, with the addition of a faster optimistic path toward an agreement, the idea of dual-mode consensus has been proposed as a promising approach to enhance the performance of asynchronous BFT protocols. However, this opportunity is not completely exploited by existing dual-mode protocols as the fast path can be followed only in a nonfaulty and synchronous network. This article presents a novel dual-mode protocol consisting of fast and backup subprotocols. To create different consensus committees for fast and backup-mode operations, the network contains both active and passive nodes. A consensus can be expedited through a fast-mode operation when majority of the active nodes can communicate synchronously. Under non-ideal conditions, the backup protocol takes over the agreement process from its fast-mode counterpart without starting over the suspended round. The safety and liveness of the proposed protocol are guaranteed with lower communication costs, which balance the trade-off between protocol efficiency and availability.

Predictive topology refinements in distributed stream processing system.

Cloud computing has evolved the big data technologies to a consolidated paradigm with SPaaS (Streaming processing-as-a-service). With a number of enterprises offering cloud-based solutions to end-users and other small enterprises, there has been a boom in the volume of data, creating interest of both industry and academia in big data analytics, streaming applications, and social networking applications. With the companies shifting to cloud-based solutions as a service paradigm, the competition grows in the market. Good quality of service (QoS) is a must for the enterprises, as they strive to survive in a competitive environment. However, achieving reasonable QoS goals to meet SLA agreement cost-effectively is challenging due to variation in workload over time. This problem can be solved if the system has the ability to predict the workload for the near future. In this paper, we present a novel topology-refining scheme based on a workload prediction mechanism. Predictions are made through a model based on a combination of SVR, autoregressive, and moving average model with a feedback mechanism. Our streaming system is designed to increase the overall performance by making the topology refining robust to the incoming workload on the fly, while still being able to achieve QoS goals of SLA constraints. Apache Flink distributed processing engine is used as a testbed in the paper. The result shows that the prediction scheme works well for both workloads, i.e., synthetic as well as real traces of data.

Cost-aware orchestration of applications over heterogeneous clouds.

The orchestration of applications and their components over heterogeneous clouds is recognized as being critical in solving the problem of vendor lock-in with regards to distributed and cloud computing. There have been recent strides made in the area of cloud application orchestration with emergence of the TOSCA standard being a definitive one. Although orchestration by itself provides a considerable amount of benefit to consumers of cloud computing services, it remains impractical without a compelling reason to ensure its utilization by cloud computing consumers. If there is no measurable benefit in using orchestration, then it is likely that clients may opt out of using it altogether. In this paper, we present an approach to cloud orchestration that aims to combine an orchestration model with a cost and policy model in order to allow for cost-aware application orchestration across heterogeneous clouds. Our approach takes into consideration the operating cost of the application on each provider, while performing a forward projection of the operating cost over a period of time to ensure that cost constraints remain unviolated. This allows us to leverage the existing state of the art with regards to orchestration and model-driven approaches as well as tie it to the operations of cloud clients in order to improve utility. Through this study, we were able to show that our approach was capable of providing not only scaling features but also orchestration features of application components distributed across heterogeneous cloud platforms.

From Smart Homes to Smart-Ready Homes and Communities.

BACKGROUND: People have various and changing needs as they age, and the number of people living with some form of dementia is steadily increasing. Smart homes have a unique potential to provide assisted living but are often designed rigidly with a specific and fixed problem in mind. OBJECTIVES: To make smart-ready homes and communities that can be adaptively and easily updated over time to support varying user needs and to deliver the needed assistance, empowerment, and living independence. METHOD: The design and deployment of programmable assistive environment for older adults. RESULTS: The use of platform technology (a special form of what is known today as the Internet of Things or IoT) has enabled the decoupling of goal setting and application development from sensing and assistive technology deployment and insertion in the assistive environment. Personalising a smart home or changing its applications and its interfaces dynamically as the user needs change was possible and has been demonstrated successfully in one house - the Gator Tech Smart House. Scaling up the platform technology approach to a planned living community is underway at one of UK's National Health Services (NHS) Healthy New Town projects. CONCLUSIONS: There is a great need to integrate technology with living spaces to provide assistance and independent living, but to smarten these spaces for lifelong living, the technology and the smart home applications must be flexible, adaptive, and changeable over time. However, people do not just live at home, they live in communities. Looking at the big picture (communities), as well as the small (homes), we consider how to progress beyond smart-ready homes towards smart-ready communities.

VM Capacity-Aware Scheduling within Budget Constraints in IaaS Clouds.

Recently, cloud computing has drawn significant attention from both industry and academia, bringing unprecedented changes to computing and information technology. The infrastructure-as-a-Service (IaaS) model offers new abilities such as the elastic provisioning and relinquishing of computing resources in response to workload fluctuations. However, because the demand for resources dynamically changes over time, the provisioning of resources in a way that a given budget is efficiently utilized while maintaining a sufficing performance remains a key challenge. This paper addresses the problem of task scheduling and resource provisioning for a set of tasks running on IaaS clouds; it presents novel provisioning and scheduling algorithms capable of executing tasks within a given budget, while minimizing the slowdown due to the budget constraint. Our simulation study demonstrates a substantial reduction up to 70% in the overall task slowdown rate by the proposed algorithms.

Participatory and persuasive telehealth.

Technological advances in telehealth systems are primarily focused on sensing and monitoring. However, these systems are limited in that they only rely on sensors and medical devices to obtain vital signs. New research and development are urgently needed to offer more effective and meaningful interactions between patients, medical professionals and other individuals around the patients. Social networking with Web 2.0 technologies and methods can meet these demands, and help to develop a more complete view of the patient. Also many people, including the elderly, may be resistant to change, which can reduce the efficacy of telehealth systems. Persuasive technology and mechanisms are urgently needed to counter this resistance and promote healthy lifestyles. In this paper, we propose the participatory and persuasive telehealth system as a solution for these two limitations. By integrating connected health solutions with social networking and adding persuasive influence, we increase the chances for effective interventions and behavior alterations.