Article
Telecommunications
Monolina Dutta, Anoop Thomas
Summary: This study focuses on addressing network congestion caused by temporal variance in client demands in the client-server framework, and proposes a decentralized shared caching scheme. The proposed scheme, utilizing index coding techniques, is shown to be optimal among all linear schemes, achieving a comparable rate to existing centralized prefetching schemes.
IEEE COMMUNICATIONS LETTERS
(2021)
Article
Computer Science, Theory & Methods
Hai Zhou, Dan Feng, Yuchong Hu
Summary: This article proposes a single-node multi-level forwarding repair technique for heterogeneous networks, as well as a multi-node scheduling repair technique. SMFRepair accelerates single-node recovery time by selecting helper nodes and utilizing idle nodes. On the other hand, MSRepair reduces multi-node recovery time by scheduling repair links on multiple nodes.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
(2022)
Article
Computer Science, Theory & Methods
Mi Zhang, Qiuping Wang, Zhirong Shen, Patrick P. C. Lee
Summary: Stragglers are common in large-scale storage systems and cause performance instability. We propose an erasure-coded caching design that achieves robust straggler tolerance by caching only parity blocks, reducing read latency.
JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING
(2022)
Article
Computer Science, Theory & Methods
Xiaolu Li, Keyun Cheng, Zhirong Shen, Patrick P. C. Lee
Summary: Erasure coding provides a storage-efficient redundancy mechanism for large-scale storage clusters, but it incurs high performance overhead in failure repair. Recent developments in accurate disk failure prediction allow repairing soon-to-fail nodes in advance, opening new opportunities for accelerating failure repair in erasure-coded storage. FastPR is a fast proactive repair solution that fully parallelizes the repair operation by coupling migration and reconstruction methods. FastPR significantly reduces repair time for both Reed-Solomon codes and Azure's Local Reconstruction Codes.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
(2022)
Article
Engineering, Electrical & Electronic
Kai Wan, Minquan Cheng, Mari Kobayashi, Giuseppe Caire
Summary: Caching at the wireless edge nodes is an efficient way to improve network efficiency. This paper proposes a solution for the coded caching problem based on location-based content. The proposed scheme can effectively reduce network load and enhance communication efficiency.
IEEE TRANSACTIONS ON COMMUNICATIONS
(2022)
Article
Computer Science, Information Systems
Guillaume Ruty, Hana Baccouch, Victor Nguyen, Andre Surcouf, Jean-Louis Rougier, Nadia Boukhatem
Summary: The study proposes a new caching policy for erasure-coded storage systems, aiming to cache full objects to improve cache hit ratio and reduce waste. Simulation evaluation demonstrates that the full replica solution outperforms traditional caching methods in terms of cache performance.
CLUSTER COMPUTING-THE JOURNAL OF NETWORKS SOFTWARE TOOLS AND APPLICATIONS
(2021)
Article
Physics, Multidisciplinary
Frederique Oggier, Anwitaman Datta
Summary: This study focuses on designing grid quorum systems for MDS erasure code based distributed storage systems, showing how to build these systems, investigate their load characteristics, and demonstrate achieving sequential consistency even in the presence of storage node failures.
Article
Engineering, Electrical & Electronic
Guangyu Zhu, Caili Guo, Tiankui Zhang, Yulin Shao
Summary: This paper focuses on a cache-enable device-to-device (D2D) communication network with user mobility and proposes a mobility-aware coded caching scheme to reduce network traffic. By assigning dynamic cache memory to mobile users, content exchange is enabled via relaying even among users who never meet. The use of network coding effectively reduces network traffic and improves decoding efficiency. The numerical results demonstrate the superiority of the proposed algorithm compared to random and greedy algorithms and the standard Ford-Fulkerson algorithm in terms of broadcasting data and successful decoding ratio.
PHYSICAL COMMUNICATION
(2023)
Article
Computer Science, Hardware & Architecture
Minquan Cheng, Longsong Liu, Jinyu Wang, Qingyong Deng
Summary: This article focuses on designing a coded caching scheme for fixed subpacketization and dynamic number of users by constructing an appropriate matrix. By removing columns, a new flexible PDA with dynamic column number is obtained, and the transmission load of the scheme is order optimal. Furthermore, an improved PDA can be obtained using combinatorial methods to reduce transmission load.
JOURNAL OF SYSTEMS ARCHITECTURE
(2022)
Article
Engineering, Electrical & Electronic
Mohammed S. Al-Abiad, Md. Zoheb Hassan, Md. Jahangir Hossain
Summary: One of the main advantages of multi-level cache-enabled networks is pushing content proximity to the network edge and proactively caching them at multiple transmitters, reducing fronthaul congestion and optimizing throughput by jointly optimizing network-coded user scheduling and power allocation.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY
(2021)
Article
Computer Science, Theory & Methods
Tian Xie, Sanchal Thakkar, Ting He, Patrick McDaniel, Quinn Burke
Summary: This study focuses on jointly optimizing caching and routing in networks with arbitrary topology. It analyzes the complexity of the problem and develops polynomial-time algorithms with approximation guarantees in important special cases. It also proposes an alternating optimization algorithm with good empirical performance and fast convergence.
IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS
(2023)
Article
Computer Science, Information Systems
Minquan Cheng, Jie Li, Xiaohu Tang, Ruizhong Wei
Summary: Coded caching systems aim to reduce data transmission during peak traffic, with transmission rate and subpacketization being important parameters. Designing schemes with lower transmission rates and subpacketization can improve transmission efficiency.
IEEE TRANSACTIONS ON INFORMATION THEORY
(2021)
Article
Computer Science, Information Systems
Eleftherios Lampiris, Antonio Bazco-Nogueras, Petros Elia
Summary: This paper investigates the optimal tradeoff between feedback costs and DoF performance in multi-antenna cache-aided wireless networks. It proposes a novel algorithm and information-theoretic converse to achieve optimal linear DoF performance with partial channel state information. In practice, caching can provide unbounded gains without additional feedback costs, enhancing the performance of multi-antenna systems.
IEEE TRANSACTIONS ON INFORMATION THEORY
(2022)
Article
Computer Science, Information Systems
Shushi Gu, Zichao Yu, Qinyu Zhang, Tao Huang
Summary: With the rapid development of future networks, the desire for low-energy system design and transmission strategy becomes more compelling. This letter investigates the energy consumption issue of hierarchical cooperative caching networks (HCCNs) and proposes a three-stage coded transmission strategy. A total energy consumption minimization problem is formulated with cache size constraints, and an algorithm is provided to obtain the optimal cache placement matrix. Numerical results demonstrate the effectiveness of the optimized three-stage coded transmission strategy in reducing energy consumption compared to various multicast transmission strategies under different users' cache sizes.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2023)
Article
Engineering, Electrical & Electronic
Hui Yang, Ling Fu Xie, Juan Liu, Lu Lu
Summary: This paper investigates a scheduling problem in decentralized coded caching, taking into account sequential and random request arrivals. To achieve a tradeoff between energy efficiency and delay penalty, the problem is formulated as an optimal stopping problem and an iterative method is proposed to approximate the optimal rule.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY
(2022)
Article
Computer Science, Information Systems
Zhixiong Yang, Jing-Yuan Xia, Junshan Luo, Shuanghui Zhang, Deniz Gunduz
Summary: This paper proposes a learning aided gradient descent algorithm for maximizing the weighted sum rate in MISO beamforming. By dynamically determining the optimization strategy through a neural network, the algorithm shows superior performance. Numerical results indicate that it outperforms other methods with modest computational complexity.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2022)
Article
Engineering, Electrical & Electronic
Deniz Gunduz, Zhijin Qin, Inaki Estella Aguerri, Harpreet S. Dhillon, Zhaohui Yang, Aylin Yener, Kai Kit Wong, Chan-Byoung Chae
Summary: This tutorial summarizes the efforts in communication systems to integrate message semantics and goals of communication into their designs, as well as incorporating the context of communication exchange. It covers the foundations, algorithms, and potential implementations, with a focus on utilizing information theory and learning in semantics and task-aware communications.
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS
(2023)
Article
Computer Science, Information Systems
Haotian Wu, Yulin Shao, Krystian Mikolajczyk, Deniz Gunduz
Summary: This research proposes a learning-based channel-adaptive joint source and channel coding (CA-JSCC) scheme for wireless image transmission over multipath fading channels. The proposed method is able to adapt to channel-gain and noise-power variations by utilizing estimated channel state information (CSI). Experimental results demonstrate that CA-JSCC achieves state-of-the-art performance compared to existing JSCC schemes, and it is robust to varying channel conditions.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2022)
Article
Physics, Multidisciplinary
Sreejith Sreekumar, Deniz Gunduz
Summary: This paper investigates a two-terminal distributed binary hypothesis testing problem over a noisy channel. The observer and decision maker each have access to independent and identically distributed samples. The observer communicates with the decision maker through a discrete memoryless channel. The trade-off between the exponents of the type I and type II error probabilities is studied, and two inner bounds are obtained using separation-based and joint schemes respectively. The results show that the joint scheme achieves a tighter bound than the separation-based scheme for certain points of the error-exponents trade-off.
Article
Engineering, Electrical & Electronic
Yulin Shao, Deniz Gunduz, Soung Chang Liew
Summary: Over-the-air computation (OAC) is an important component of future wireless networks, enabling efficient function computation in multiple-access edge computing. Traditional OAC using maximum likelihood (ML) estimation is susceptible to noise and error propagation. To address this, a Bayesian approach is proposed in this paper, where each edge device transmits statistical information to the fusion center for misalignment handling. Numerical and simulation results show the superior performance of the proposed Bayesian estimators in different scenarios.
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Baturalp Buyukates, Emre Ozfatura, Sennur Ulukus, Deniz Gunduz
Summary: Distributed implementations are crucial in speeding up large scale machine learning applications. Distributed gradient descent is widely used to parallelize the learning task, but straggling workers can cause performance bottlenecks. Recent techniques in coded distributed computation have been introduced to mitigate straggling workers and improve the completion time of iterations.
IEEE TRANSACTIONS ON COMMUNICATIONS
(2023)
Article
Engineering, Electrical & Electronic
Qiao Lan, Qunsong Zeng, Petar Popovski, Deniz Gunduz, Kaibin Huang
Summary: This paper investigates the scenario of inference at the wireless edge, where devices connect to an edge server and request remote classification. Due to the limited resources of wireless channels, the devices need to upload high-dimensional features, causing a communication bottleneck. To address this issue, a "Progressive Feature Transmission" (ProgressFTX) protocol is proposed to minimize overhead by gradually transmitting features until a target confidence level is achieved.
IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS
(2023)
Article
Computer Science, Information Systems
Yulin Shao, Deniz Gunduz
Summary: Recent progress in DeepJSCC, a deep learning-based joint source-channel coding, has introduced a new paradigm of semantic communications. It leverages semantic-aware features directly from the source signal and utilizes discrete-time analog transmission. Compared to traditional digital communications, DeepJSCC-based semantic communications offer superior receiver reconstruction performance, graceful degradation with diminishing channel quality, but also exhibit a large peak-to-average power ratio (PAPR) in the transmitted signal. This letter explores PAPR reduction techniques to retain DeepJSCC's superior image reconstruction performance while suppressing PAPR to an acceptable level, paving the way for practical implementation of DeepJSCC in semantic communication systems.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2023)
Article
Computer Science, Interdisciplinary Applications
Isaac Shiri, Behrooz Razeghi, Alireza Vafaei Sadr, Mehdi Amini, Yazdan Salimi, Sohrab Ferdowsi, Peter Boor, Deniz Guenduez, Slava Voloshynovskiy, Habib Zaidi
Summary: A federated learning (FL) framework was developed for multi-institutional PET/CT image segmentation. The FL algorithms showed comparable performance to centralized learning methods and outperformed the single center-based baseline. The results have promising implications for HN tumor segmentation from PET/CT images.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2023)
Article
Engineering, Electrical & Electronic
Ecenaz Erdemir, Tze-Yang Tung, Pier Luigi Dragotti, Deniz Gunduz
Summary: Recent works have shown that DeepJSCC, a joint source-channel coding scheme using deep neural networks (DNNs), provides promising results for wireless image transmission. However, existing methods mostly focus on the distortion of reconstructed signals rather than human perception. This work proposes two novel JSCC schemes, InverseJSCC and GenerativeJSCC, that leverage deep generative models for improved perceptual quality in extreme conditions. Simulation results demonstrate that both InverseJSCC and GenerativeJSCC outperform DeepJSCC in terms of perceptual quality and distortion.
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Jing-Yuan Xia, Shengxi Li, Jun-Jie Huang, Zhixiong Yang, Imad M. Jaimoukha, Deniz Gunduz
Summary: In this article, a meta-learning based alternating minimization (MLAM) method is proposed to solve nonconvex problems of multiple variables. By iteratively minimizing a part of the global losses, the method learns an adaptive strategy for superior performance while maintaining interpretability.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Information Systems
Wing Fei Lo, Nitish Mital, Haotian Wu, Deniz Gunduz
Summary: We propose two novel deep learning-based joint source and channel coding (JSCC) schemes for collaborative image retrieval problem at the wireless edge. The proposed schemes outperform the single-device JSCC and separation-based multiple-access benchmarks. We also propose a channel state information-aware JSCC scheme with attention modules to adapt to varying channel conditions.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2023)
Article
Engineering, Electrical & Electronic
Yulin Shao, Soung Chang Liew, Deniz Gunduz
Summary: This article investigates a fundamental problem of NoisyNNs, which is how to reconstruct the DNN weights from noise. A denoising approach is proposed to maximize the inference accuracy of the reconstructed models. Experimental results demonstrate that our denoiser outperforms the maximum likelihood estimation in small-scale problems and shows significantly better performance when applied to advanced learning tasks with modern DNN architectures.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2023)
Article
Computer Science, Theory & Methods
Behrooz Razeghi, Flavio P. Calmon, Deniz Gunduz, Slava Voloshynovskiy
Summary: Bottleneck problems are gaining attention in machine learning and information theory. In this work, the complexity-leakage-utility bottleneck (CLUB) model is proposed, which provides a unified theoretical framework, new interpretation of generative and discriminative models, insights for generative compression models, and fair generative models. The CLUB model is formulated as a complexity-constrained privacy-utility optimization problem and is connected to related problems such as information bottleneck (IB), privacy funnel (PF), and conditional entropy bottleneck (CEB). The deep variational CLUB (DVCLUB) models are introduced using neural networks and can be applied to generative models and fair representation learning problems. Extensive experiments are conducted on colored-MNIST and CelebA datasets.
IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
(2023)
Article
Engineering, Electrical & Electronic
Yuxuan Sun, Fan Zhang, Junlin Zhao, Sheng Zhou, Zhisheng Niu, Deniz Gunduz
Summary: This paper investigates a distributed computing scenario with multiple master hosts and heterogeneous workers, aiming to reduce the straggler effect and minimize task delays through coded computation and optimized task allocation. The proposed algorithms improve task completion delay compared to benchmarks.
IEEE TRANSACTIONS ON SIGNAL PROCESSING
(2022)