Distributed Artificial Intelligence Empowered by End-Edge-Cloud Computing: A Survey

IEEE Communications Surveys & Tutorials, Journal Year: 2022, Volume and Issue: 25(1), P. 591 - 624

Published: Nov. 4, 2022

As the computing paradigm shifts from cloud to end-edge-cloud computing, it also supports artificial intelligence evolving a centralized manner distributed one. In this paper, we provide comprehensive survey on (DAI) empowered by (EECC), where heterogeneous capabilities of on-device edge and are orchestrated satisfy diverse requirements raised resource-intensive AI computation. Particularly, first introduce several mainstream paradigms benefits EECC in supporting AI, as well fundamental technologies for AI. We then derive holistic taxonomy state-of-the-art optimization that boost training inference, respectively. After that, point out security privacy threats DAI-EECC architecture review shortcomings each enabling defense technology accordance with threats. Finally, present some promising applications enabled highlight research challenges open issues toward immersive performance acquisition.

Language: Английский

---

A Review of Trustworthy and Explainable Artificial Intelligence (XAI)

IEEE Access, Journal Year: 2023, Volume and Issue: 11, P. 78994 - 79015

Published: Jan. 1, 2023

The advancement of Artificial Intelligence (AI) technology has accelerated the development several systems that are elicited from it. This boom made vulnerable to security attacks and allows considerable bias in order handle errors system. puts humans at risk leaves machines, robots, data defenseless. Trustworthy AI (TAI) guarantees human value environment. In this paper, we present a comprehensive review state-of-the-art on how build eXplainable AI, taking into account is black box with little insight its underlying structure. paper also discusses various TAI components, their corresponding bias, inclinations make system unreliable. study necessity for many verticals, including banking, healthcare, autonomous system, IoT. We unite ways building trust all fragmented areas protection, pricing, expense, reliability, assurance, decision-making processes utilizing diverse industries differing degrees. It emphasizes importance transparent post hoc explanation models construction an lists potential drawbacks pitfalls AI. Finally, policies developing vehicle sectors thoroughly examined eclectic reliable, interpretable, eXplainable, explained guarantee safe systems.

Language: Английский

---

A Human-Centric Metaverse Enabled by Brain-Computer Interface: A Survey

IEEE Communications Surveys & Tutorials, Journal Year: 2024, Volume and Issue: 26(3), P. 2120 - 2145

Published: Jan. 1, 2024

The growing interest in the Metaverse has generated momentum for members of academia and industry to innovate toward realizing world. is a unique, continuous, shared virtual world where humans embody digital form within an online platform. Through avatar, users should have perceptual presence environment can interact control around them. Thus, human-centric design crucial element Metaverse. human are not only central entity but also source multi-sensory data that be used enrich ecosystem. In this survey, we study potential applications Brain-Computer Interface (BCI) technologies enhance experience users. By directly communicating with brain, most complex organ body, BCI hold intuitive human-machine system operating at speed thought. enable various innovative through neural pathway, such as user cognitive state monitoring, avatar control, interactions, imagined speech communications. This survey first outlines fundamental background technologies. We then discuss current challenges potentially addressed by BCI, motion sickness when environments or negative emotional states immersive applications. After that, propose new research direction called Human Digital Twin, which twins create intelligent interactable from user's brain signals. present solutions synchronizing between physical entities Finally, highlight challenges, open issues, future directions BCI-enabled systems.

Language: Английский

---

Edge Learning for 6G-Enabled Internet of Things: A Comprehensive Survey of Vulnerabilities, Datasets, and Defenses

IEEE Communications Surveys & Tutorials, Journal Year: 2023, Volume and Issue: 25(4), P. 2654 - 2713

Published: Jan. 1, 2023

The deployment of the fifth-generation (5G) wireless networks in Internet Everything (IoE) applications and future (e.g., sixth-generation (6G) networks) has raised a number operational challenges limitations, for example terms security privacy. Edge learning is an emerging approach to training models across distributed clients while ensuring data Such when integrated network infrastructures 6G) can potentially solve challenging problems such as resource management behavior prediction. However, edge (including deep learning) are known be susceptible tampering manipulation. This survey article provides holistic review extant literature focusing on learning-related vulnerabilities defenses 6G-enabled Things (IoT) systems. Existing machine approaches 6G–IoT learning-associated threats broadly categorized based modes, namely: centralized, federated, distributed. Then, we provide overview enabling technologies intelligence. We also existing research attacks against classify threat into eight categories, backdoor attacks, adversarial examples, combined poisoning Sybil byzantine inference dropping attacks. In addition, comprehensive detailed taxonomy comparative summary state-of-the-art defense methods vulnerabilities. Finally, new realized, overall prospects IoT discussed.

Language: Английский

---

Secure and Efficient Federated Learning With Provable Performance Guarantees via Stochastic Quantization

IEEE Transactions on Information Forensics and Security, Journal Year: 2024, Volume and Issue: 19, P. 4070 - 4085

Published: Jan. 1, 2024

Federated learning is a popular distributed machine paradigm that enables collaborative model training at multiple entities via exchanging intermediate results. Security and communication efficiency are crucial for successful applications of federated in various privacy-sensitive services. However, existing work focused on gradient defense separately, also incurred additional computation, signaling, accuracy overhead. A lightweight (in terms time-complexity signaling) technique simultaneously achieves security critical massive resource-constrained devices (e.g., Internet-of-Things generating the data), but has yet to be established. This paper proposes secure efficient framework with provable communication-accuracy-security performance guarantees. low-complexity signaling-free stochastic quantization module added client side quantizes original local gradients discrete values communication-efficient global aggregation. The shown interpreted as triangular or Gaussian-multiply-triangular noises under uniform Gaussian distributions gradients, hence protecting data privacy. We prove proposed exhibits an { O (log ₂ 1/δ), (δ ² ), (1/δ)}-tradeoff between overhead, accuracy, protection, where δ adjustable interval. Experimental results validate tradeoff superiority (only 14.1% differential privacy 0.2% homomorphic encryption) computation complexity (similar only 0.03% encryption). Under same protection performance, approach outperforms accuracy) all 9 comparison settings CIFAR10 dataset.

Language: Английский

---

Bidirectional domain transfer knowledge distillation for catastrophic forgetting in federated learning with heterogeneous data

Knowledge-Based Systems, Journal Year: 2025, Volume and Issue: unknown, P. 113008 - 113008

Published: Jan. 1, 2025

Language: Английский

---

Multi-Agent AI: From Isolated Agents to Cooperative Ecosystems

Published: Jan. 1, 2025

Language: Английский

---

Intentions and Premises

Advances in public policy and administration (APPA) book series, Journal Year: 2025, Volume and Issue: unknown, P. 321 - 360

Published: Jan. 24, 2025

This chapter evaluates the sustainability premises of strategy documents on AI for U.S. security institutions regarding intent, and foundational assumptions. While doing so, paper covers a discussion position in public administration how is captured within these strategies sector. These results indicate that work toward enhancing effectiveness operations using AI, but also address long-term, critical challenges energy efficiency, social equity, economic responsibility. The concept based assumptions, such as fact should serve today's needs while protecting those future generations. can be interpreted to mean far stronger integration secure-by-design principles much wider focus minimizing negative societal impacts AI. It therefore helps provide an insight into governance evolves through complex adaptive system framing strategic foresight developing balanced ethical applications.

Language: Английский

---

Over-the-Air Computation for Distributed Systems: Something Old and Something New

IEEE Network, Journal Year: 2023, Volume and Issue: 37(5), P. 240 - 246

Published: June 19, 2023

Facing the upcoming era of Internet-of-Things and connected intelligence, efficient information processing, computation, communication design becomes a key challenge in large-scale intelligent systems. Recently, Over-the-Air (OtA) computation has been proposed for data aggregation distributed functions over large set network nodes. Theoretical foundations this concept exist long time, but it was mainly investigated within context wireless sensor networks. There are still many open questions when applying OtA different types systems where modern technology is applied. In article, we provide comprehensive overview principle its applications learning, control, inference systems, both server-coordinated fully decentralized architectures. Particularly, highlight importance statistical heterogeneity channels, temporal evolution model updates, choice performance metrics, federated learning (FL) Several challenges privacy, security, robustness aspects FL also identified further investigation.

Language: Английский

---

I/O Efficient Label-Constrained Reachability Queries in Large Graphs

Proceedings of the VLDB Endowment, Journal Year: 2024, Volume and Issue: 17(10), P. 2590 - 2602

Published: June 1, 2024

Computing the reachability between two vertices in a graph is fundamental problem data analysis. Most of existing works assume that edges have no labels, but many real application scenarios, naturally come with edge-labels, and label constraints may be placed on appearing valid path query vertices. Therefore, we study label-constrained (LCR) queries this paper, where are given source vertex s , target t set Δ, goal to check whether there exists any from such all labels belong Δ. A plethora methods been proposed literature support LCR queries. All these take assumption resident main memory machine. Nevertheless, graphs scenarios generally big not reside memory. In cases, suffer serious scalability problem, i.e., result huge I/O costs. Motivated by this, efficient aim efficiently answer when cannot fit To achieve goal, propose reduction-based indexing approach. We introduce elegant reduction operators which aims reduce size loaded while preserving information among remaining With operators, devise an index named LCR-Index algorithms adaptively construct based available Equipped LCR-Index, can only scanning sequentially. Experiments demonstrate our processing algorithm handle billions edges.

Language: Английский

---