Large language models (LLMs) often generate hallucinations—fluent yet factually incorrect responses—that undermine reliability in knowledge-intensive tasks. Existing approaches for hallucination mitigation typically rely on external retrieval modules or probability heuristics, which either require additional resources or lack interpretability. In this work, we propose a diffusion-based hallucination detection framework (DHDF) that leverages U-Net denoising to reconstruct consensus answers from multiple LLM outputs. If the diffusion process exhibits spurious convergence away from factual ground truth, it provides a clear signal of hallucination. To quantify factual correctness, we incorporate TruthfulQA scores as a fact-grounded evaluation metric, distinguishing well-aligned models (high scores) from hallucination-prone models (low scores). Experimental results demonstrate that convergence dynamics under diffusion, combined with fact-grounded QA evaluation, offer an effective and interpretable pathway for hallucination detection without relying on external knowledge bases.

End-to-end (E2E) autonomous driving must maintain global consistency while preserving local precision. However, existing E2E approaches rarely achieve both goals simultaneously. Therefore, we propose a multimodal coarse-to-local transformer (MC2L-Transformer), which is composed of a hierarchical transformer architecture. Multimodal inputs are fused into a shared embedding, and global waypoints are produced. Local refinement is then utilized to capture fine interactions around the vehicle. Furthermore, a temporal encoder summarizes recent context, and navigation target and velocity are embedded to guide route- and speed-aware decoding. We evaluate in CARLA, and the results show lower collision and off-route rates even under sudden events. These results indicate that combining a coarse-to-local hierarchical transformer with a lightweight temporal context provides a practical step toward reliable E2E autonomous driving.

In adaptive cruise control (ACC), balancing safety, comfort, and sustainability still remains challenging. Accordingly, we propose a hybrid reinforcement learning framework combining proximal policy optimization (PPO) and deep Q-network (DQN) with a multi-objective reward for autonomous carbon-neutral eco-driving. Experimental results revealed the contrasts between eco and non-eco modes, underscoring how reward design shapes driving behaviors.

Quantum machine learning (QML) has attracted growing interest for their ability to achieve superior performance with significantly fewer parameters. However, the high cost and scarcity of current hardware push inference to cloud-hosted quantum devices, creating a tension between verifiability and confidentiality. This work proposes a novel framework that converts quantum neural network operations into classical arithmetic circuits that faithfully approximate genuine quantum computations. By encrypting these circuits with zero-knowledge proofs, it ensures computational validity while concealing internal parameters. Experimental results show that our classical circuits achieve fidelity above 0.9996 and total variation distance below 1% compared to actual quantum computations, verifying the practicality of trustworthy and privacy-preserving quantum inference.

• This work proposes a practical initial layout method that utilizes matrix formulation to determine enhanced qubit assignments, without iterative optimization techniques and combinatorial search. • The proposed method computes the gradient of the objective function through a single matrix differentiation to obtain assignment suitability information between logical and physical qubits. • The gradient matrix encodes the interaction between physical-qubit centrality and logical-qubit connectivity, enabling direct mapping decisions that minimize error costs while satisfying hardware constraints. • Simulation results show that the proposed method improves computational efficiency while maintaining comparable execution reliability compared to existing qubit mapping methods. Qubit mapping is a crucial compilation process that assigns logical qubits from quantum circuits to physical qubits on quantum hardware, ensuring efficient and reliable execution on noisy intermediate-scale quantum (NISQ) computers. However, as the scale of quantum computers and circuit complexity increase, existing mapping approaches face significant computational complexity and suboptimal mapping quality. This work formulates the logic-to-physical qubit mapping problem as a matrix-form optimization problem to address scalability and computational efficiency challenges. The proposed method computes the gradient of the matrix-formulated problem through a single matrix differentiation and uses it as guidance to determine the enhanced qubit assignment, without the need for iterative combinatorial exploration and optimization solvers. Simulation results show that the proposed method achieves comparable execution reliability while significantly reducing the compilation time, even for multi-programming scenarios and large-scale quantum computers.

Low Earth Orbit (LEO) satellite constellations are crucial for global connectivity by providing extensive coverage and reduced delays. However, scheduling data transmission in these dynamic networks is challenging due to rapidly changing satellite positions. This research introduces BoostRL (boosted reinforcement learning), a novel framework integrating large language models (LLMs) with reinforcement learning (RL) for efficient scheduling in LEO satellite constellations. BoostRL leverages LLM-generated initial policies to accelerate convergence, thereby guiding the early stages of policy learning while adapting swiftly to dynamic network conditions. It employs a hybrid policy approach which transitions smoothly from LLM recommendations to autonomous RL policies. A tailored initialization of Q-value parameters and an enhanced loss function further optimize learning efficiency, by aligning the initial learning phase with LLM-generated insights. Simulations using two-line element (TLE) orbital data demonstrate that BoostRL achieves rapid convergence and improved efficiency, thereby validating its potential as a scalable, adaptive solution for managing satellite communication networks.

Spurred by the limited availability of quantum resources, known as qubits, in recent quantum computers, quantum federated learning (QFL) is drawing attention. Due to its ability to fully utilize distributed qubits, QFL is suitable for developing quantum algorithms. QFL achieves local quantum gradients using the parameter-shift rule (P-S rule) and aggregates them, effectively coping with the limited number of qubits in each quantum computer. However, realizing QFL remains challenging due to the characteristics of the P-S rule, which requires two forward passes to compute the quantum gradients in each local quantum computer. These challenges become even more severe when the aggregation of the local quantum gradients occurs under heterogeneous channel conditions and data distributions. Motivated by this, this paper proposes the joint P-S rule, which eliminates the aggregation process in QFL and instead directly achieves the global quantum gradients. Furthermore, this paper proposes joint efficient quantum federated learning (Joint EQFL) that leverages successive interference cancellation and divergence-based clustering for achieving stability under heterogeneous channel conditions and robustness to heterogeneous data distributions. This paper analyzes the convergence and corroborates the superiority of Joint EQFL.

The increasing demand for ultra-low-latency and high-throughput communication necessitates network architectures capable of ensuring reliable real-time connectivity. To address this requirement, integrated architectures combining terrestrial networks (TN) and non-terrestrial networks (NTN)—particularly hierarchical multi-layer satellite systems composed of geostationary Earth orbit (GEO), medium Earth orbit (MEO), and low Earth orbit (LEO) satellites—have attracted considerable attention. Such hierarchical systems offer global coverage, low latency, and high capacity. However, their highly dynamic topology, fast-moving satellites, and fluctuating link quality pose significant challenges to establishing efficient end-to-end routing paths. Traditional routing methods based on static paths or centralized control lack the adaptability required for such environments. To address these challenges, this paper introduces a neural hierarchical reinforcement learning (HRL)-based routing algorithm for satellite networks. The framework exploits the layered structure of GEO, MEO, and LEO satellites, enabling gateway-assisted routing and adaptive path selection. Control responsibilities are distributed across orbital layers: the GEO layer allocates hop-count budgets, the MEO layer selects feasible LEO paths, and the LEO layer forwards data packets. When inter-satellite links are unavailable, terrestrial gateways provide alternative routes. Routing decisions further account for satellite-to-satellite and satellite-to-ground link quality, with an end-to-end delay-based reward function capturing transmission, processing, and propagation effects. Simulation results show that the proposed neural hierarchical framework, supported by integrated terrestrial and non-terrestrial networking, achieves faster convergence, greater route stability, and better adaptability than conventional routing algorithms and HRL approaches.

Providing seamless global Internet connectivity remains a significant challenge due to geographical barriers, economic constraints, and vulnerabilities to catastrophic events, including natural disasters and warfare. Non-terrestrial Networks (NTN), particularly satellite communication systems, are crucial for overcoming these limitations, but traditional geostationary Earth orbit (GEO) satellites experience inherent latency issues due to their high-altitude positioning. Low Earth orbit (LEO) satellites have emerged as superior alternatives, significantly reducing communication delays and enhancing network performance. Despite their advantages, LEO satellite networks face significant challenges arising from rapidly changing network topologies driven by high orbital speeds, resulting in frequent shifts in inter-satellite link (ISL) availability. Conventional routing algorithms struggle to adapt effectively to these dynamic conditions, resulting in suboptimal routing decisions and increased latency due to congestion-related issues. To address these complexities, this paper proposes an adaptive reinforcement learning-based NTN routing (ARL-NTNR) algorithm designed explicitly for dynamic LEO environments. The proposed ARL-NTNR algorithm autonomously learns and adapts optimal routing policies in real-time by continually interacting with the evolving satellite network. It effectively balances shortest-path routing with real-time queue backlog congestion management, significantly reducing packet delay and minimizing packet loss.

Achieving global space-air-ground integrated network (SAGIN) access only with CubeSats presents significant challenges such as the access sustainability limitations in specific regions (e.g.,polar regions) and the energy efficiency limitations in CubeSats. To tackle these problems, high-altitude long-endurance unmanned aerial vehicles (HALE-UAVs) can complement these CubeSat shortcomings for providing cooperatively global access sustainability and energy efficiency. However, as the number of CubeSats and HALE-UAVs, increases, the scheduling dimension of each ground station (GS) increases. As a result, each GS can fall into the curse of dimensionality, and this challenge becomes one major hurdle for efficient global access. Therefore, this paper provides a quantum multi-agent reinforcement Learning (QMARL)-based method for scheduling between GSs and CubeSats/HALE-UAVs in order to improve global access availability and energy efficiency. The main reason why the QMARL-based scheduler can be beneficial is that the algorithm facilitates a logarithmic-scale reduction in scheduling action dimensions, which is one critical feature as the number of CubeSats and HALE-UAVs expands. Additionally, individual GSs have different traffic demands depending on their locations and characteristics, thus it is essential to provide differentiated access services. The superiority of the proposed scheduler is validated through data-intensive experiments in realistic CubeSat/HALE-UAV settings.

Satellites have been envisioned as primary non-terrestrial networks capable of seamless global network and surveillance services. Among various satellite types, Cube Satellites (CubeSats) have been actively researched because multiple CubeSats can be conveniently positioned in a target orbit simultaneously and in proximity to Earth. However, CubeSats are small-scale, and thus, they are not able to accommodate a sizable battery, imposing constraints on the duration of their mission. Considering this energy limitation, in order to realize global network services using multiple CubeSats, this paper proposes a novel two-stage Reinforcement Learning (RL) algorithm for energy-efficient CubeSats where RL is utilized for dynamic control under uncertainty. Firstly, sustainable control for single-CubeSat orbital maneuver is considered using deep deterministic policy gradient for vertical position adjustment over a continuous action domain. Secondly, a novel quantum multi-agent RL algorithm for multi-CubeSat cooperative scheduling is designed to realize action dimension reduction into a logarithmic scale based on our proposed Projection-Valued Measure (PVM) over the quantum domain. It is highlighted that our considering two single- and multi-CubeSat problems cannot be separately considered for extreme energy management. The performance evaluation results demonstrate that the proposed algorithm outperforms other benchmarks with 1.51× higher performance in orbital control, 2.71× higher converged reward in enormous action dimensions, and 2.27× higher average network performance.

This paper proposes a novel quantum style transfer (QST) in hybrid quantum-classical computing. QST leverages quantum computing's ability to process high-dimensional data efficiently. Our approach aims to decrease both inference time and complexity while maintaining performance, presenting a viable solution that enhances the scalability and efficiency of image generation technologies.

Large language models (LLMs) require substantial resources for task-specific adaptation, that motivates the development of parameter-efficient fine-tuning (PEFT) methods. This paper presents quantum-amplitude embedded adaptation (QAA), a novel PEFT framework that logarithmically compresses activation vectors using quantum-amplitude embedding and applies expressive non-linear transformations via parameterized quantum circuits (PQCs). By replacing linear adapters in attention modules with compact quantum modules, QAA achieves high expressivity while drastically reducing the number of trainable parameters. Empirical results demonstrate that QAA performs on par with or better than existing PEFT under constrained memory and compute budgets, highlighting its potential for efficient LLM fine-tuning.

Large-scale university courses face significant challenges. Teaching assistants are overwhelmed by the large number of student questions, limiting their ability to provide detailed and individualized support. As a result, students-especially those who are struggling-receive less tailored assistance, further widening gaps in academic performance. To address these challenges, we propose student-centered AI learning assistant (SCALA), a large language model (LLM)-based interactive tutoring system that incorporates student needs and learning expectations. SCALA consists of two main components, i.e., predictive query management and student-centered fine tuning. The first component anticipates common student questions via LLM agent debate. Each agent interacts using a combination of lecture content and student interactions in chat logs, collaboratively predicting what the students will likely ask. This fosters learning in students by generating and presenting relevant queries that guide their learning. On the other hand, the second component is fine-tuned on a 14k-question Python-tutoring dataset, curated based on in-depth student interviews to reflect real learning expectations. Our real-world experiments with 1500-student large-scale Python classes demonstrate that SCALA delivers more helpful and accurate responses compared to closed-form models (e.g., GPT-4o), while significantly reducing latency.

This paper proposes SafeGPT, a two-tiered framework that integrates generative pretrained transformers (GPTs) with reinforcement learning (RL) for efficient and reliable un-manned aerial vehicle (UAV) last-mile deliveries. In the proposed design, a Global GPT module assigns high-level tasks such as sector allocation, while an On-Device GPT manages real-time local route planning. An RL-based safety filter monitors each GPT decision and overrides unsafe actions that could lead to battery depletion or duplicate visits, effectively mitigating hallucinations. Furthermore, a dual replay buffer mechanism helps both the GPT modules and the RL agent refine their strategies over time. Simulation results demonstrate that SafeGPT achieves higher delivery success rates compared to a GPT-only baseline, while substantially reducing battery consumption and travel distance. These findings validate the efficacy of combining GPT-based semantic reasoning with formal safety guarantees, contributing a viable solution for robust and energy-efficient UAV logistics.

Reinforcement learning (RL) using conventional neural networks (NN) has significantly progressed in various applications. However, conventional RL needs help training in environments with large-scale action dimensions, such as coordinated mobility/satellite systems. Quantum reinforcement learning (QRL) with quantum NN (QNN) can address this problem through superposition and entanglement, one of the great features of quantum mechanics. Based on its ‘i) fast convergence’ and ‘ii) high scalability’, unique advantages of QRL that distinguish it from conventional RL, this paper highlights the potential for QRL utilization in coordinated mobility and satellite systems.

As context windows expands, Large Language Models (LLMs) are being more capable of handling long-context tasks. However, long-context inference is hindered by high latency and memory usage due to the linearly growing Key-Value (KV) cache. Prior work has shown that a dynamic subset of the KV cache dominates attention outcomes and selectively loads these subsets instead of the full cache. These methods partition the KV cache into segments (block-wise and token-wise) and identify critical ones for recall. However, for block-wise ones, their recall accuracy is often compromised by semantic inconsistencies among consecutive tokens, and for token-wise ones, they suffer from inefficiency in memory management and offloading. To these ends, we propose Dispenser, a hierarchical KV cache management algorithm inspired by multi-level page tables. Dispenser organizes the KV cache based on token semantic similarity and partitions it into pages and blocks, which enables straightforward integration into modern serving engines such as vLLM. It employs a two-level lookup mechanism to identify and load the most relevant blocks for the current query. Experimental results demonstrate that Dispenser achieves up to a$5.75 \times$speedup in self-attention, reducing inference latency by$2.1 \times$on tasks with long dependencies with negligible accuracy loss.

Quantum multi-programming, which is a method to run multiple quantum programs concurrently, has the potential to significantly improve the overall throughput and resource utilization. However, previous approaches do not always guarantee consistent performance (e.g., circuit complexity and reliability) over all possible hardware platforms due to qubit connectivity and qubit error rates in individual hardware platforms. This paper proposes hardware-Agnostic QUbit Allocation, named$A Q U A$, that allows multiple quantum programs to run simultaneously while ensuring high reliability regardless of hardware platforms. The proposed AQUA optimizes quantum programs at the logicallevel, without physical qubit mappings and allocations, to ensure that they consist of fewer gates and shorter depths. By attempting to optimize quantum programs at the logical-level, regardless of hardware connectivity, the AQUA can ensure consistent circuit complexity and reliability. Then, the AQUA dynamically allocates quantum programs to more reliable physical qubits depending on circuit depths and qubit error rates. The effectiveness of the AQUA is demonstrated using several quantum hardware platforms, showing that AQUA ensures high reliability and consistent circuit complexity.

This paper proposes a novel algorithm for distributed multi unmanned aerial vehicles (UAVs) cooperation in dynamic and unstable network environments by employing joint multi-agent reinforcement learning (MARL) and message-passing. To realize MARL, our proposed algorithm utilizes a centralized training with distributed execution (CTDE) framework. However, CTDE-based algorithms should be able to recognize the communications between UAVs and centralized server, which is not possible in every single time step. Therefore, after conducting centralized training for MARL, the distribution of the model for distributed execution should be re-designed. For this objective, a conflict graph-based approach is used, which enables graph-edge if two UAVs can talk to each other. Based on this conflict graph construction, message-passing is used to select UAVs for communication with the server. The non-selected UAVs can receive their models from conflict graph-connected UAVs.

Polarization reconfigurable (PR) antennas provide a promising technology for next-generation wireless communications, due to their capability of reconfiguring wireless channel state. This paper pursues integrating PR antenna technology into massive multiple-input-multiple-output (MIMO) system in the scenario that channel state information (CSI) is not provided. We aim to jointly design the polarization and beamforming vectors on both transceivers for simultaneous channel reconfiguration and beam alignment to maximize the signal-to-noise-ratio (SNR). However, the joint optimization over the polarization and beamforming vectors without perfect CSI is a challenging task. Depolarization increases the dimension of CSI and massive MIMO systems typically have low-dimensional pilot measurement from limited number of radio frequency (RF) chains. This leads to pilot overhead because the transceivers can fulfill only the low-dimensional measurement of the high-dimensional channel. Substantial PR beamforming gain is achieved with significantly reduced pilot overhead, by employing a transformer-based framework on both transceivers to actively design the polarization and beamforming vectors for the pilot stage and transmission stage based on the sequence of received pilots. Numerical experiments demonstrate that the proposed framework remarkably outperforms the prevalent non-adaptive data-driven methods.

The stability of aircraft remains vulnerable to sudden external disturbances and unpredictable vortices. The aircraft's attitude angles undergo rapid changes due to random turbulence. Consequently, to ensure safety, it is essential to control the aircraft's control surfaces, i.e., ailerons, elevators, and rudder angles, to maintain its static stability. Although classical closed-loop control methods have been widely adopted, their limited adaptability to changing dynamics calls for more robust solutions. Reinforcement learning (RL) offers adaptive capabilities but often demands a large number of training parameters and substantial computational resources, which may be impractical for real-time lightweight aircraft applications. To overcome these limitations, this paper introduces a quantum aircraft with the quantum actorcritic networks-based aircraft control (QACN-AC) algorithm. By utilizing quantum neural networks (QNN), QACN-AC significantly reduces the number of parameters required for training, thus mitigating computational overhead while preserving robust control performance. The QACN-AC's effectiveness is validated through realistic simulations leveraging Boeing's B777 specifications. The results highlight QACN-AC's superiority over conventional RL, evidenced by a$1.25 \times$higher control performance and a$760 \times$reduction in the number of required parameters.

Cloud-edge collaboration provides an efficient way to promote the development and application of the Artificial Intelligence of Things (AIoT) by addressing the limited computing, communication, and storage capabilities of the Internet of Things (IoT) devices. The compute-intensive task of developing an Artificial Intelligence (AI) model is performed on the cloud-server over a public dataset while it is deployed on the edge-server to analyze the IoT data. The merit in processing data locally is the privacy preservation. However, to keep the model relevant to the ever-growing IoT data, it is necessary to share it with the cloud-server, which can raise certain privacy concerns. Also, it is challenging to adapt the model to the IoT data characteristics. Therefore, we propose a privacy-preserving cloud-edge paradigm incorporating personalized context learning and incremental learning strategies to adapt the trained model to the heterogeneous and dynamic IoT environment. To avoid any privacy issues, we propose a lightweight image cryptosystem parameterized to protect image contents according to an application requirement. The simulation results show that our proposed image cryptosystem has favorable encryption properties and the cipher-images are learnable by an AI model. In addition, our incremental learning strategy efficiently kept the model up-to-date, and with proposed personalized context learning strategy, the cloud-server model’s performance improved up to 13% for plain-images and 18% for cipher-images.

The rapid growth of edge computing has enabled low-latency and high-efficiency processing for a wide range of applications; however, it also leads to significant energy consumption and carbon emissions. In this context, this study investigates a CO2 emission minimisation problem in a digital twin-aided edge computing system, aiming to optimise task offloading decisions, transmit power, and processing rates of Internet of Things (IoT) devices. To address the formulated mixed-integer non-linear programming problem, we propose two solutions: an alternating optimisation method based on the successive convex approximation framework and a deep reinforcement learning (DRL) approach. Extensive simulations validate the effectiveness of the proposed solutions, demonstrating significant reductions in CO2 emissions, robust optimisation performance, and superior results compared to benchmark schemes. The findings highlight the feasibility of integrating advanced optimisation and artificial intelligence-driven techniques to achieve environmentally sustainable and high-performance edge computing systems, paving the way for greener technological innovation.

The integration of the non-terrestrial network (NTN) and the terrestrial network (TN) aims to provide seamless and ubiquitous connectivity by using satellites, unmanned aerial vehicles, and high-altitude platforms to expand coverage in remote and underserved areas. Despite its potential, the integration of NTN and TN faces significant challenges in interference management and resource allocation, primarily due to the shared spectrum and the distinct operational characteristics of both network types. To address these issues, this paper proposes a reinforcement learning-based framework for dynamic resource allocation and interference management in integrated NTN-TN environments. The framework is designed to maintain balanced performance between NTN and TN while achieving a high signal-to-interference-plus-noise ratio (SINR) and coverage probability (CP). An interference approximation model, validated via Monte Carlo simulations, ensures reliability and accuracy across various scenarios. The simulation results demonstrate that the framework effectively mitigates interference, adapting to varying user equipment densities while minimizing performance degradation. The proposed approach significantly enhances the co-existence of NTN and TN, achieving higher SINR and CP compared to existing methods. These findings establish a robust foundation for scalable and efficient NTN-TN integration in next-generation networks.

Low Earth orbit (LEO) satellite networks have emerged as a promising solution, offering advantages such as lower propagation delay, broader coverage, and rapid deployment capabilities. However, the dynamic topology and frequent handovers inherent in LEO satellite systems, coupled with limited onboard computational resources, necessitate the development of efficient and lightweight routing algorithms. Therefore, this paper proposes quantum reinforcement learning-based satellite routing (QRL-SR) tailored for LEO satellite networks. The QRL-SR algorithm addresses three critical considerations: (i) adapting to the dynamic and time-varying environment of LEO satellite networks; (ii) incorporating LEO satellite geometry by transforming celestial coordinate data, specifically two-line element, into orbital coordinate systems for accurate LEO satellite positioning over time; and (iii) being designed to be lightweight by leveraging QRL to reduce the number of training parameters. The proposed QRL-SR efficiently trains routing policies with fewer parameters, aligning with LEO satellites’ small-size, weight, and power (SWaP) constraints. The primary purpose of the QRL-SR-based LEO satellites is to reduce free space path loss, delay time, and the number of hops needed for routing through the inter-satellite links. Finally, experimental results demonstrate that the QRL-SR achieves routing performance comparable to or outperforms conventional algorithms while significantly reducing computational resources.

In order to build realistic digital-twin systems, this article proposes a novel two-stage algorithm for high-quality digital-twin services in cloud-assisted multitier networks. In our proposed algorithm, the first stage is quantum multiagent reinforcement learning (QMARL)-based scheduling for differentiated quality control of individual segments of digital-twin virtual objects in our cloud. As the number of segments selected by each edge increases, the edge’s action dimension expands exponentially, posing significant challenges to learning with conventional MARL. To solve this problem, the quantum-inspired MARL-based scheduler is considered in order to reduce the scheduling action dimensions into a logarithmic-scale. For the scheduling formulation, age-of-information (AoI) is also considered for low-latency high-quality digital-twin services. Additionally, the second stage is for the fast and seamless distribution of differentiated quality-controlled segments of virtual objects. For this objective, each user requests its desired segments and one of nearby edges is selected. Among various approaches, this second stage considers second price auction for truthful and distributed computation. Furthermore, low-complexity computation can be realized by avoiding integer-programming-based computation which is NP-hard. The proposed two-stage algorithm achieves performance levels that are 8.33 and 1.18 times higher in terms of reward value in high dimensions and revenue, respectively, compared to other benchmarks.

According to the advances in quantum computing and distributed learning, quantum federated learning (QFL) has recently become an emerging field of study. In QFL, each quantum computer or device locally trains its quantum neural network (QNN) with trainable gates, and communicates only these gate parameters over classical channels, without costly quantum communications. To successfully opeate QFL under various and dynamic channel conditions in Internet of Things (IoT) environments, this article develops a novel depth-controllable architecture of entangled slimmable QNNs (eSQNNs), and thus, proposes an entangled slimmable QFL (eSQFL) that communicates the superposition-coded parameters of eSQNNs. Even though the proposed eSQNN-based eSQFL is superior, training the depth-controllable eSQNN architecture is challenging due to high-entanglement entropy and interdepth interference. Therefore, the proposed method in this article mitigates the interference using entanglement controlled universal (CU) gates and an inplace fidelity distillation (IPFD) regularizer penalizing interdepth quantum state differences, respectively. Furthermore, the proposed method optimizes the superposition coding power allocation by deriving and minimizing the convergence bound of eSQFL. The novelty of this work is evaluated via extensive simulations in terms of prediction accuracy, fidelity, and entropy compared to Vanilla QFL as well as under different channel conditions and various data distributions.

This paper presents an advanced excavation automation framework that improves trajectory efficiency and improves operational stability in diverse and complex soil conditions. The framework integrates the fundamental equation of earthmoving (FEE) with the proximal policy optimization (PPO) algorithm to enable adaptive trajectory planning during excavation. Using these estimates, the PPO algorithm iteratively optimizes excavation strategies to minimize applied force and improve trajectory efficiency, contributing to potential energy savings. The validation of the proposed framework is performed through simulations conducted under varying soil conditions and initial height configurations. The results show an 18.1% reduction in total excavation resistance compared to baseline models, achieving a mean absolute error (MAE) of 0.065 m. These findings confirm the effectiveness of the framework in reducing excavation resistance and improving task precision in simulated environments.

Object detection applications, especially in autonomous driving, have drawn attention due to the advancements in deep learning. Additionally, with continuous improvements in classical convolutional neural networks (CNNs), there has been a notable enhancement in both the efficiency and speed of these applications, making autonomous driving more reliable and effective. However, due to the exponentially rapid growth in the complexity and scale of visual signals used in object detection, there are limitations regarding computation speeds while conducting object detection solely with classical computing. Motivated by this, this paper proposes the quantum object detection engine (QODE), which implements a quantum version of CNN, named QCNN, in object detection. Furthermore, this paper proposes a novel fast quantum convolution algorithm that processes the multi-channel of visual signals based on a small number of qubits and constructs the output channel data, thereby achieving relieved computational complexity. Our QODE, equipped with fast quantum convolution, demonstrates feasibility in object detection with multi-channel data, addressing a limitation of current QCNNs due to the scarcity of qubits in the current era of quantum computing. Moreover, this paper introduces a heterogeneous knowledge distillation training algorithm that enhances the performance of our QODE.

According to recent advances in cellular vehicle-to-everything (cellular-V2X) communication networks, vehicle-to-vehicle (V2V) networks can be performed without infrastructure support. The corresponding wireless standard has been actively studied by the 3rd generation partnership project (3GPP) and has been proposed to realize V2V communications via sidelink interfaces. This standard specifically introduces cellular-V2X Mode 4, where individual vehicles access wireless resources in a distributed manner by using sensing-based semi-persistent scheduling (SPS) for avoiding collisions in cooperative awareness messages (CAMs). However, the legacy SPS scheme is able to be challenged by resource scheduling collisions, which obviously degrade performance. To tackle this problem, this paper proposes intelligent adaptive and additive extra resource allocation strategy that responds in real time to estimate traffic density fluctuations, aiming to decrease resource collisions as well as improve utilization performance. Moreover, our additional intelligent solution, which is fundamentally inspired by Lyapunov optimization-based drift-plus-penalty framework, dynamically controls the number of resources to minimize transmission outage probability subject to resource constraints. The corresponding data-intensive performance analysis results verify that the proposed algorithm is able to improve performance compared to the other benchmarks.

Recent advances in deep learning have successfully replaced classical algorithms with machine learning models based on neural networks (NNs). This is particularly prevalent in proactive caching. As NNs grows more capable as their size in terms of storage and computation increases, NN-based proactive caching achieves performance improvement. Nonetheless, there remain challenges in implementing NN-based proactive caching in realistic environments with dynamic user movement. These are due to the fixed structure of NNs that should expand the input size to match the dimensions of the input with the dimensions of the NN’s input units. To address these challenges, this paper proposes a scalable proactive caching framework, named slimmable federated reinforcement learning (SlimFRL). By adopting slimmable neural networks (SNNs) in FRL, our SlimFRL easily adjusts the widths of the SNNs during training according to the number of users. Moreover, due to the scalability of SNNs, our SlimFRL can set the appropriate input dimension while not using imputation, leading to performance improvement. This paper also validates the performance and advantages of SlimFRL in terms of reward and additional cost functions. Additionally, this paper proposes several training algorithms for SlimFRL and corroborates their superiority with convergence analysis and various experiments.

This tutorial offers a hands-on introduction into the captivating field of quantum machine learning (QML). Beginning with the bedrock of quantum information science (QIS)-including essential elements like qubits, single and multiple qubit gates, measurements, and entanglement-the session swiftly progresses to foundational QML concepts. Participants will explore parameterized or variational circuits, data encoding or embedding techniques, and quantum circuit design principles. Delving deeper, attendees will examine various QML models, including the quantum support vector machine (QSVM), quantum feed-forward neural network (QNN), and quantum convolutional neural network (QCNN). Pushing boundaries, the tutorial delves into cutting-edge QML models such as quantum recurrent neural networks (QRNN) and quantum reinforcement learning (QRL), alongside privacy-preserving techniques like quantum federated machine learning, bolstered by concrete programming examples. Throughout the tutorial, all topics and concepts are brought to life through practical demonstrations executed on a quantum computer simulator. Designed with novices in mind, the content caters to those eager to embark on their journey into QML. Attendees will also receive guidance on further reading materials, as well as software packages and frameworks to explore beyond the session.

This article proposes a distributed angular routing algorithm in time-varying dynamic low Earth orbit (LEO) satellite constellation networks. For designing satellite routing algorithms, it is essential to consider 1) distributed operation due to the difficulty in global centralized computation and 2) angle-based computation under the consideration of orbit coordinate systems. Therefore, our proposed routing algorithm is based on distributed angular computation. Moreover, the proposed algorithm is designed by the Markov decision process (MDP) for discrete-time sequential decision making in time-varying LEO satellite networks. As a result, this article proposes an MDP-based distributed angular routing (MDAR) algorithm for seamless LEO routing. Based on the reward formulation in terms of angular differences in MDP formulation, our proposed distributed angular routing algorithm pursues orbit-geometrically straight-line data delivery from the source to its associated destination. Finally, our proposed routing algorithm is evaluated in the realistic environment with real-world satellite data, i.e., two line elements (TLEs), and the results confirm that our proposed algorithm outperforms the others in terms of routing success rate, reward convergence, and successful throughput.

The extensive use of smartphones and wearable devices has facilitated many useful applications. For example, with Global Positioning System (GPS)-equipped smart and wearable devices, many applications can gather, process, and share rich metadata, such as geolocation, trajectories, elevation, and time. For example, fitness applications, such as Runkeeper and Strava, utilize the information for activity tracking and have recently witnessed a boom in popularity. Those fitness tracker applications have their own web platforms and allow users to share activities on such platforms or even with other social network platforms. To preserve the privacy of users while allowing sharing, several of those platforms may allow users to disclose partial information, such as the elevation profile for an activity, which supposedly would not leak the location of the users. In this work, and as a cautionary tale, we create a proof of concept where we examine the extent to which elevation profiles can be used to predict the location of users. To tackle this problem, we devise three plausible threat settings under which the city or borough of the targets can be predicted. Those threat settings define the amount of information available to the adversary to launch the prediction attacks. Establishing that simple features of elevation profiles, e.g., spectral features, are insufficient, we devise both natural language processing (NLP)-inspired text-like representation and computer vision-inspired image-like representation of elevation profiles, and we convert the problem at hand into text and image classification problem. We use both traditional machine learning- and deep learning-based techniques and achieve a prediction success rate ranging from 59.59% to 99.80%. The findings are alarming, highlighting that sharing elevation information may have significant location privacy risks.

In order to build realistic metaverse systems, enabling high synchronization between physical-space and virtual meta-space is essentially required. For this purpose, this paper proposes a novel system-wide coordination algorithm for high synchronization under characteristics (i.e., highly realistic meta-space construction under the constraints of physical-space). The proposed algorithm consists of the following three stages. The first stage is quantum multi-agent reinforcement learning (QMARL)-based scheduling for low-delay temporal-synchronization using differentiated age-of-information (AoI) during data gathering in physical-space by observers for meta-space construction. This is beneficial for scalability according to action dimension reduction in reinforcement learning computation. The second stage is for creating virtual contents under delay constraints in meta-space based on the gathered data. When rendering regions that have received more user attention, avatar-popularity is considered for spatio-synchronization. Thus, a stabilized control mechanism is designed for time-average reality quality maximization for each region. The last stage is for caching based on avatar-popularity and AoI which can be helpful in constructing low-delay realistic meta-space. Furthermore, the concept of AoI is divided into two separate sub-concepts of physical AoI and virtual AoI such that the AoI in virtual meta-space can be thoroughly implemented.

This article aims to provide a hierarchical reinforcement learning (RL)-based solution to the automated drug infusion field. The learning policy is divided into the tasks of: 1) learning trajectory generative model and 2) planning policy model. The proposed deep infusion assistant policy gradient (DIAPG) model draws inspiration from adversarial autoencoders (AAEs) and learns latent representations of hypnotic depth trajectories. Given the trajectories drawn from the generative model, the planning policy infers a dose of propofol for stable sedation of a patient under total intravenous anesthesia (TIVA) using propofol and remifentanil. Through extensive evaluation, the DIAPG model can effectively stabilize bispectral index (BIS) and effect site concentration given a potentially time-varying target sequence. The proposed DIAPG shows an increased performance of 530% and 15% when a human expert and a standard reinforcement algorithm are used to infuse drugs, respectively.

This paper proposes a novel multi-agent deep reinforcement learning (MADRL)-based positioning algorithm for multiple unmanned aerial vehicles (UAVs) collaboration in mobile access applications where the UAVs work as mobile base stations. The primary objective of the proposed algorithm is to establish reliable mobile access networks for vehicle-to-everything (V2X) communications. This paper jointly considers energy-efficient UAV operation and reliable wireless communication services for realizing robust mobile access services. For the energy-efficient UAV operation, the reward function formulation of our proposed MADRL algorithm contains the features for UAV energy consumption models in order to realize efficient operations. Furthermore, for the reliable wireless communication services, the quality of service (QoS) requirements of individual users are considered as a part of reward function. Furthermore, this paper considers 60, GHz millimeter-wave (mmWave) mobile access for utilizing the benefits of i) ultra-wide-bandwidth for multi-Gbps high-speed communications and ii) high-directional communications for spatial reuse that is obviously good for avoiding interference among densely deployed users. Lastly, the comprehensive and data-intensive performance evaluation of the proposed MADRL-based algorithm for multi-UAV positioning is conducted. The results of these evaluations demonstrate that the proposed algorithm outperforms other existing algorithms.

According to the challenges related to the limited availability of quantum bits (qubits) in the era of noisy intermediate-scale quantum (NISQ), the immediate replacement of all components in existing network architectures with quantum computing devices may not be practical. As a result, implementing a hybrid quantum-classical system is regarded as one of effective strategies. In hybrid quantum-classical systems, quantum computing devices can be used for computation-intensive applications, such as massive scheduling in dynamic environments. Furthermore, one of most popular network applications is advanced social media services such as metaverse. Accordingly, this paper proposes an advanced multi-metaverse dynamic streaming algorithm in hybrid quantum-classical systems. For this purpose, the proposed algorithm consists of three stages. For the first stage, three-dimensional (3D) point cloud data gathering should be conducted using spatially scheduled observing devices from physical-spaces for constructing virtual multiple meta-spaces in metaverse server. This is for massive scheduling over dynamic situations, i.e., quantum multi-agent reinforcement learning-based scheduling is utilized for scheduling dimension reduction into a logarithmic-scale. For the second stage, a temporal low-delay metaverse server’s processor scheduler is designed for region-popularity-aware multiple virtual meta-spaces rendering contents allocation via modified bin-packing with hard real-time constraints. Lastly, a novel dynamic dynamic streaming algorithm is proposed for high-quality, differentiated, and stabilized meta-spaces rendering contents delivery to individual users via Lyapunov optimization theory. Our performance evaluation results verify that the proposed spatio-temporal algorithm outperforms benchmarks in various aspects over hybrid quantum-classical systems.

This paper investigates the framework of cross-entropy (CE) based clustering and beamforming for mmWave-non-orthogonal multiple access (NOMA) system taking into consideration the impact of imperfect successive interference cancellation (SIC). For the design of clustering and beamforming, we adopt CE based machine learning algorithm that has the objective to obtain the statistical parameters by minimizing the cross-entropy between optimal and sampling distributions. By using CE based clustering, the number of clusters can be adjusted to strike a balance between the inter-cluster interference and intra-cluster interference introduced by imperfect SIC. Furthermore, the inter-cluster interference induced by spatial beamforming is further reduced using CE based beamforming, which can significantly enhance the system performance of mmWave-NOMA. Based on the result, we compute the power allocation by dividing it into the intra-cluster and inter-cluster power allocation problems. In particular, we derive the optimal intra-cluster power allocation in a closed form and obtain the condition to guarantee the minimum rate requirements of all the users. We next solve the inter-cluster power allocation using convex optimization technique. Data-intensive simulation results illustrate that our proposed algorithm outperforms the conventional algorithm such as$K$-mean based clustering and the number of clusters can be controlled using CE based clustering algorithm.

Quantum convolutional neural network (QCNN) has just become as an emerging research topic as we experience the noisy intermediate-scale quantum (NISQ) era and beyond. As convolutional filters in QCNN extract intrinsic feature using quantum-based ansatz, it should use only finite number of qubits to prevent barren plateaus, and it introduces the lack of the feature information. In this paper, we propose a novel QCNN training algorithm to optimize feature extraction while using only a finite number of qubits, which is called fidelity-variation training (FV-Training).

Although quantum supremacy is yet to come, there has recently been an increasing interest in identifying the potential of quantum machine learning (QML) in the looming era of practical quantum computing. Motivated by this, in this article we re-design multi-agent reinforcement learning (MARL) based on the unique characteristics of quantum neural networks (QNNs) having two separate dimensions of trainable parameters: angle parameters affecting the output qubit states, and pole parameters associated with the output measurement basis. Exploiting this dyadic trainability as meta-learning capability, we propose quantum meta MARL (QM2ARL) that first applies angle training for meta-QNN learning, followed by pole training for few-shot or local-QNN training. To avoid overfitting, we develop an angle-to-pole regularization technique injecting noise into the pole domain during angle training. Furthermore, by exploiting the pole as the memory address of each trained QNN, we introduce the concept of pole memory allowing one to save and load trained QNNs using only two-parameter pole values. We theoretically prove the convergence of angle training under the angle-to-pole regularization, and by simulation corroborate the effectiveness of QM2ARL in achieving high reward and fast convergence, as well as of the pole memory in fast adaptation to a time-varying environment.

In recent years, quantum neural network (QNN) based on quantum computing has attracted attention due to its potential for computation-acceleration and parallelism. However, the intrinsic limitations of QNN, where the output (i.e., observables) can only be obtained through a measurement process, pose scalability challenges. Motivated by this, this paper aims to address the scalability challenges by incorporating Pauli-Z measurement and Basis measurement. In conventional frameworks, QNN typically relies on classical fully connected networks (FCNs) or increases the number of qubits to achieve large output dimensions. However, by leveraging our proposed framework, this paper successfully expands the output dimensions to an exponential scale, surpassing the limitations imposed by the limited number of qubits without relying on FCNs. Through extensive experiments, this paper demonstrates that the proposed framework outperforms existing QNN frameworks in multi-class classification tasks that require numerous output dimensions.

Recently, research on quantum neural network (QNN) architectures has been attracted in various fields. Among them, the distributed computation of QNN has been actively discussed for privacy-preserving information management due to data and model distribution over multiple computing devices. Based on this concept, this paper proposes quantum split learning (QSL) which splits a single QNN architecture across multiple distributed computing devices to avoid entire QNN architecture exposure. In order to realize QSL design, this paper also proposes cross-channel pooling, which utilizes quantum state tomography. Our evaluation results verifies that QSL preserves privacy in classification tasks and also improves accuracy at most by 6.83% compared to existing methods.

It has been considered that urban air mobility (UAM), also known as drone-taxi or electrical vertical takeoff and landing (eVTOL), will play a key role in future transportation. By putting UAM into practical future transportation, several benefits can be realized, i.e., (i) the total travel time of passengers can be reduced compared to traditional transportation and (ii) there is no environmental pollution and no special labor costs to operate the system because electric batteries will be used in UAM system. However, there are various dynamic and uncertain factors in the flight environment, i.e., passenger sudden service requests, battery discharge, and collision among UAMs. Therefore, this paper proposes a novel cooperative multiagent deep reinforcement learning (MADRL) algorithm based on centralized training and distributed execution (CTDE) concepts for reliable and efficient passenger delivery in UAM networks. According to the performance evaluation results, we confirm that the proposed algorithm outperforms other existing algorithms in terms of the number of serviced passengers increase (30%) and the waiting time per serviced passenger decrease (26% ).

This paper proposes an efficient quantum train engine (EQuaTE), a novel tool for quantum machine learning software which plots gradient variances to check whether our quantum neural network (QNN) falls into local minima (called barren plateaus in QNN). EQuaTE can be realized via dynamic analysis of the undetermined probabilistic qubit states. Furthermore, the proposed EQuaTE is capable of HCI-based visual feedback such that software engineers can recognize barren plateaus via visualization, allowing the modification of QNN based on this information.

This paper proposes a novel centralized training and distributed execution (CTDE)-based multi-agent deep reinforcement learning (MADRL) method for multiple unmanned aerial vehicles (UAVs) control in autonomous mobile access applications. For the purpose, a single neural network is utilized in centralized training for cooperation among multiple agents while maximizing the total quality of service (QoS) in mobile access applications.

The prodigious growth of digital health data has precipitated a mounting interest in harnessing machine learning methodologies, such as natural language processing (NLP), to scrutinize medical records, clinical notes, and other text-based health information. Although NLP techniques have exhibited substantial potential in augmenting patient care and informing clinical decision-making, data privacy and adherence to regulations persist as critical concerns. Federated learning (FL) emerges as a viable solution, empowering multiple organizations to train machine learning models collaboratively without disseminating raw data. This paper proffers a pragmatic approach to medical NLP by amalgamating FL, NLP models, and the NVFlare framework, developed by NVIDIA. We introduce two exemplary NLP models, the Long-Short Term Memory (LSTM)-based model and Bidirectional Encoder Representations from Transformers (BERT), which have demonstrated exceptional performance in comprehending context and semantics within medical data. This paper encompasses the development of an integrated framework that addresses data privacy and regulatory compliance challenges while maintaining elevated accuracy and performance, incorporating BERT pretraining, and comprehensively substantiating the efficacy of the proposed approach.

This article proposes a novel algorithm, named quantum multiagent actor–critic networks (QMACN) for autonomously constructing a robust mobile access system employing multiple unmanned aerial vehicles (UAVs). In the context of facilitating collaboration among multiple UAVs, the application of multiagent reinforcement learning (MARL) techniques is regarded as a promising approach. These methods enable UAVs to learn collectively, optimizing their actions within a shared environment, ultimately leading to more efficient cooperative behavior. Furthermore, the principles of quantum computing (QC) are employed in our study to enhance the training process and inference capabilities of the UAVs involved. By leveraging the unique computational advantages of QC, our approach aims to boost the overall effectiveness of the UAV system. However, employing a QC introduces scalability challenges due to the near intermediate-scale quantum (NISQ) limitation associated with qubit usage. The proposed algorithm addresses this issue by implementing a quantum centralized critic, effectively mitigating the constraints imposed by NISQ limitations. Additionally, the advantages of the QMACN with performance improvements in terms of training speed and wireless service quality are verified via various data-intensive evaluations. Furthermore, this article validates that a noise injection scheme can be used for handling environmental uncertainties in order to realize robust mobile access.

As one of the latest fields of interest in both academia and industry, quantum computing has garnered significant attention. Among various topics in quantum computing, variational quantum circuits (VQCs) have been noticed for their ability to carry out quantum deep reinforcement learning (QRL). This article verifies the potential of QRL, which will be further realized by implementing quantum multiagent reinforcement learning (QMARL) from QRL, especially for Internet-connected autonomous multirobot control and coordination in smart factory applications. However, the extension is not straightforward due to the nonstationarity of classical MARL. To cope with this, the centralized training and decentralized execution (CTDE) QMARL framework is proposed under the Internet connection. A smart factory environment with the Internet of Things (IoT)-based multiple agents is used to show the efficacy of the proposed algorithm. The simulation corroborates that the proposed QMARL-based autonomous multirobot control and coordination performs better than the other frameworks.

Guaranteeing real-time and accurate object detection simultaneously is paramount in autonomous driving environments. However, the existing object detection neural network systems are characterized by a tradeoff between computation time and accuracy, making it essential to optimize such a tradeoff. Fortunately, in many autonomous driving environments, images come in a continuous form, providing an opportunity to use optical flow. In this paper, we improve the performance of an object detection neural network utilizing optical flow estimation. In addition, we propose a Lyapunov optimization framework for time-average performance maximization subject to stability. It adaptively determines whether to use optical flow to suit the dynamic vehicle environment, thereby ensuring the vehicle’s queue stability and the time-average maximum performance simultaneously. To verify the key ideas, we conduct numerical experiments with various object detection neural networks and optical flow estimation networks. In addition, we demonstrate the self-configurable stabilized detection with YOLOv3-tiny and FlowNet2-S, which are the real-time object detection network and an optical flow estimation network, respectively. In the demonstration, our proposed framework improves the accuracy by 3.02%, the number of detected objects by 59.6%, and the queue stability for computing capabilities.

Federated learning (FL) is a key enabler for efficient communication and computing, leveraging devices’ distributed computing capabilities. However, applying FL in practice is challenging due to the local devices’ heterogeneous energy, wireless channel conditions, and non-independently and identically distributed (non-IID) data distributions. To cope with these issues, this paper proposes a novel learning framework by integrating FL and width-adjustable slimmable neural networks (SNN). Integrating FL with SNNs is challenging due to time-varying channel conditions and data distributions. In addition, existing multi-width SNN training algorithms are sensitive to the data distributions across devices, which makes SNN ill-suited for FL. Motivated by this, we propose a communication and energy-efficient SNN-based FL (namedSlimFL) that jointly utilizessuperposition coding (SC)for global model aggregation andsuperposition training (ST)for updating local models. By applying SC, SlimFL exchanges the superposition of multiple-width configurations decoded as many times as possible for a given communication throughput. Leveraging ST, SlimFL aligns the forward propagation of different width configurations while avoiding inter-width interference during backpropagation. We formally prove the convergence of SlimFL. The result reveals that SlimFL is not only communication-efficient but also deals with non-IID data distributions and poor channel conditions, which is also corroborated by data-intensive simulations.

In this paper, we propose a Gaussian Random Trajectory guided Hierarchical Reinforcement Learning (GRT-HL) method for autonomous furniture assembly. The furniture assembly problem is formulated as a comprehensive human-like long-horizon manipulation task that requires a long-term planning and a sophisticated control. Our proposed model, GRT-HL, draws inspirations from the semi-supervised adversarial autoencoders, and learns latent representations of the position trajectories of the end-effector. The high-level policy generates an optimal trajectory for furniture assembly, considering the structural limitations of the robotic agents. Given the trajectory drawn from the high-level policy, the low-level policy makes a plan and controls the end-effector. We first evaluate the performance of GRT-HL compared to the state-of-the-art reinforcement learning methods in furniture assembly tasks. We demonstrate that GRT-HL successfully solves the long-horizon problem with extremely sparse rewards by generating the trajectory for planning.

Augmented reality (AR) is one of emerging applications in modern multimedia systems research. Due to intensive time-consuming computations for AR visualization in mobile devices, quality-aware real-time computing under delay constraints is essentially required. Inspired by Lyapunov optimization framework, this paper proposes a time-average quality maximization method for the AR visualization under delay considerations.

We consider a scenario that utilizes road side units (RSUs) as distributed caches in connected vehicular networks. The goal of the use of caches in our scenario is for rapidly providing contents to connected vehicles under various traffic conditions. During this operation, due to the rapidly changed road environment and user mobility, the concept of age-of-information (AoI) is considered for (1) updating the cached information as well as (2) maintaining the freshness of cached information. The frequent updates of cached information maintain the freshness of the information at the expense of network resources. Here, the frequent updates increase the number of data transmissions between RSUs and MBS; and thus, it increases system costs, consequently. Therefore, the tradeoff exists between the AoI of cached information and the system costs. Based on this observation, the proposed algorithm in this paper aims at the system cost reduction which is fundamentally required for content delivery while minimizing the content AoI, based on Markov Decision Process (MDP) and Lyapunov optimization.

In recent years, quantum computing (QC) has been getting a lot of attention from industry and academia. Especially, among various QC research topics, variational quantum circuit (VQC) enables quantum deep reinforcement learning (QRL). Many studies of QRL have shown that the QRL is superior to the classical reinforcement learning (RL) methods under the constraints of the number of training parameters. This paper extends and demonstrates the QRL to quantum multi-agent RL (QMARL). However, the extension of QRL to QMARL is not straightforward due to the challenge of the noise intermediate-scale quantum (NISQ) and the non-stationary properties in classical multi-agent RL (MARL). Therefore, this paper proposes the centralized training and decentralized execution (CTDE) QMARL framework by designing novel VQCs for the framework to cope with these issues. To corroborate the QMARL framework, this paper conducts the QMARL demonstration in a single-hop environment where edge agents offload packets to clouds. The extensive demonstration shows that the proposed QMARL framework enhances 57.7% of total reward than classical frameworks.

This paper aims to integrate two synergetic technologies, federated learning (FL) and width-adjustable slimmable neural network (SNN) architectures. FL preserves data privacy by exchanging the locally trained models of mobile devices. By adopting SNNs as local models, FL can flexibly cope with the time-varying energy capacities of mobile devices. Combining FL and SNNs is however non-trivial, particularly under wireless connections with time-varying channel conditions. Furthermore, existing multi-width SNN training algorithms are sensitive to the data distributions across devices, so are ill-suited to FL. Motivated by this, we propose a communication and energy efficient SNN-based FL (named SlimFL) that jointly utilizes superposition coding (SC) for global model aggregation and superposition training (ST) for updating local models. By applying SC, SlimFL exchanges the superposition of multiple width configurations that are decoded as many as possible for a given communication throughput. Leveraging ST, SlimFL aligns the forward propagation of different width configurations, while avoiding the inter-width interference during back propagation. We formally prove the convergence of SlimFL. The result reveals that SlimFL is not only communication-efficient but also can counteract non-IID data distributions and poor channel conditions, which is also corroborated by simulations.

Recently, emerging media services over edge-assisted vehicular-to-infrastructure (V2I) networks are widely studied. A joint design of scheduling and streaming algorithms is essential for seamless and high-quality media services. In scheduling, the algorithm is designed under the consideration of heterogeneous computing at edges. In addition, the streaming algorithm is also proposed for maximizing the time-average streaming quality maximization subject to stability. It is verified that our proposed joint scheduling and streaming algorithm achieves desired performance improvements.

A mega-constellation of low-altitude earth orbit (LEO) satellites (SATs) are envisaged to provide a global coverage SAT network in beyond fifth-generation (5G) cellular systems. However, such wide coverage rather makes it difficult to apply existing multiple access protocols, such as random access channel (RACH). To overcome this issue, in this paper, we propose a novel random access solution for LEO SAT networks, called as S-RACH. In contrast to existing standardized protocols, S-RACH is a model-free approach using deep reinforcement learning (DRL). Compared to RACH, we show from various simulations that our proposed S-RACH yields around 2x lower average access delay.

This paper proposes a cooperative and dynamic quality adaptation scheme for delay-sensitive video streaming between the transmitter and the receiver. We present a novel adaptive super-resolution (SR) technique that adaptively controls the quality enhancement rate and computation time. Due to the capability of enhancing the quality of video chunks at the user device side, the transmitter can aggressively transcode video chunks to reduce the delivery latency and power consumption. Also, adaptive SR can control the tradeoff between playback stall rate and CPU consumption of the user device. Simulation results verify the performance of adaptive SR and show that the proposed video delivery scheme is very good to balance tradeoff among the following performance metrics for online video services: 1) playback stall rate, 2) average quality measure, 3) transmission power, and 4) CPU consumption of the user device.

CCTV-based surveillance using unmanned aerial vehicles (UAVs) is considered a key technology for security in smart city environments.This article creates a case where the UAVs with CCTV-cameras fly over the city area for flexible and reliable surveillance services. UAVs should be deployed to cover a large area while minimizing overlapping and shadow areas for a reliable surveillance system. However, the operation of UAVs is subject to high uncertainty, necessitating autonomous recovery systems. This article develops a multiagent deep reinforcement learning-based management scheme for reliable industry surveillance in smart city applications. The core idea this article employs is autonomously replenishing the UAV's deficient network requirements with communications. Via intensive simulations, our proposed algorithm outperforms the state-of-the-art algorithms in terms of surveillance coverage, user support capability, and computational costs.

We present${\sf Supremo}$, a cloud-assisted system for low-latency image super-resolution (SR) in mobile devices. As SR is extremely compute-intensive, we first further optimize state-of-the-art DNN to reduce the inference latency. Furthermore, we design a mobile-cloud cooperative execution pipeline composed of specialized data compression algorithms to minimize end-to-end latency with minimal image quality degradation. Finally, we extend${\sf Supremo}$to video applications by formulating a dynamic optimal control algorithm to design${\sf Supremo-Opt}$, which aims to maximize the impact of SR while satisfying latency and resource constraints under practical network conditions.${\sf Supremo}$upscales 360p image to 1080p in 122 ms, which is 43.68× faster than on-device GPU execution. Compared to cloud offloading-based solutions,${\sf Supremo}$reduces wireless network bandwidth consumption and end-to-end latency by 15.23× and 4.85× compared to baseline approach of sending and receiving whole images, and achieves 2.39 dB higher PSNR compared to using conventional JPEG to achieve similar data size compression. Furthermore,${\sf Supremo-Opt}$guarantees robust performance in practical scenarios.

In real-time strategy (RTS) game artificial intelligence research, various multi-agent deep reinforcement learning (MADRL) algorithms are widely and actively used nowadays. Most of the research is based on StarCraft II environment because it is the most well-known RTS games in world-wide. In our proposed MADRL-based algorithm, distributed MADRL is fundamentally used that is called QMIX. In addition to QMIX-based distributed computation, we consider state categorization which is a novel preprocessing method for representation of graph attention. Furthermore, self-attention mechanisms are used for identifying the relationship among agents in the form of graphs. Based on these approaches, we propose a categorized state graph attention policy (CSGA-policy). As observed in the performance evaluation of our proposed CSGA-policy with the most well-known StarCraft II simulation environment, our proposed algorithm works well in various settings, as expected.

Machine learning (ML) is a promising enabler for the fifth-generation (5G) communication systems and beyond. By imbuing intelligence into the network edge, edge nodes can proactively carry out decision-making and, thereby, react to local environmental changes and disturbances while experiencing zero communication latency. To achieve this goal, it is essential to cater for high ML inference accuracy at scale under the time-varying channel and network dynamics, by continuously exchanging fresh data and ML model updates in a distributed way. Taming this new kind of data traffic boils down to improving the communication efficiency of distributed learning by optimizing communication payload types, transmission techniques, and scheduling, as well as ML architectures, algorithms, and data processing methods. To this end, this article aims to provide a holistic overview of relevant communication and ML principles and, thereby, present communication-efficient and distributed learning frameworks with selected use cases.

This paper comprehensively studies a content-centric mobile network based on a preference learning framework, where each mobile user is equipped with a finite-size cache. We consider a practical scenario where each user requests a content file according to its own preferences, which is motivated by the existence of heterogeneity in file preferences among different users. Under our model, we consider a single-hop-based device-to-device (D2D) content delivery protocol and characterize the average hit ratio for the following two file preference cases: the personalized file preferences and the common file preferences. By assuming that the model parameters such as user activity levels, user file preferences, and file popularity are unknown and thus need to be inferred, we present a collaborative filtering (CF)-based approach to learn these parameters. Then, we reformulate the hit ratio maximization problems into a submodular function maximization and propose two computationally efficient algorithms including a greedy approach to efficiently solve the cache allocation problems. We analyze the computational complexity of each algorithm. Moreover, we analyze the corresponding level of the approximation that our greedy algorithm can achieve compared to the optimal solution. Using a real-world dataset, we demonstrate that the proposed framework employing the personalized file preferences brings substantial gains over its counterpart for various system parameters.

Wireless caching networks have been extensively researched as a promising technique for supporting the massive data traffic of multimedia services. Many of the existing studies on real-data traffic have shown that users of a multimedia service consecutively request multiple contents and this sequence is strongly dependent on the related list of the first content and/or the top referrer in the category. This paper thus introduces the notion of “temporary preference”, characterizing the behavior of users who are highly likely to request the next content from a certain target category (i.e., related content list). Based on this observation, this paper proposes both probabilistic caching and dynamic delivery policies for categorized contents and consecutive user demands. The proposed caching scheme maximizes the minimum of the cache hit rates for all users. In the delivery phase, a dynamic helper association policy for receiving multiple contents in a row is designed to reduce the delivery latency. By comparing with the content placement optimized for one-shot requests, numerical results verify the effects of categorized contents and consecutive user demands on the proposed caching and delivery policies.

This paper proposes a user scheduling and power allocation method for content delivery in wireless caching helper networks without any stringent constraint on the interference model. For supporting delay-sensitive and time-varying user demands, the actual delivery quantity of the requested content should be dynamically controlled by advanced scheduling and power allocation. In addition, it is difficult for a central unit to control the content delivery due to a lack of knowledge of the entire time-varying network; therefore, a belief-propagation (BP)-based algorithm that facilitates distributed decisions on user scheduling and power allocation at every caching helper is presented. The proposed delivery scheme maximizes power efficiency while limiting the average delay of user request satisfactions by managing interference among users well. Simulation results show that the proposed scheme provides almost the same delay performance as the exhaustively found optimal one at the expense of little power consumption.

The extensive use of smartphones and wearable devices has facilitated many useful applications. For example, with Global Positioning System (GPS)-equipped smart and wearable devices, many applications can gather, process, and share rich metadata, such as geolocation, trajectories, elevation, and time. For example, fitness applications, such as Runkeeper and Strava, utilize information for activity tracking, and have recently witnessed a boom in popularity. Those fitness tracker applications have their own web platforms, and allow users to share activities on such platforms, or even with other social network platforms. To preserve privacy of users while allowing sharing, several of those platforms may allow users to disclose partial information, such as the elevation profile for an activity, which supposedly would not leak the location of the users. In this work, and as a cautionary tale, we create a proof of concept where we examine the extent to which elevation profiles can be used to predict the location of users. To tackle this problem, we devise three plausible threat settings under which the city or borough of the targets can be predicted. Those threat settings define the amount of information available to the adversary to launch the prediction attacks. Establishing that simple features of elevation profiles, e.g., spectral features, are insufficient, we devise both natural language processing (NLP)-inspired text-like representation and computer vision-inspired image-like representation of elevation profiles, and we convert the problem at hand into text and image classification problem. We use both traditional machine learning-and deep learning-based techniques, and achieve a prediction success rate ranging from 59.59% to 95.83%. The findings are alarming, and highlight that sharing elevation information may have significant location privacy risks.

Generating images from sound is a challenging task. This paper proposes a novel deep learning model that generates bird images from their corresponding sound information. Our proposed model includes a sound encoder in order to extract suitable feature representations from audio recordings, and then it generates bird images that corresponds to its calls using conditional generative adversarial networks (cGANs) with auxiliary classifiers. We demonstrate that our model produces better image generation results which outperforms other state-of-the-art methods in a similar context.

In wireless caching networks, a user generally has a concrete purpose of consuming contents in a certain preferred category, and requests multiple contents in sequence. While most existing research on wireless caching and delivery has focused only on one-shot requests, the popularity distribution of contents requested consecutively is definitely different from the one-shot request and has been not considered. Also, especially from the perspective of the service provider, it is advantageous for users to consume as many contents as possible. Thus, this paper proposes two cache allocation policies for categorized contents and consecutive user demands, which maximize 1) the cache hit rate and 2) the number of consecutive content consumption, respectively. Numerical results show how categorized contents and consecutive content requests have impacts on the cache allocation.

This article proposes a novel multiagent deep reinforcement learning-based algorithm which can realize federated learning (FL) computation with Internet-of-Underwater-Things (IoUT) devices in the ocean environment. According to the fact that underwater networks are relatively not easy to set up reliable links by huge fading compared to wireless free-space air medium, gathering all training data for conducting centralized deep learning training is not easy. Therefore, FL-based distributed deep learning can be a suitable solution for this application. In this IoUT network (IoUT-Net) scenario, the FL system needs to construct a global learning model by aggregating the local model parameters that are obtained from individual IoUT devices. In order to reliably deliver the parameters from IoUT devices to a centralized FL machine, base station like devices are needed. Therefore, a joint cell association and resource allocation (JCARA) method is required and it is designed inspired by multiagent deep deterministic policy gradient (MADDPG) to deal with distributed situations and unexpected time-varying states. The performance evaluation results show that our proposed MADDPG-based algorithm achieves 80% and 41% performance improvements than the standard actor–critic and DDPG, respectively, in terms of the downlink throughput.

This article proposes a novel multiagent deep reinforcement learning method for the energy management of distributed electric vehicle charging stations with a solar photovoltaic system and energy storage system. In the literature, the conventional method is to calculate the optimal electric vehicle charging schedule in a centralized manner. However, in general, the centralized approach is not realistic under certain environments where the system operators for multiple electric vehicle charging stations handle dynamically varying data, such as the status of the energy storage system and electric vehicle-related information. Therefore, this article proposes a method that can compute the scheduling solutions of multiple electric vehicle charging stations in a distributed manner while handling run-time time-varying dynamic data. As shown in the data-intensive performance evaluation, it can be observed that the proposed method achieves a desirable performance in terms of reducing the operation costs of electric vehicle charging stations.

In wireless caching networks, the design of the content delivery method must consider random user requests, caching states, network topology, and interference management. In this article, we establish a general framework for content delivery in wireless caching networks without stringent assumptions that restrict the network structure and interference model. Based on the framework, we propose a dynamic and distributed link scheduling and power allocation scheme for content delivery that is assisted by belief-propagation (BP) algorithms. The proposed scheme achieves three critical purposes of wireless caching networks: 1) limiting the delay of user request satisfactions, 2) maintaining the power efficiency of caching nodes, and 3) managing interference among users. In addition, we address the intrinsic problem of the BP algorithm in our network model, proposing a matching algorithm for one-to-one link scheduling. Simulation results show that the proposed scheme provides almost the same delay performance as the optimal scheme found through an exhaustive search at the expense of a little additional power consumption and does not require a clustering method and orthogonal resources in a large-scale D2D network.

Millimeter-wave (mmWave) base station can offer abundant high capacity channel resources toward connected vehicles so that quality-of-service (QoS) of them in terms of downlink throughput can be highly improved. The mmWave base station can operate among existing base stations (e.g., macro- cell base station) on non-overlapped channels among them and the vehicles can make decision what base station to associate, and what channel to utilize on heterogeneous networks. Furthermore, because of the non-omni property of mmWave communication, the vehicles decide how to align the beam direction toward mmWave base station to associate with it. However, such joint problem requires high computational cost, which is NP-hard and has combinatorial features. In this paper, we solve the problem in 3-tier heterogeneous vehicular network (HetVNet) with multi-agent deep reinforcement learning (DRL) in a way that maximizes expected total reward (i.e., downlink throughput) of vehicles. The multi-agent deep deterministic policy gradient (MADDPG) approach is introduced to achieve optimal policy in continuous action domain.

In wireless caching networks, each user generally consumes more than one content in a row, and the number of consecutive demands could vary for different users. In addition, popular contents are usually classified into several categories. In this case for consecutive user demands, the content popularity model largely depends on the previously consumed contents, i.e., contents that belong to the same category as the previously consumed content would be highly popular. Based on this observation, this paper proposes an optimal probabilistic caching policy for consecutive user demands in categorized contents. The proposed caching scheme maximizes the minimum of the success probabilities for content delivery of all users when individual users request different numbers of contents in a row. Comparing with the content placement optimized for one-shot request, intensive numerical results verify the impacts of categorized contents and consecutive user demands on the caching policy.

With the evolution of various advanced driver assistance system (ADAS) platforms, the design of autonomous driving system is becoming more complex and safety-critical. The autonomous driving system simultaneously activates multiple ADAS functions; and thus it is essential to coordinate various ADAS functions. This paper proposes a randomized adversarial imitation learning (RAIL) method that imitates the coordination of autonomous vehicle equipped with advanced sensors. The RAIL policies are trained through derivative-free optimization for the decision maker that coordinates the proper ADAS functions, e.g., smart cruise control and lane keeping system. Especially, the proposed method is also able to deal with the LIDAR data and makes decisions in complex multi-lane highways and multi-agent environments.

Deep learning techniques not only have surpassed humans in several computer vision tasks, but also have achieved state-of-the-art performance on the super-resolution (SR) task, which enhances the resolution of reconstructed images from the observed low-resolution (LR) images. Very Deep Super- Resolution (VDSR) is a popular architecture with decent performance on the SR task. In general, the deeper the VDSR network, the better the performance. Say, on a computing device with limited computational resources, the SR task must support multiple resolutions or multiple output rates. For this multi-rate SR task, the state-of-the-art architectures require multiple networks with varying depths. However as the number of supported rates increases, so does the number of networks imposing incremental computational burden on the computing device. We propose a depth-controllable network and training principles for the multi-rate SR task. The proposed network is configured as a single network regardless of the number of supported rates. The inverse auxiliary loss and contiguous/progressive skip connections are presented to train the network end-to-end throughout the varying number of layers without biasing the performances of specific depths. With three data sets and three scaling factors, the proposed network is compared to the baseline network VDSR, along with the Super-Resolution Convolutional Neural Network (SRCNN). Our network not only requires a single network at varying rates, but also performs as well as the baseline networks in terms of Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity (SSIM). Our model outperforms the baseline networks when the depth of layers is more than eleven.

Developing agents that can perform challenging complex tasks is the goal of reinforcement learning. The model-free reinforcement learning has been considered as a feasible solution. However, the state of the art research has been to develop increasingly complicated techniques. This increasing complexity makes the reconstruction difficult. Furthermore, the problem of reward dependency is still exists. As a result, research on imitation learning, which learns policy from a demonstration of experts, has begun to attract attention. Imitation learning directly learns policy based on data on the behavior of the experts without the explicit reward signal provided by the environment. However, imitation learning tries to optimize policies based on deep reinforcement learning such as trust region policy optimization. As a result, deep reinforcement learning based imitation learning also poses a crisis of reproducibility. The issue of complex model-free model has received considerable critical attention. A derivative-free optimization based reinforcement learning and the simplification on policies obtain competitive performance on the dynamic complex tasks. The simplified policies and derivative free methods make algorithm be simple. The reconfiguration of research demo becomes easy. In this paper, we propose an imitation learning method that takes advantage of the derivative-free optimization with simple linear policies. The proposed method performs simple random search in the parameter space of policies and shows computational efficiency. Experiments in this paper show that the proposed model, without a direct reward signal from the environment, obtains competitive performance on the MuJoCo locomotion tasks.

This demo abstract introduces a new light-weight programming language koa which is suitable for blockchain system design and implementation. In this abstract, the basic features of koa are introduced including working system (with playground), architecture, and virtual machine operations. Rum-time execution of software implemented by koa will be presented during the session.

The vehicles take a role of mobile devices so that resource management issue among them in cellular networks is getting highlighted. In addition, the millimeter-wave (mmWave) base station is expected to be included in existing heterogeneous networks. In this regard, we present a multi-agent deep reinforcement learning (MADRL) based resource allocation strategy for mobile devices in heterogeneous vehicular networks, which suffer from shortage issue of shared frequency band. The downlink throughput of each mobile device is cooperatively enhanced by the proposed MADRL method, and thus each mobile device associates with a base station and uses a frequency band within low delay.

With the rapid evolution of electric systems, there has been a significant demand for energy Internet (EI) systems that allow sustainable and environmentally friendly energy management. Several research efforts regarding EI systems have been aimed at providing reliable, efficient, and cost-effective techniques. In this paper, we propose a novel algorithm and system for real-time electricity pricing and scheduling. Our algorithm consists of a two-stage operation. The first stage performs real-time pricing to determine the maximum electricity consumption while the second stage performs Internet of Things (IoT) device scheduling. In the second stage, the optimization framework for scheduling is modeled as a 0–1 Knapsack problem; therefore, the solutions to the optimization problem are computed using a dynamic programming framework. Through intensive simulations with well-defined parameters, it is verified that the proposed scheme provides several features, especially reductions in electricity bills with the appropriate parameter settings.

Temperature effect inversion (TEI) phenomenon in ultralow power (ULP) very large scale integration circuits has been identified as an important effect by both academia and industry. Although a number of ULP methods that attempt to exploit the TEI phenomenon have been proposed, the small size of the design exploration space when applying these methods to ULP circuits hinders them from achieving their full potential. This is mainly due to the limited granularity of the supply voltage level control. Starting with an intuition that the body biasing (BB) technique is a key to overcome this limitation, this paper exploits the BB technique along with the TEI-aware voltage scaling (TEI-VS) method and TEI-aware frequency scaling (TEI-FS) method, so as to substantially increase the design spaces of these methods. Techniques for optimally combining the BB technique with TEI-VS and TEI-FS are introduced. Simulation results with the latest commercial CMOS process technologies for ULP designs demonstrate the effectiveness of the proposed methodology.

Exploiting both LTE and Wi-Fi links simultaneously enhances the performance of video streaming services in a smartphone. However, it is challenging to achieve seamless and high quality video while saving battery energy and LTE data usage to prolong the usage time of a smartphone. In this paper, we propose REQUEST, a video chunk request policy for Dynamic Adaptive Streaming over HTTP (DASH) in a smartphone, which can utilize both LTE and Wi-Fi. REQUEST enables seamless DASH video streaming with near optimal video quality under given budgets of battery energy and LTE data usage. Through extensive simulation and measurement in a real environment, we demonstrate that REQUEST significantly outperforms other existing schemes in terms of average video bitrate, rebuffering, and resource waste.

In this paper, we propose a joint dynamic power control and user pairing algorithm for power-efficient and delay-constrained hybrid multiple access systems. In a hybrid multiple access system, user pairing determines whether the transmitter serves as a certain user by orthogonal multiple access (OMA) or non-orthogonal multiple access (NOMA). The proposed optimization framework minimizes the long-term time-average transmit power expenditure while reducing the queuing delay and guaranteeing the minimum time-average data rates. The proposed technique observes both channel and queue state information and adjusts queue backlogs to avoid an excessive queueing delay by appropriate user pairing and power allocation. Furthermore, the flexible use of resources is captured in the proposed algorithm by employing NOMA. The data-intensive simulation results show that the proposed scheme for power allocation and user scheduling achieves a balance among multiple performance goals, i.e., power efficiency, queueing delay, and data rate.

This paper proposes a video delivery strategy for dynamic streaming services which maximizes time-average streaming quality under a playback delay constraint in wireless caching networks. The network where popular videos encoded by scalable video coding are already stored in randomly distributed caching nodes is considered under adaptive video streaming concepts, and distance-based interference management is investigated in this paper. In this network model, a streaming user makes delay-constrained decisions depending on stochastic network states: 1) caching node for video delivery, 2) video quality, and 3) the quantity of video chunks to receive. Since wireless link activation for video delivery may introduce delays, different timescales for updating caching node association, video quality adaptation, and chunk amounts are considered. After associating with a caching node for video delivery, the streaming user chooses combinations of quality and chunk amounts in the small timescale. The dynamic decision making process for video quality and chunk amounts at each slot is modeled using Markov decision process, and the caching node decision is made based on the framework of Lyapunov optimization. Our intensive simulations verify that the proposed video delivery algorithm works reliably and also can control the tradeoff between video quality and playback latency.

As interest in cryptocurrency has increased, problems have arisen with Proof-of-Work (PoW) and Proof-of-Stake (PoS) methods, the most representative methods of acquiring cryptocurrency in a blockchain. The PoW method is uneconomical and the PoS method can be easily monopolized by a few people. To cope with this issue, this paper introduces a Proof-of-Probability (PoP) method. The PoP is a method where each node sorts the encrypted actual hash as well as a number of fake hash, and then the first node to decrypt actual hash creates block. In addition, a wait time is used when decrypting one hash and then decrypting the next hash for restricting the excessive computing power competition. In addition, the centralization by validaters with many stakes can be avoided in the proposed PoP method.

This paper proposes a reliable and self-adaptive operating system management policy for CCTV-based security applications which controls arrival image compression rates. After receiving image sequences via CCTV cameras, the system enqueues the sequences of images and processes them for face recognition. High compression rates in CCTV-recorded images provide low recognition performance due to quantization while it is beneficial in terms of queue stability. On the other hand, low compression rates in the images provide high recognition performance while it may introduce queue overflows. Therefore, this paper designs a queue-aware, self-adaptive, and reliable operating system management scheme which aims at face identification performance maximization while avoiding queue overflow by controlling CCTV-recorded image compression rates based on the theory of Lyapunov optimization.

One of the reasons behind the tremendous success of deep learning theory and applications in the recent days is advances in distributed and parallel high performance computing (HPC). This paper proposes a new distributed deep learning platform, named ShmCaffe, which utilizes remote shared memory for communication overhead reduction in massive deep neural network training parameter sharing. ShmCaffe is designed based on Soft Memory Box (SMB), a virtual shared memory framework. In the SMB framework, the remote shared memory is used as a shared buffer for asynchronous massive parameter sharing among many distributed deep learning processes. Moreover, a hybrid method that combines asynchronous and synchronous parameter sharing methods is also discussed in this paper for improving scalability. As a result, ShmCaffe is 10.1 times faster than Caffe and 2.8 times faster than Caffe-MPI for deep neural network training when Inception_v1 is trained with 16 GPUs. We verify the convergence of the Inception_v1 model training using ShmCaffe-A and ShmCaffe-H by varying the number of workers. Furthermore, we evaluate scalability of ShmCaffe by analyzing the computation and communication times per one iteration of deep learning training in four convolutional neural network (CNN) models.

This paper presents a new approach for avoiding double-spending attacks via the concept of recipient-oriented concepts in private blockchain networks. Together with the concepts of stealth address and masternode at recipient sides, transactional privacy is assured and transaction recipients become active which receive transaction propagation unilaterally. Based on this approach, it can be shown that recipient-oriented method is suggested for preventing the double-spending attacks in this paper. In addition, the proposed method is possible to prevent double spending attacks using the verification time of the recipient and blocking time of the transaction.

This paper proposes a low-complexity online model adaptation algorithm which dynamically selects an object detection algorithm among given/implemented algorithms in the system depending on workload-backlog. As well-studied in literature, there exists tradeoff between object detection accuracy and computation time (i.e., delay) because highly accurate algorithms generally take more time due to complicated deep neural network architectures. In our proposed algorithm, the accuracy is reformulated as reward; and the delay is modeled with queue. Based on this queue-based model, Lyapunov control inspired stochastic optimization is utilized for designing time-average reward maximization subject to stability in real-time object detection deep learning platforms. Moreover, our proposed algorithm solves closed-form equation in each model selection interval, thus the proposed algorithm takes low computational complexity. The performance of our proposed algorithm is evaluated via data-intensive real-world implementations under heavy workloads; and is verified that our proposed algorithm works as desired.

In industrial Internet of Things (IIoT) applications for smart manufacturing system, efficient allocation of the carrier and computing resources is crucial. However, existing resource assignment schemes in smart manufacturing system cannot provide timely provision of resources to the inherently dynamic and bursty user demands. To reflect real-time supply and demand for smart manufacturing resources, several research results on auction-style resource assignments have been introduced; however, security, privacy, and trust computing related issues are not actively discussed in the results. The resources should be assigned to devices according to the system policy, which depends on the information provided by IIoT devices. If there are any resource demanding devices, they can report manipulated malicious information for their own interest to obtain more resources. That is, the smart manufacturing system may be vulnerable due to selfish smart manufacturing devices’ behaviors. This reduces the efficiency of the entire system and moreover ceases the plant-wide process. While many research contributions related to the trust computing aim at detecting malicious nodes, this paper presents a novel view of trust computing by showing why devices inside the smart manufacturing system have to act honestly. In this paper, a Vickrey–Clarke–Groves auction-based hierarchical trust computing algorithm is proposed for: 1) computing carrier resources required for wireless communication between IIoT devices and gateways and 2) distributing CPU resources for processing data at central processing controller. Last, simulation results demonstrate that the utilities of each participant are maximized when the IIoT devices and gateways are trustful.

The Internet of Things (IoT) is expected to play an important role in the construction of next generation mobile communication services, and is currently used in various services. However, the power-hungry battery significantly limits the lifetime of IoT devices. Among the various lifetime extension techniques, this paper discusses mobile charging, which enables wireless power transfer based on radio frequency with mobile chargers (MCs). MCs function as traveling target IoT networks that provide energy to battery-operated IoT devices. However, MCs with an energy-constrained battery result in limitation of travel-time. This paper formulates a problem to minimize energy consumption for charging IoT devices by determining the path of motion of an MC and efficient charging points, and proves that the problem is NP-hard. An efficient algorithm, named best charging efficiency (BCE), is proposed to solve the problem and the upper bound of the BCE algorithm is guaranteed using the duality of linear programming. In addition, an improved BCE algorithm called branching second best efficiency algorithm with additional searching techniques is introduced. Finally, this paper analyzes the difference in performance among the proposed algorithms, optimal solutions, and the existing algorithm and concludes that the performance of the proposed algorithm is near optimal, within 1% of difference ratio in terms of charging efficiency and delay.

This paper considers one-hop device-to-device-assisted wireless caching networks that cache video files of varying quality levels, with the assumption that the base station can control the video quality but cache-enabled devices cannot. Two problems arise in such a caching network: file placement problem and node association problem. This paper suggests a method to cache videos of different qualities, and thus of varying file sizes, by maximizing the sum of video quality measures that users can enjoy. There exists an interesting tradeoff between video quality and video diversity, i.e., the ability to provision diverse video files. By caching high-quality files, the cache-enabled devices can provide high-quality video, but cannot cache a variety of files. Conversely, when the device caches various files, it cannot provide a good quality for file-requesting users. In addition, when multiple devices cache the same file but their qualities are different, advanced node association is required for file delivery. This paper proposes a node association algorithm that maximizes time-averaged video quality for multiple users under a playback delay constraint. In this algorithm, we also consider request collision, the situation where several users request files from the same device at the same time, and we propose two ways to cope with the collision: scheduling of one user and non-orthogonal multiple access. Simulation results verify that the proposed caching method and the node association algorithm work reliably.

The next-generation mobile network anticipates integrated heterogeneous fronthaul and backhaul technologies referred to as a unified crosshaul architecture. The crosshaul enables a flexible and cost-efficient infrastructure for handling mobile data tsunami from dense Internet of things (IoT). However, stabilization, energy efficiency, and latency have not been jointly considered in the optimization of crosshaul performance. To overcome these issues, we propose an orchestration scheme referred to as the stabilized green crosshaul orchestration (SGCO). SGCO utilizes a Lyapunov-theory-based drift-plus-penalty policy to determine the optimal amount of offloaded data that should be processed either at the eastbound or westbound computing platforms to minimize energy consumption. To achieve system stability, the cache buffer is considered as the main constraint in developing the optimization process. Moreover, the amount of offloaded data transmitted via crosshaul links is selected by adopting the binary min-knapsack problem. Accordingly, a lightweight heuristic algorithm is proposed. As the cache buffer is stabilized and the computations are controlled, the SGCO ensures adjustable computing latency threshold for various IoT services. The performance analysis shows that the proposed SGCO scheme exposes effective energy consumption compared to other existing schemes while maintaining system stability considering latency.

Exploiting both LTE and Wi-Fi links simultaneously enhances the performance of video streaming services in a smartphone. However, it is challenging to achieve seamless and high quality video while saving battery energy and LTE data usage to prolong the usage time of a smartphone. In this paper, we propose REQUEST, a video chunk request policy for Dynamic Adaptive Streaming over HTTP (DASH) in a smartphone, which can utilize both LTE and Wi-Fi. REQUEST enables seamless DASH video streaming with near optimal video quality under given budgets of battery energy and LTE data usage. Through extensive simulation and measurement in a real environment, we demonstrate that REQUEST significantly outperforms other existing schemes in terms of average video bitrate, rebuffering, and resource waste.

This paper conducts the feasibility study for discussing about the interference impacts in advanced fog computing (FC) networks with 60 GHz millimeter-wave wireless technology. With the concept of FC, cloud computing services can be utilized with local computing devices (FC devices) those are located nearby user terminals [edge devices (EDs)] for low-latency communications. Since EDs can be densely deployed in urban areas, wireless hyperconnection should be supportable in FC networks. For the system, 60 GHz millimeter-wave technology can be used as one of wireless networking methods because it can provide: 1) high rates for visual information delivery and 2) high directionality for spatial reuse in hyperconnected FC networks. Therefore, this performance simulation study under the consideration of interference impacts determines whether utilizing 60 GHz millimeter-wave wireless technology for hyperconnected FC networks is feasible or not. Under the consideration of various interference scenarios, the interference impacts are calculated. As presented in simulation-based performance evaluation results, we verify that 1.5 Gb/s high rate can be supportable even though more than 1000 EDs are deployed in the 100 m-by-100 m size small-scale networks (upper bound). Therefore, we confirm that hyperconnection can be realized with 60 GHz millimeter-wave technology for FC networks.

With the current state of development in demand response (DR) programs in smart grid systems, there have been great demands for automated energy scheduling for residential customers. Recently, energy scheduling in smart grids have focused on the minimization of electricity bills, the reduction of the peak demand, and the maximization of user convenience. Thus, a user convenience model is proposed under the consideration of user waiting times, which is a nonconvex problem. Therefore, the nonconvex is reformulated as convex to guarantee optimal solutions. Moreover, mathematical formulations for DR optimization are derived based on the reformulated convex problem. In addition, two types of pricing policies for electricity bills are designed in the mathematical formulations, i.e., real-time pricing policy and progressive policy. With real-time pricing policy, convexity is guaranteed whereas progressive policy cannot. Then, heuristic algorithms are finally designed for obtaining approximated optimal solutions in progressive policy.

The Invisible Internet Project (I2P) is an overlay network that provides secure and anonymous communication channels. EepSites are the anonymous websites hosted in the I2P network. To access the eepSites, DNS requests of a domain name suffixed with the {\sf .i2p} pseudo top-level domain (TLD) are routed within the I2P network. However, not only that {\sf .i2p} queries are leaking in the public DNS infrastructure, but also such leakage has various plausible root causes and implications that are different from other related leakage. In this paper, we analyze the leaked {\sf .i2p} requests captured in the A and J root name servers of the public DNS, showing that a large number of queries are observed and outlining various potential directions of addressing such leakage.

On-demand video streaming is becoming a killer application for wireless networks. Recent information-theoretic results have shown that a combination of caching on the users' devices and device-to-device (D2D) communications yields throughput scalability for very dense networks, which represent critical bottlenecks for conventional cellular and wireless local area network (WLAN) technologies. In this paper, we consider the implementation of such caching D2D systems where each device pre-caches a subset of video files from a library, and users requesting a file that is not already in their library obtain it from neighboring devices through D2D communication. We develop centralized and distributed algorithms for the delivery phase, encompassing a link scheduling and a streaming component. The centralized scheduling is based on the max-weighted independent set (MWIS) principle and uses message-passing to determine max-weight independent sets. The distributed scheduling is based on a variant of the FlashLinQ link scheduling algorithm, enhanced by introducing video-streaming specific weights. In both cases, the streaming component is based on a quality-aware stochastic optimization approach, reminiscent of current Dynamic Adaptive Streaming over HTTP (DASH) technology, for which users sequentially request video “chunks” by choosing adaptively their quality level. The streaming and the scheduling components are coupled by the length of the users' request queues. Through extensive system simulation, the proposed approaches are shown to provide sizeable gains with respect to baseline schemes formed by the concatenation of off-the-shelf FlashLinQ with proportional fair link scheduling and DASH at the application layer.

This paper proposes a polynomial-time algorithm for energy-efficient dynamic packet downloading from medical cloud storage to medical Internet-of-Things (IoT) devices. The medical cloud can distribute its own medical data to medical IoT devices via access points. Therefore, network disconnection can happen between the medical cloud and medical IoT devices when power/energy management in each access point is not efficient. This situation is especially harmful in in-hospital network architectures, because the architecture usually has strict requirements in terms of reliability. Therefore, this paper proposes a dynamic energy-efficient algorithm, which computes the amount of power allocation in each access point based on the buffer backlog size and channel states under the consideration of buffer stability. With the proposed adaptive algorithm, each access point calibrates its own parameters for more adaptive power/energy management. The performance of the proposed algorithm is evaluated in terms of network lifetime, and it is observed that the proposed algorithm achieves the desired performance.

This paper proposes two novel algorithms for adaptive crowdsourcing in 60-GHz medical imaging big-data platforms, namely, a max-weight scheduling algorithm for medical cloud platforms and a stochastic decision-making algorithm for distributed power-and-latency-aware dynamic buffer management in medical devices. In the first algorithm, medical cloud platforms perform a joint queue-backlog and rate-aware scheduling decisions for matching deployed access points (APs) and medical users where APs are eventually connected to medical clouds. In the second algorithm, each scheduled medical device computes the amounts of power allocation to upload its own medical data to medical big-data clouds with stochastic decision making considering joint energy-efficiency and buffer stability optimization. Through extensive simulations, the proposed algorithms are shown to achieve the desired results.

Millimeter-wave (mm-wave) transmission is a promising method for increasing capacity in next-generation cellular systems. However, since mm-wave signals are too weak to (i) do long-distance communication and (ii) survive in non-line-of-sight situations, multi-hop relaying is required. We thus study in this paper a multi-hop routing protocol at mm-wave (38 GHz and 28GHz) frequencies. Due to the great and steadily increasing importance of video streaming in cellular applications, we pay special attention to the maximization of video transmission quality, which makes the optimization problem different from the usual sum-rate maximization. Specifically, we develop a protocol that optimizes the “sum quality” of the transmission of different video streams, subject to constraints (minimum quality) of separate streams. We take the unique properties of mm-waves (sparse angular power spectrum, leading to low interference) into account in our method. Extensive simulation results show the superiority of our approach compared to max-min flow routing which is widely used in QoS-sensitive applications.

Wireless streaming of high-definition video is a promising application for 60GHz links, since multi-Gigabit/s data rates are possible. In particular we consider a sports stadium broadcasting system where video signals from multiple cameras are transmitted to a central location. Due to the high pathloss of 60GHz radiation over the large distances encountered in this setting, the use of relays is required. This paper designs a quality-aware coding and relaying algorithm for maximization of the overall video quality. We consider the setting that the source can split its data stream into parallel streams, which can be transmitted via different relays to the destination. For this, we derive the related formulation and re-formulate it as convex programming, which can guarantee optimal solutions.

This demo abstract describes an initial design of a new adaptive video streaming protocol for device-to-device WiFi-based mobile platforms and its software implementation. For the demonstration, two mobile servers and two mobile users will be deployed verifying that our device-to-device adaptive video streaming implementation works with desirable user experience.

This paper proposes a joint optimization framework for minimizing high-definition (HD) video coding rates and selecting optimal relay nodes in 60 GHz millimeter-wave (mmWave) IEEE 802.11ad very high throughput (VHT) wireless systems. While IEEE 802.11ad VHT aims to support uncompressed HD video wireless transmission, its major limitation is the extremely high attenuation even in line-of-sight situations, which leads to a short admissible distance between transmitter and receivers and/or the necessity to compress video. To deal with this problem, the IEEE 802.11ad VHT draft standard defines efficient relaying protocols to extend the network coverage. When multiple source-destination pairs and multiple relays are present, a key question is which relay should help in the forwarding of which source flow. This selection should be done in such a way that the average video quality of the streams, which is related to the throughput in a nonlinear way, is maximized. We solve this problem by an integer programming framework that selects optimal relay nodes, their cooperation modes (i.e., amplify-and-forward, decode-and-forward, or non-cooperation), and the video coding (compression) rates which can maximize transmission quality.

This paper provides concept of data transmission protocol for millimeter wave (mmWave) wireless systems operating in Non-Line-of-Sight environment. This concept is designed to provide an effective and practical functioning of Multiple-Input Multiple-Output (MIMO) transmission mode that exploits combination of Singular Value Decomposition (SVD) of channel matrix and non-adaptive beamforming. The proposed protocol reduces complexity of implementation of the beamformed MIMO system with SVD by transmitting the special signals, called by MIMO postambles. Using these MIMO postambles allows stations to obtain the channel state information and operate in SVD mode without any feedback and simplifies MIMO system architecture significantly.

Many emerging smartphone applications require position information to provide location-based or context-aware services. In these applications, GPS is often preferred over its alternatives such as GSM/WiFi based positioning systems because it is known to be more accurate. However, GPS is extremely power hungry. Hence a common approach is to periodically duty-cycle GPS. However, GPS duty-cycling trades-off positioning accuracy for lower energy. A key requirement for such applications, then, is a positioning system that provides accurate position information while spending minimal energy.

This paper addresses optimal beacon fragmentation schemes for high-performance MAC in 60 GHz millimeter wave (mmWave) networks. The proposed optimal beacon scheme can be classified into two methods, i.e., differentiated and fragmentized beaconing. Differentiated and fragmentized beaconing schemes are for directional and omni-directional data transmission. By the result of performance evaluation, the novelty of the proposed schemes is verified.

This demo presentation shows the usefulness of a multi-Gbps wireless video and audio network for video streaming in conference room applications. In this demo, two notebooks share one display (e.g., Projector) using wireless uncompressed video links. The video quality, switching delay, and how easy to swap the video stream sources (notebooks) will be key features in the application like this.

This paper proposes an energy-aware distributed topology control (EDTC) algorithm for coverage-time optimization unique to clustering-based heterogeneous sensor networks. The coverage-time is defined as the time until one of cluster heads (CHs) runs out of energy. CHs which are operated by EDTC move virtual picket nodes, indicators that determine the boundaries of cluster regions, for the load-balanced regulation of cluster regions in a distributed manner. Regulation of cluster regions leads the load-balanced energy consumption of CH which is in the region for coverage-time maximization. Through a performance evaluation of EDTC, we show that our algorithm for clustering-based heterogeneous sensor networks achieves desired properties.

This paper proposes an adaptive and dynamic localized scheme unique to hierarchical clustering in RFID networks, while reducing the overlapping areas of clusters and consequently reducing collisions among RFID readers. Drew on our LLC scheme that adjusts cluster coverage to minimize energy consumption, low-energy localized clustering for RFID networks (LLCR) addresses RFID reader anti-collision problem in this paper. LLCR is a RFID reader anti-collision algorithm that minimizes collisions by minimizing overlapping areas of clusters that each RFID reader covers. LLCR takes into account each RFID reader's energy state as well as RFID reader collisions. For the energy state factor, we distinguish homogeneous RFID networks from heterogeneous ones according to computing power of each RFID reader. Therefore, we have designed efficient homo-LLCR and hetero-LLCR schemes for each case. Our simulation-based performance evaluation shows that LLCR minimizes energy consumption and overlapping areas of clusters of RFID readers.