AI in health and medicine

Nature Medicine, Journal Year: 2022, Volume and Issue: 28(1), P. 31 - 38

Published: Jan. 1, 2022

Language: Английский

---

Federated Learning for Internet of Things: A Comprehensive Survey

IEEE Communications Surveys & Tutorials, Journal Year: 2021, Volume and Issue: 23(3), P. 1622 - 1658

Published: Jan. 1, 2021

The Internet of Things (IoT) is penetrating many facets our daily life with the proliferation intelligent services and applications empowered by artificial intelligence (AI). Traditionally, AI techniques require centralized data collection processing that may not be feasible in realistic application scenarios due to high scalability modern IoT networks growing privacy concerns. Federated Learning (FL) has emerged as a distributed collaborative approach can enable applications, allowing for training at devices without need sharing. In this article, we provide comprehensive survey emerging FL networks, beginning from an introduction recent advances discussion their integration. Particularly, explore analyze potential enabling wide range services, including sharing, offloading caching, attack detection, localization, mobile crowdsensing, security. We then extensive use various key such smart healthcare, transportation, Unmanned Aerial Vehicles (UAVs), cities, industry. important lessons learned review FL-IoT are also highlighted. complete highlighting current challenges possible directions future research booming area.

Language: Английский

---

Model-Contrastive Federated Learning

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Journal Year: 2021, Volume and Issue: unknown

Published: June 1, 2021

Federated learning enables multiple parties to collaboratively train a machine model without communicating their local data. A key challenge in federated is handle the heterogeneity of data distribution across parties. Although many studies have been proposed address this challenge, we find that they fail achieve high performance image datasets with deep models. In paper, propose MOON: model-contrastive learning. MOON simple and effective framework. The idea utilize similarity between representations correct training individual parties, i.e., conducting contrastive model-level. Our extensive experiments show significantly outperforms other state-of-the-art algorithms on various classification tasks.

Language: Английский

---

Federated Learning on Non-IID Data Silos: An Experimental Study

2022 IEEE 38th International Conference on Data Engineering (ICDE), Journal Year: 2022, Volume and Issue: unknown, P. 965 - 978

Published: May 1, 2022

Due to the increasing privacy concerns and data regulations, training have been increasingly fragmented, forming distributed databases of multiple "data silos" (e.g., within different organizations countries). To develop effective machine learning services, there is a must exploit from such without exchanging raw data. Recently, federated (FL) has solution with growing interests, which enables parties collaboratively train model their local A key common challenge on heterogeneity distribution among parties. The are usually non-independently identically (i.e., non-IID). There many FL algorithms address effectiveness under non-IID settings. However, lacks an experimental study systematically understanding advantages disadvantages, as previous studies very rigid partitioning strategies parties, hardly representative thorough. In this paper, help researchers better understand setting in learning, we propose comprehensive cover typical cases. Moreover, conduct extensive experiments evaluate state-of-the-art algorithms. We find that does bring significant challenges accuracy algorithms, none existing outperforms others all Our provide insights for future addressing silos".

Language: Английский

---

Swarm Learning for decentralized and confidential clinical machine learning

Nature, Journal Year: 2021, Volume and Issue: 594(7862), P. 265 - 270

Published: May 26, 2021

Fast and reliable detection of patients with severe heterogeneous illnesses is a major goal precision medicine1,2. Patients leukaemia can be identified using machine learning on the basis their blood transcriptomes3. However, there an increasing divide between what technically possible allowed, because privacy legislation4,5. Here, to facilitate integration any medical data from owner worldwide without violating laws, we introduce Swarm Learning-a decentralized machine-learning approach that unites edge computing, blockchain-based peer-to-peer networking coordination while maintaining confidentiality need for central coordinator, thereby going beyond federated learning. To illustrate feasibility Learning develop disease classifiers distributed data, chose four use cases diseases (COVID-19, tuberculosis, lung pathologies). With more than 16,400 transcriptomes derived 127 clinical studies non-uniform distributions controls substantial study biases, as well 95,000 chest X-ray images, show outperform those developed at individual sites. In addition, completely fulfils local regulations by design. We believe this will notably accelerate introduction medicine.

Language: Английский

---

Multimodal biomedical AI

Nature Medicine, Journal Year: 2022, Volume and Issue: 28(9), P. 1773 - 1784

Published: Sept. 1, 2022

Language: Английский

---

Generalizing to Unseen Domains: A Survey on Domain Generalization

IEEE Transactions on Knowledge and Data Engineering, Journal Year: 2022, Volume and Issue: unknown, P. 1 - 1

Published: Jan. 1, 2022

Machine learning systems generally assume that the training and testing distributions are same. To this end, a key requirement is to develop models can generalize unseen distributions. Domain generalization (DG), i.e., out-of-distribution generalization, has attracted increasing interests in recent years. deals with challenging setting where one or several different but related domain(s) given, goal learn model an test domain. Great progress been made area of domain for This paper presents first review advances area. First, we provide formal definition discuss fields. We then thoroughly theories carefully analyze theory behind generalization. categorize algorithms into three classes: data manipulation, representation learning, strategy, present popular detail each category. Third, introduce commonly used datasets, applications, our open-sourced codebase fair evaluation. Finally, summarize existing literature some potential research topics future.

Language: Английский

---

AI applications to medical images: From machine learning to deep learning

Physica Medica, Journal Year: 2021, Volume and Issue: 83, P. 9 - 24

Published: March 1, 2021

Language: Английский

---

Federated learning for predicting clinical outcomes in patients with COVID-19

Nature Medicine, Journal Year: 2021, Volume and Issue: 27(10), P. 1735 - 1743

Published: Sept. 15, 2021

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining anonymity, thus removing many barriers to sharing. Here we 20 institutes across the globe train FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts future oxygen requirements of symptomatic patients COVID-19 using inputs vital signs, laboratory and X-rays. achieved an average area under curve (AUC) >0.92 predicting outcomes at 24 72 h time initial presentation emergency room, it provided 16% improvement in AUC measured all participating sites increase generalizability 38% when compared trained single site site's data. For prediction mechanical ventilation treatment or death largest independent test site, sensitivity 0.950 specificity 0.882. In this study, facilitated rapid science collaboration without exchange generated model generalized heterogeneous, unharmonized datasets clinical COVID-19, setting stage broader use healthcare.

Language: Английский

---

Artificial intelligence based anomaly detection of energy consumption in buildings: A review, current trends and new perspectives

Applied Energy, Journal Year: 2021, Volume and Issue: 287, P. 116601 - 116601

Published: Feb. 9, 2021

Enormous amounts of data are being produced everyday by sub-meters and smart sensors installed in residential buildings. If leveraged properly, that could assist end-users, energy producers utility companies detecting anomalous power consumption understanding the causes each anomaly. Therefore, anomaly detection stop a minor problem becoming overwhelming. Moreover, it will aid better decision-making to reduce wasted promote sustainable efficient behavior. In this regard, paper is an in-depth review existing frameworks for building based on artificial intelligence. Specifically, extensive survey presented, which comprehensive taxonomy introduced classify algorithms different modules parameters adopted, such as machine learning algorithms, feature extraction approaches, levels, computing platforms application scenarios. To best authors' knowledge, first article discusses consumption. Moving forward, important findings along with domain-specific problems, difficulties challenges remain unresolved thoroughly discussed, including absence of: (i) precise definitions consumption, (ii) annotated datasets, (iii) unified metrics assess performance solutions, (iv) reproducibility (v) privacy-preservation. Following, insights about current research trends discussed widen applications effectiveness technology before deriving future directions attracting significant attention. This serves reference understand technological progress

Language: Английский

---