Distributed Foundation Models for Multi-Modal Learning in 6G Wireless Networks

IEEE Wireless Communications, Journal Year: 2024, Volume and Issue: 31(3), P. 20 - 30

Published: June 1, 2024

Benefiting from the ability to process and integrate data various modalities, multi-modal foundation models (FMs) facilitate potential applications across a range of fields, including computer vision (CV), natural language processing (NLP), diverse such as imagetext retrieval. Currently, FMs are deployed on computing clusters for training inference meet their considerable computational demands. In foreseeable future, parameter size is expected evolve further, posing challenges both computation resources energy supply. Fortunately, leveraging next-generation wireless networks (6G) aggregate substantial myriad devices holds promise handling aforementioned challenges. this work, we delve into state-of-the-art artificial intelligence (AI) techniques, specifically focusing pipeline parallelism, learning, with aim supporting sustainable development distributed in 6G era. context compressing activations gradients while intelligently allocating communication can overcome bottlenecks caused by unstable links. For federated learning (FL) over-the-air (AirComp) seamlessly integrates computation, significantly expediting gradient aggregation. Furthermore, following recent success large (LLMs) incorporating FMs, NLP CV, along broader AI community, establishing cornerstone intrinsic within networks.

Language: Английский

---

Joint communication and computation design for secure integrated sensing and semantic communication system

Science China Information Sciences, Journal Year: 2025, Volume and Issue: 68(3)

Published: Feb. 11, 2025

Language: Английский

---

A Joint Communication and Computation Design for Probabilistic Semantic Communications

Entropy, Journal Year: 2024, Volume and Issue: 26(5), P. 394 - 394

Published: April 30, 2024

In this paper, the problem of joint transmission and computation resource allocation for a multi-user probabilistic semantic communication (PSC) network is investigated. considered model, users employ information extraction techniques to compress their large-sized data before transmitting them multi-antenna base station (BS). Our model represents through substantial knowledge graphs, utilizing shared probability graphs between BS efficient compression. The formulated as an optimization with objective maximizing sum equivalent rate all users, considering total power budget limit constraints. load in PSC non-smooth piecewise function respect compression ratio. To tackle non-convex challenge, three-stage algorithm proposed, where solutions received beamforming matrix BS, transmit each user, ratio user are obtained stage by stage. numerical results validate effectiveness our proposed scheme.

Language: Английский

---

Spectral-Efficient RIS-Aided RSMA URLLC: Toward Mobile Broadband Reliable Low Latency Communication (mBRLLC) System

IEEE Transactions on Wireless Communications, Journal Year: 2023, Volume and Issue: 23(4), P. 3507 - 3524

Published: Sept. 1, 2023

Next-generation wireless applications are expected to enable extended ultra-reliable low latency communication (xURLLC) support high data rates along with ultra-high reliability and end-to-end features beyond the capabilities of existing core services. These consolidated URLLC requirements in resource-constrained systems necessitate shift from conventional architectures more powerful robust multiple access schemes. This paper investigates a multi-reconfigurable intelligent surface (RIS)-assisted rate-splitting (RSMA) prompt an unconventional xURLLC service called mobile broadband reliable (mBRLLC) for spectral efficiency under finite block-length (FBL) transmission constraints. To spectral-efficient resource allocation, we formulate sum throughput maximization problem joint optimization precoder design at base-station (BS), common private symbols each user, passive beamforming RIS. solve NP-hardness non-convexity formulated problem, use alternating technique decouple original into three sub-problems: active BS, optimization, RIS which solved using general convex approximations. Simulations demonstrate effectiveness proposed allocation algorithm over The considered RSMA system achieves even lower higher reliability. Additionally, investigation encompasses evaluation deployment implications, analysis worst-case scenario, assessment influence channel estimation errors.

Language: Английский

---

Physical-Layer Adversarial Robustness for Deep Learning-Based Semantic Communications

IEEE Journal on Selected Areas in Communications, Journal Year: 2023, Volume and Issue: 41(8), P. 2592 - 2608

Published: June 23, 2023

End-to-end semantic communications (ESC) rely on deep neural networks (DNN) to boost communication efficiency by only transmitting the semantics of data, showing great potential for high-demand mobile applications. We argue that central success ESC is robust interpretation conveyed at receiver side, especially security-critical applications such as automatic driving and smart healthcare. However, robustifying challenging extremely vulnerable physical-layer adversarial attacks due openness wireless channels fragileness models. Toward robustness in practice, we ask following two questions: Q1: For attacks, it possible generate semantic-oriented are imperceptible, input-agnostic controllable? Q2: Can develop a defense strategy against distortions previously proposed adversaries? To this end, first present MobileSC , novel framework considers computation memory environments. Equipped with framework, propose xmlns:xlink="http://www.w3.org/1999/xlink">SemAdv perturbation generator aims craft adversaries over air abovementioned criteria, thus answering Q1. better characterize real-world effects training evaluation, further introduce method $\texttt {SemMixed}$ harden existing strong threats, Q2. Extensive experiments three public benchmarks verify effectiveness our methods various physical attacks. also show some interesting findings, e.g., can even be more than classical block-wise systems low SNR regime.

Language: Английский

---

Resilient Machine Learning-Based Semantic-Aware MEC Networks for Sustainable Next-G Consumer Electronics

IEEE Transactions on Consumer Electronics, Journal Year: 2023, Volume and Issue: 70(1), P. 2188 - 2199

Published: Dec. 4, 2023

In this paper, we investigate a semantic-aware mobile edge computing (MEC) network for sustainable next-G consumer electronics, which leverages advanced semantic communication technology to overcome the limitations of available bandwidth and thereby improve efficiency. For network, electronic devices can offload information extracted from task instead transmitting whole in conventional MEC networks, where latency energy consumption be significantly reduced through proper encoding, offloading, computation allocation decisions. However, non-convexity issue makes it difficult obtain optimal decision. To address issue, two-level optimization framework is proposed. Specifically, upper-level optimization, resilient deep reinforcement learning (DRL) approach utilized enable adaptive offloading encoding decisions within dynamic network. lower-level design three criteria allocating resources by carefully considering trade-off between computational complexity implementation Finally, extensive simulations are conducted validate effectiveness our proposed strategy. The findings paper help reduce hence supporting development electronics.

Language: Английский

---

Compressive transmission scheme for power regulation of embedded 5G communication devices

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Feb. 18, 2025

Power management for embedded devices in Fifth Generation (5G) networks is mandatory synchronizing the communication between devices. In such cases, need integration power optimization recommended aiding lossless and high-speed communications. To suppress issues hardware-based failures during transmissions, this article proposes a Compressive Transmission Scheme (CTS) through Regulation (PR). The proposed scheme identifies multiple transmission possibilities under low high throughput constraints of 5G single interval. device integrations are decided by available power-efficient slots. Such allocation slots defined integrated using neural-diffracted networks. learning network defines objectives hardware power. This pursued until completed; adverse energy drain impact handled offloading to active available. balances prevent loss satisfying constraints. For maximum slots/device, achieves 11.46% slot allocation, 12.47% latency, 9.99% less consumption.

Language: Английский

---

Semantic Communication with Probability Graph: A Joint Communication and Computation Design

2021 IEEE 94th Vehicular Technology Conference (VTC2021-Fall), Journal Year: 2023, Volume and Issue: unknown

Published: Oct. 10, 2023

In this paper, we present a probability graph-based semantic information compression system for scenarios where the base station (BS) and user share common background knowledge. We employ graphs to represent shared knowledge between communicating parties. During transmission of specific text data, BS first extracts from text, which is represented by graph. Subsequently, omits certain relational based on graph reduce data size. Upon receiving compressed can automatically restore missing using predefined rules. This approach brings additional computational resource consumption while effectively reducing communication consumption. Considering limitations wireless resources, address problem joint computation allocation design, aiming at minimizing total energy network adhering latency, transmit power, constraints. Simulation results demonstrate effectiveness proposed system.

Language: Английский

---

Optimizing Synchronization Delay for Digital Twin over Wireless Networks

ICASSP 2022 - 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Journal Year: 2024, Volume and Issue: unknown

Published: March 18, 2024

In this paper, the problem of low-latency communication and computation resource allocation for digital twin (DT) over wireless networks is investigated. considered model, multiple physical devices in network (PN) needs to frequently offload task related data (DNT), which generated controlled by central server. Due limited energy budget devices, both accuracy transmission power must be during DT procedure. This joint formulated as an optimization whose goal minimize overall delay system under total PN DNT model constraints. To solve problem, alternating algorithm with iteratively solving device scheduling, control, offloading subproblems. For scheduling subproblem, optimal solution obtained closed form through dual method. Numerical results verify that proposed can reduce up 51.2% compared conventional schemes.

Language: Английский

---

SCAN: Semantic Communication With Adaptive Channel Feedback

IEEE Transactions on Cognitive Communications and Networking, Journal Year: 2024, Volume and Issue: 10(5), P. 1759 - 1773

Published: April 29, 2024

Language: Английский

---