Earticle

클라우드 뿐 아니라 모바일 기기들이 함께 학습과 추론을 위한 연산들을 나누어 수행하고 이를 다시 종합하는 distributed learning은 주목 받고 있는 learning 기법들 중 한 형태이다. 이러한 형태의 learning의 경우, 모바일 기기들을 활용하여 학습하고 추론 연산을 수행할 때, 응용의 성능뿐 아니라 통신 상황과 모바일 기기들에서 소모되는 전력을 고려하는 것은 매우 중요하다. 따라서 본 논문에서는 엣지 클라우드와 연결된 멀티 홉 기반의 네트워크에서 distributed learning을 수행할 때, 통신 상황과 소모 전력 관점에서 학습 및 추론 연산 분배에 따른 주요 case들에 대한 분석을 수행한다.

MANET은 노드들의 이동성으로 인한 동적 토폴로지와 hop-by-hop 데이터 전달 방식의 특징으로 인해 많은 공격들 의 대상이 된다. 그리고 MANET에서는 침입탐지시스템의 위치 설정이 어렵고, 지역적으로 수집된 정보로는 공격 탐지 가 더욱 어렵다. 또한 트래픽 양이 많아지면 침입탐지 성능이 현저히 떨어지게 된다. 따라서 본 논문에서는 MANET을 zone 형태로 구성한 후 대용량의 트래픽에도 안정된 침입탐지를 수행할 수 있도록 하기 위하여 정보 손실 없이 차원을 축소할 수 있는 random projection 기법을 사용하였다. 그리고 지역적인 정보만으로 탐지가 어려운 공격 탐지를 위해서 전역 탐지 노드를 이용하였다. 전역 탐지 노드에서는 IDS 에이전트들로부터 수신한 정보와 노드들의 패턴을 이용하여 공격 탐지를 수행하게 된다. 본 논문에서 제안한 기법의 성능 평가를 위하여 k-NN 기법과 ZBIDS 기법과 비교 실험하 였으며, 실험을 통해 성능의 우수성을 확인하였다.

MANET(Mobile Ad-hoc Network) is target of many attacks because of dynamic topology and hop-by-hop data transmission method. In MANET, location setting of intrusion detection system is difficult and attack detection using information collected locally is more difficult. The amount of traffic grow, intrusion detection performance will be decreased. In this paper, MANET is composed of zone form and we used random projection technique which reduces dimension without loss of information in order to perform stable intrusion detection in even massive traffic. Global detection node is used to detect attacks which are difficult to detect using only local information. In the global detection node, attack detection is performed using received information from IDS agent and pattern of nodes. k-NN and ZBIDS were experimented to evaluate performance of the proposed technique in this paper. The superiority of performance was confirmed through the experience.

The rapid growth of IoT devices has heightened the risk of botnet attacks, calling for scalable and distributed detection solutions. In this context, this study proposes a distributed optimization system for IoT attack detection using CNN model utilizing federated learning. After optimizing the hyperparameters of the model at the server, the Siberian Tiger Optimization (STO) method distributes these values to clients for dispersed training. Our model achieves accuracy, recall, and precision of 0.89978, 0.94355, and 0.94455, respectively, using the N-BaIoT dataset. These findings show, in spite of latency issues, the efficiency of federated learning in distributed IoT security systems.

To effectively manage the escalating traffic, attention is drawn to the coexistence mechanism of cellular and Wi-Fi networks which offload the cellular traffic from the licensed band to the unlicensed band. For the coexistence of 5G new radio-unlicensed and Wi-Fi, we tackle the energy efficiency maximization problem by sequentially determining the duty cycle and transmit power through the use of distributed deep Q-network (DQN) techniques. When utilizing the existing unlicensed band, the proposed method minimizes the impact on Wi-Fi networks while achieving optimal energy efficiency. Simulations validate the superior energy efficiency of the proposed coexistence mechanism over various benchmark methods.

In this paper, we propose a novel successive cancellation flip (SCF) decoding to reduce the computational complexity compared to the conventional SCF decoding by using distributed CRC bits. The proposed decoding reduces the number of estimations for information bits by early termination of decodings for failed frames of the first SC decoding, while trying to minimize the additional sorting operations. Simulation results show that the proposed SCF decoding reduces the computational complexity of repeated SC decoding at least 27% compared to the conventional SCF decoding.

The urban aerial mobility (UAM) system, such as drone taxi or air taxi, is one of future on-demand transportation networks. Among them, electric vertical takeoff and landing (eVTOL) is one of UAM systems that is for identifying the locations of passengers, flying to the positions where the passengers are located, loading the passengers, and delivering the passengers to their destinations. In this paper, we propose a distributed deep reinforcement learning where the agents are formulated as eVTOL vehicles that can compute the optimal passenger transportation routes under the consideration of passenger behaviors, collisions among eVTOL, and eVTOL battery status.

As the cornerstone of IoT-based systems, WSN connecting a wide range of intelligent sensor nodes is expected to bring significant changes in the near future. Due to the limited battery capacity of the sensor node, WSN considers maximizing network lifetime by minimizing the power consumption to be the most important challenge. To this end, we propose a Distributed Adaptive Communication with On/Off switching and Dual queuing for Energy efficiency (DACODE) as a novel asynchronous duty cycling mechanism. The performance evaluation shows that the proposed mechanism significantly reduces power consumption while maintaining network throughput and guaranteeing data urgency and queue stability.

Although deep learning (DL) has already become a state-of-the-art technology for various data processing tasks, data security and computational overload problems often arise due to their high data and computational power dependency. To solve this problem, quantum deep learning (QDL) and distributed deep learning (DDL) has emerged to complement existing DL methods. Furthermore, a quantum distributed deep learning (QDDL) technique that combines and maximizes these advantages is getting attention. This paper compares several model structures for QDDL and discusses their possibilities and limitations to leverage QDDL for some representative application scenarios.

This paper presents a new 60 GHz amplifier design method using TSMC 90 nm CMOS technology. The proposed design scheme is based on the way of the distributed amplifier design with open gate & open drain. The proposed architecture is analyzed and we define the upper limit to adopt proposed method in designing 60 GHz amplifier. To verify the effectiveness of the proposed scheme, the 60 GHz power amplifier with three cascade structure is designed and measured. Thanks to the proposed design structure, the fabricated power amplifier shows 29 dB peak gain, 25 dB gain over the 56~65 GHz, 11.5 dBm OP1dB, 13 dBm saturation power and peak 18.8 % PAE(Power Added Efficiency) under 2V supply voltage. Judging from the measurement results, the suggested architecture is a promising way for the future millimeter wave CMOS amplifier.

Distributed machine learning utilization in the metaverse exposes many potential benefits. However, the combination of these advanced technologies raises significant privacy concerns due to the potential exploitation of sensitive user and system data. This paper provides a systematic investigation of over 100 recent studies across key academic databases obtained by initial keyword-filter screening followed by a thorough full-text review. Particularly, metaverse evolution and enabling infrastructure technologies are briefly summarized. Subsequently, the distributed learning architectures and their features are analyzed as well as possibly associated vulnerability discussions. Then, envisioned metaverse applications and future research challenges are highlighted before concluding remarks.

As the works on quantum computing have seen substantial development in recent years, we are on the verge of seeing the fruitful result of the preliminary, early version of quantum computers. Nevertheless, in order to achieve full-fledged, large-scale quantum computers, two aspects still lacking in the existing quantum computers, i.e., scalability and reliability, will need to be more carefully considered and investigated. In this survey, we present a review of existing literature that aims to alleviate the scalability and reliability issues in quantum computers. In particular, we discuss how existing research leads to two main solutions: for advancing scalability, a distributed quantum computing (DQC) paradigm will be the primary direction instead of the general centralized quantum computing (CQC), whereas for attaining reliability, a fault-tolerant quantum computation (FTQC) approach would need to be leveraged instead of the existing noisy intermediate-scale quantum computers (NISQ). Combining both solutions, we highlight the essentiality of fault-tolerant distributed quantum computation (FT-DQC) and present related progress in the field. Furthermore, from the papers that we have gathered, we provide the taxonomy of the works to give a clearer landscape of the field and discuss the key issues in realizing FT-DQC.

Consider a large-scale distributed system in which each computing device is observing triggers from an external source. Distributed Trigger Counting (DTC) algorithm is used to detect the state of the system when the aggregated number of the observed triggers reaches a predefined value. In this paper, we propose a simple and efficient DTC algorithm: Cascading Thresholds (CT). We mathematically show that CT is an optimal DTC algorithm in terms of the total number of exchanged messages among the devices (message complexity). For the maximum number of received messages per device (MaxRcv), CT is sub-optimal. The average message complexity of CT is , and MaxRcv of it is , where is the number of triggers to be detected, is the number of devices, and is the degree of a node in the tree-like structure. Compared to the previous optimal algorithm (TreeFill), CT is much simpler: in our implementation the code size is about 2.5 times smaller. Also, unlike TreeFill CT does not require complicated mechanisms including distributed locking. Experimental results show that CT has a lower message complexity and MaxRcv compared to the previous work (CoinRand and RingRand). Furthermore, CT and TreeFill show a similar performance. From its simplicity, CT is more practical than previous work including TreeFill, CoinRand and RingRand.

In this work, we model multi-user distributed channel access as a game with channels and users, and propose the Multi-Agent Thompson Sampling (MA-TS) algorithm. It uses Bayes’ theorem to dynamically optimize action selection. This optimization aims to maximize throughput. We derive the algorithm’s computational complexity as . Simulations show that MA-TS converges to a pure strategy Nash equilibrium (PNE) and outperforms existing methods in average throughput.

A distributed congestion control must be adaptable to varying target communication ranges as cellular V2X (C-V2X) is evolving to support flexible coverage suitable for various service scenarios. This study proposes range-adaptive distributed power control (Ra-DPC) based on deep reinforcement learning (DRL) with the Monte Carlo policy gradient algorithm. A key finding is that the agents learn Ra-DPC more effectively when the cumulative interference power of the subchannels is adopted as the state of the DRL model, rather than the channel busy ratio. The proposed Ra-DPC algorithm performs better in energy efficiency and packet delivery ratio than the existing technologies.

The content distributed networks (CDNs) have been extending their market coverage on the conventional Internet environment by providing better quality of service (QoS) to the consumers. We consider the CDN as an economic player that competes with transit service provider (TSP), and model it as a platform that mediates content providers (CPs) and Internet service providers (ISPs) that make an independent decision on their participation to the CDN. Given the information about CPs and ISPs, the CDN can maximize its revenue by optimizing the prices at the both sides. This problem can be formulated as a combinatorial optimization problem. However, all possible combinations of participating CPs and ISPs should be considered, which require intensive computation, in particular, under CP dynamics where CPs arrive and depart the market over time. In this work, we develop a low-complexity pricing scheme that aims to find close-to-optimal prices.

Power output from large-scale solar farms is often plagued by anomalies that can adversely impact grid integration. This paper presents an anomaly detection system that used Siamese-twin neural networks for anomaly detection on edge devices in a solar farm. The model achieved an F1-score of 0.88 and was evaluated using two multi-threading schemes on a Raspberry PI, Nvidia Nano and Google Coral. A single analytics edge device could service 512 solar panels at 1 Hz. The best hardware platform was Nvidia’s Nano using a TensorFlow Lite model consuming about 35 Wh over 12 h, and with maximum CPU utilization not exceeding 60%.

In this paper, a sparsity-aware hybrid target localization method in multiple-input-multiple-output (MIMO) radars from time difference of arrival (TDOA) and angle of arrival (AOA) measurements is proposed. This method provides a maximum likelihood estimate of target position by employing compressive sensing techniques. A blockwise approach is addressed in order to achieve better accuracy for a constant computational complexity. The mismatch problem due to grid discretization is also tackled by a dictionary learning technique. The Cramer-Rao lower bound for this model is derived as a benchmark. Numerical simulations are included to corroborate the theoretical developments.