PaperlayerView exact revision on arXiv

Source-linked AI summary

A Survey on Semantic Communications in Internet of Vehicles

Sha Ye, Qiong Wu, Pingyi Fan, Qiang Fan

arXiv:2503.03767v2cs.NIcs.LGeess.SP

TL;DR

IoV data growth puts pressure on traditional communication through scarce spectrum and high latency. This survey reviews semantic communication technologies and applications in the IoV. Its synthesis reports promise across IoV applications while identifying unresolved challenges, including knowledge-base updating and security.

  • Problem

    IoV data growth leaves traditional communication facing scarce spectrum, high latency, and limited semantic-information processing.

    Source: §1 · p. 2

  • Method

    The paper systematically reviews IoV semantic communication, covering technical background, key technologies, application case studies, challenges, and future directions.

    Source: §1 · p. 2, §6 · p. 43

  • Results

    Case studies report that task-oriented semantic frameworks outperform traditional methods in low-SNR scenarios, while generalization to complex environments remains challenging.

    Source: §4.5 · p. 40

  • Takeaways & Limitations

    The survey identifies edge computing and distributed knowledge sharing as critical for scalability, and privacy-preserving mechanisms as a priority for sensitive applications.

    Source: §4.5 · p. 40

    The survey identifies unresolved challenges in semantic IoV security, real-time knowledge-base updates, and standardization.

    Source: §5.6 · p. 43

Abstract

from arXiv · show

Internet of Vehicles (IoV), as the core of intelligent transportation system, enables comprehensive interconnection between vehicles and their surroundings through multiple communication modes, which is significant for autonomous driving and intelligent traffic management. However, with the emergence of new applications, traditional communication technologies face the problems of scarce spectrum resources and high latency. Semantic communication, which focuses on extracting, transmitting, and recovering some useful semantic information from messages, can reduce redundant data transmission, improve spectrum utilization, and provide innovative solutions to communication challenges in the IoV. This paper systematically reviews state of art of semantic communications in the IoV, elaborates the technical background of IoV and semantic communications, and deeply discusses key technologies of semantic communications in IoV, including semantic information extraction, semantic communication architecture, resource allocation and management, and so on. Through specific case studies, it demonstrates that semantic communications can be effectively employed in the scenarios of traffic environment perception and understanding, intelligent driving decision support, IoV service optimization, and intelligent traffic management. Additionally, it analyzes the current challenges and future research directions. This survey reveals that semantic communications has broad application prospects in IoV, but it is necessary to solve the real existing problems by combining advanced technologies to promote its wide application in IoV and contributing to the development of intelligent transportation system.

1. Introduction

IoV data growth is straining traditional communication through scarce spectrum, high latency, and limited semantic-processing capability. The survey reviews semantic communication technologies and applications in the IoV, alongside challenges and future directions.

Motivation

  • Explosive IoV data growth exposes traditional communication limits in spectrum, latency, and semantic-information processing.These limitations restrict IoV performance and hinder intelligent transportation systems.Source: §1 · p. 2

Semantic communication in IoV

  • Semantic communication extracts, transmits, and recovers message meaning, with potential to improve transmission efficiency and support IoV services.The paper also identifies challenges in contextual understanding and efficient decoding while maintaining concealment.Source: §1 · p. 2

Survey scope

  • The survey reviews IoV semantic communication through key technologies and applications, and discusses challenges and future research directions.Its technology coverage includes semantic information extraction, communication architecture, and resource allocation and management; its application case studies span perception, driving support, service optimization, and traffic management.Source: §1 · pp. 2–3

2.1. Internet of Vehicles

The IoV connects vehicles, infrastructure, pedestrians, and roadside units to support information sharing and collaborative interaction. This section outlines IoV communication modes, architectural frameworks, and communication technologies.

2.1. Internet of Vehicles

  • IoV connects vehicles with vehicles, infrastructure, pedestrians, and roadside units; its communication modes include V2V, V2I, V2P, V2R, V2D, and V2G.A mature IoV system typically incorporates these modes.Source: §2.1 · pp. 4–5

2.1.1. IoV Architectures

  • IoV architectures include three-tier, in-vehicle/intervehicle, and vehicle–road–cloud frameworks, with layers or components supporting sensing, information exchange, and services.The three-tier model has perception, network, and application layers; the vehicle–road–cloud model comprises onboard, road-facility, and cloud-control units.Source: §2.1.1 · pp. 5–6

2.1.2. Communication Technologies

  • IoV communications span vehicular, cellular, and short-range static technologies, including DSRC, 4G/LTE and 5G NR V2X, satellite, Zigbee, Bluetooth, and Wi-Fi.The cited examples range from DSRC-based vehicle-to-roadside communication to Wi-Fi access in parking lots and service stations.Source: §2.1.2 · pp. 6–7

2.2. Semantic Communication

Semantic communication encodes, transmits, and decodes information around its meaning and intent, conveying essential semantic content rather than simply transmitting raw data.

2.2.1. Semantic Communication System Architectures

  • The architecture combines semantic and channel encoders and decoders with a knowledge base, shifting communication from bit accuracy toward recovering the sender’s intended meaning.The knowledge base supplies a basis for semantic understanding; semantic noise can cause incorrect recognition or interpretation.Source: §2.2.1 · pp. 7–8
  • Semantic communication extracts meaning from raw data before transmission, using AI methods such as NLP, computer vision, and speech recognition to identify task-relevant content.The section describes semantic extraction as the basis for reducing transmitted data.Source: §2.2.1 · p. 8
  • Joint Source–Channel Coding jointly optimizes source and channel coding; deep-learning examples address image, text, speech, and multiuser multimodal transmission.The section reports Deep JSCC performance under low SNR and limited bandwidth, and DeepSC performance under low-SNR conditions, without providing numerical comparisons.Source: §2.2.1 · p. 9

2.2.2. Semantic Communication Types

  • The section distinguishes semantics-oriented communication, goal-oriented communication, and semantic-aware communication by their focus on semantic accuracy, task execution, or agent collaboration.The first two depend on shared local knowledge, while semantic-aware communication derives information from agent behavior and the task environment.Source: §2.2.2 · pp. 9–10

2.2.3. Semantic Metrics

  • Semantic metrics assess similarity at the semantic level and task performance at the effectiveness level, using measures suited to different media and applications.Examples include BLEU, PSNR, SSIM, AoI, and VoI.Source: §2.2.3 · p. 10

2.3. Integration of IoV and Semantic Communication

IoV communication demands span reliability, latency, bandwidth, and security, while conventional technologies face spectrum, congestion, and semantic-processing limitations. Semantic communication offers efficiency and application benefits across connected-vehicle services.

2.3. Integration of IoV and Semantic Communication

  • IoV applications require reliable, low-latency, high-bandwidth, and secure communication, while cellular networks and DSRC face spectrum limits, congestion-related latency, and weak semantic processing.Autonomous driving requires real-time traffic information; remote monitoring and diagnostics involve large data volumes.Source: §2.3 · pp. 10–11
  • Semantic communication reduces redundant transmission, can improve accuracy in constrained bandwidth or low-SNR settings, and can obscure content from people lacking shared background knowledge.The section also describes adaptation to diverse communication scenarios based on information meaning and user needs.Source: §2.3 · p. 11
  • The cited ITU-T report describes Semantic-Aware Networks as using shared semantic terms and grammars to interpret data and support heterogeneous-data interoperability and real-time decisions.The report also describes reliability through distributed semantic resource provisioning.Source: §2.3 · p. 11
  • Applications include autonomous driving and traffic management, as well as remote diagnostics and connected entertainment, with services such as route-tailored navigation and repair guidance.The section describes vehicle data transmission to service centers for fault identification and remote repair guidance.Source: §2.3 · pp. 11–12

2.4. Lessons Learned

The chapter establishes background on IoV and semantic communication, including their architectures, communication modes, and integration potential.

2.4. Lessons Learned

  • The chapter covers IoV architecture and communication modes alongside semantic-communication system architecture and types.These topics provide technical background for their integration.Source: §2.4 · p. 12
  • It frames semantic communication as a way to improve communication efficiency and security by extracting and transmitting semantic content.Source: §2.4 · p. 12
  • The chapter considers how integrating IoV with semantic communication may promote intelligent transportation systems.Source: §2.4 · p. 12

3. Key Technologies of Semantic Communication in IoV

The section introduces IoV semantic-communication technologies and an evaluation framework intended to support systematic comparison across methodologies.

3. Key Technologies of Semantic Communication in IoV

  • The section identifies semantic information extraction, communication architecture, and resource allocation as key technologies for efficient and reliable IoV semantic communication.Source: §3 · p. 12
  • A semantic layer adds semantic-feature extraction and compression, semantic-content recovery, shared-knowledge support, and coordination between semantic and channel encoding and decoding.The cited four-layer V2X architecture comprises device, network, semantic, and application layers.Source: §3 · p. 12
  • The proposed evaluation framework organizes comparisons around semantic accuracy, end-to-end latency, and resource efficiency.Resource efficiency includes computational overhead and bandwidth efficiency.Source: §3 · p. 13, Table 2 · p. 14

3.1. Semantic Information Extraction

Semantic information extraction turns vehicle sensor data into key information for perception and decision making, with research moving from single-sensor processing toward multimodal fusion. The surveyed methods offer different trade-offs in recognition, transmission efficiency, and robustness, while real-time processing and multimodal accuracy remain future priorities.

3.1. Semantic Information Extraction

  • Deep-learning-based extraction has evolved from single-sensor processing toward multimodal fusion to improve the efficiency and accuracy of extracting key information from vehicle sensor data.Extraction supports real-time perception and intelligent decision making.Source: §3.1 · p. 14

3.1.1. Single-Modal Extraction

  • Single-modal approaches use segmentation and learned compression for road-scene understanding, with methods differing in object recognition, ROI quality, and data efficiency.Reported examples include U-Net’s 94% mIoU on Camvid and a 1.19 dB ROI PSNR gain at 0.25 bpp over traditional compression.Source: §3.1.1 · pp. 14–15

3.1.2. Multimodal Fusion

  • Multimodal fusion combines visual, GPS, map, point-cloud, and radar data to broaden perception, reduce redundancy, and improve robustness under noise.The cited CNN-GNN-LSTM method maintained an F1-score of 0.93 at 0 dB SNR, while diffusion-enhanced decoding raised IoU from 0.65 to 0.80 at 0 dB.Source: §3.1.2 · pp. 15–16
  • Future extraction research prioritizes lightweight, fast models for high-speed driving and more accurate, real-time multimodal fusion.These are identified as requirements for processing large data volumes in real time.Source: §3.1.2 · p. 16

3.2. Semantic Communication Architecture

The survey groups IoV semantic-communication architectures by their application settings and optimization strategies, covering collaboration, image transmission, generative AI, and other designs. Across these approaches, reported gains in communication efficiency and semantic performance coexist with constraints in computation, real-time deployment, robustness, and adaptability.

3.2. Semantic Communication Architecture

  • The architecture section classifies IoV semantic-communication designs and summarizes their applications and optimization strategies.Table 3 is identified as the summary of architecture classifications and characteristics.Source: §3.2 · p. 16, Table 3 · p. 17

3.2.1. Multiuser Collaboration and Multitask Driving

  • Multiuser collaboration and multitask frameworks reduce redundant transmission and support varied driving tasks, but computational demands, CSI dependence, and dynamic-channel adaptability constrain deployment.Co-SC’s performance gains are limited at high SNR and depend heavily on accurate CSI; other reviewed systems also raise real-time computation concerns.Source: §3.2.1 · pp. 18–19

3.2.2. Oriented to Image Transmission

  • Under an average mIoU requirement of 60%, VIS-SemCom achieves nearly 6 dB coding gain over traditional solutions and reduces transmitted data by up to 70%.The system also improves key-object segmentation accuracy by approximately 4%, while generalization in complex environments remains to be verified.Source: §3.2.2 · pp. 19–20
  • Image-transmission architectures use semantic features, importance-aware coding, and routing to improve similarity, reduce data volume, or lower latency, with real-time and generalization limits remaining.The reviewed approaches differ in SNR performance and computational overhead, and some lack detailed latency or deployment evaluations.Source: §3.2.2 · pp. 19–21

3.2.3. Generative AI-Based Semantic Communication Architectures

  • Generative AI architectures compress or reconstruct semantic content and can improve multimodal processing, but computational cost, privacy, SNR robustness, and deployment feasibility remain concerns.The surveyed frameworks vary: some reduce transmitted data substantially, while A-GSC reports favorable delay and computational overhead but needs further deployment verification.Source: §3.2.3 · pp. 21–23

3.2.4. Other Architectures

  • Other architectures apply blockchain-assisted knowledge management, task-driven green transmission, and neural-network sharing, while leaving scalability, computational cost, or real-time feasibility open.PreCMTS is reported to reduce energy consumption and improve semantic-transmission reliability; SCKS shares model knowledge from RSUs to vehicles.Source: §3.2.4 · pp. 23–24

3.3. Resource Allocation and Management

IoV resource allocation must accommodate mobile vehicles and tasks with changing demands for spectrum, power, and computing. The survey reviews reinforcement-learning, optimization-theory, and federated-learning approaches, alongside limitations in deployment costs and evaluation.

3.3. Resource Allocation and Management

  • Resource demands vary by task: road-condition transmission needs low latency and high bandwidth, while vehicle diagnostics prioritize accuracy and completeness.The section groups resource-allocation strategies by their technical approaches.Source: §3.3 · pp. 24–25

3.3.1. Reinforcement Learning-Based Resource Allocation Methods

  • PPO-based SARADC and SAC-based SSS adapt spectrum sharing to semantic metrics; both report advantages over conventional approaches, with SSS outperforming bit-based methods on HSSE and SRS.SARADC reports improved HSSE and semantic throughput under high-resolution, low-SNR conditions; both studies raise computational-overhead concerns for real-time deployment.Source: §3.3.1 · p. 26
  • DDQN, MARL-based SAMRA [32], and MADQN-based VSRAA-SM adapt resource allocation to V2V, platoon, and in-vehicle video tasks, respectively.Reported benefits include reduced delay and improved semantic similarity, QoE and transmission success [32], and video-understanding accuracy and transmission metrics; overhead and cross-method evaluation remain concerns.Source: §3.3.1 · pp. 26–27
  • Across RL-based studies, computational overhead, real-time feasibility, multimodal fusion, and the lack of a unified evaluation framework remain unresolved concerns.The survey calls for lower overhead, greater adaptability and scalability, and quantitative comparison across methods.Source: §3.3.1 · p. 27

3.3.2. Optimization Theory-Based Resource Allocation Methods

  • For D2D networks, Lyapunov-based allocation improves robustness and power consumption, while a two-stage matching and convex-optimization scheme improves platoon video-detection accuracy over V2I-only communication.The studies leave computational overhead and real-time performance, and indicators such as semantic similarity and delay, insufficiently evaluated.Source: §3.3.2 · pp. 27–28
  • The SSTS scheme combines reservation and on-demand plans under uncertain service demand and reports lower energy consumption for semantic than non-semantic data transmission.Its analysis focuses on transmission cost and subscription strategy, with real-time computational overhead and complexity not examined in depth.Source: §3.3.2 · p. 28

3.3.3. Federated Learning-Based Resource Allocation Methods

  • MSFTL reduces vehicle training costs and outperforms traditional FL in convergence speed and final accuracy, while FVSCom improves semantic extraction and robustness to vehicle departures.MSFTL communication efficiency needs improvement for large-scale data transmission; FVSCom does not account for delay and computational overhead across channel conditions.Source: §3.3.3 · p. 29

3.4. Data Security and Privacy Protection

Semantic information in IoV communications raises security and privacy risks, including attacks on semantic content and leakage of sensitive user data. The survey reviews countermeasures while noting gaps in end-to-end protection and detection of covert attacks.

3.4. Data Security and Privacy Protection

  • Semantic information such as vehicle trajectories, driving intentions, and passenger preferences requires confidentiality, integrity, and reliability during transmission.The section frames security and privacy protection as challenges introduced by semantic communication in the IoV.Source: §3.4 · p. 30

3.4.1. Security Risk Analysis

  • The survey identifies adversarial and semantic backdoor attacks, privacy leakage, man-in-the-middle attacks, and model poisoning as distinct IoV security risks.Examples include removing traffic-sign semantics without explicit triggers, inferring travel patterns from trajectories, forging emergency-braking commands, and poisoning shared semantic knowledge.Source: §3.4.1 · p. 30
  • Current defenses focus on single points rather than the full semantic-generation, transmission, and processing chain, with gaps in covert-attack detection and dynamic privacy protection.The survey identifies cross-layer defense architectures and lightweight cryptographic primitives as areas requiring further development.Source: §3.4.1 · pp. 30–31

3.4.2. Countermeasures

  • Reported countermeasures for security and privacy in IoV semantic communication include semantic-information encryption, federated learning with differential privacy, blockchain, edge intelligence, and adversarial-sample detection.Examples include blockchain sharding to reduce knowledge-base verification delay and resist tampering, and shared and private edge knowledge bases to balance semantic understanding and vehicle privacy [14].Source: §3.4.2 · pp. 31–32

3.5. Lessons Learned

Semantic extraction and communication architectures face trade-offs among computational efficiency, accuracy, and security.

3.5. Lessons Learned

  • Deep-learning methods support multimodal fusion and noise resistance but demand substantial resources; RL and FL need real-time optimization, while blockchain introduces scalability issues.The survey prioritizes lightweight models, edge-compatible algorithms, and hybrid optimization strategies to address these trade-offs.Source: §3.5 · p. 32

4. Applications of Semantic Communication in IoV

This section surveys semantic communication across IoV scenarios with differing latency, reliability, and data-accuracy needs, including traffic perception, driving decisions, service optimization, and traffic management. It also covers intelligent traffic management approaches, deployment challenges, and lessons on task-oriented frameworks, edge and distributed collaboration, and privacy protection.

4.1. Traffic Environment Perception and Understanding

  • The perception studies show reduced transmission overhead and improved beam prediction [39], while collaborative perception improves sensing performance, especially for precise target recognition [40].The [40] study’s AP@0.5 and AP@0.7 indicators exceeded traditional methods; the review notes that both approaches may face insufficient processing resources as data volumes grow. Semantic extraction, reliable architecture, and resource management support these applications under limited bandwidth and computing resources.Source: §4.1 · pp. 33–34

4.2. Intelligent Driving Decision Support

  • The DQN-based framework exceeded ARS in training convergence and reward, raising total reward 37.11%; communication costs fell by up to 90.81%.It extracts traffic-sign concepts for CAV decisions, and reward increased with resource-block size. The roadblock-focused system shortens image transmission time and file size versus direct image transmission, with 92% traffic-optimization accuracy; deployment depends on training data and network quality.Source: §4.2 · pp. 35–36

4.3. IoV Service Optimization

  • Service optimization cases show S4 reduces queuing delay and raises throughput, receiver-centric communication [44] reduces transmitted data while meeting receiver needs, and DL-SCMT improves reconstruction and classification.S4 uses knowledge-base matching and vehicle-service pairing under reliability constraints, but may require costly computation. The receiver-centric system extracts requested information using language models and semantic tools, with processing delays possible; DL-SCMT combines smart fabrics and 6G networking but is limited to in-vehicle scenarios. The review finds DL-SCMT robust at low SNR.Source: §4.3 · pp. 36–37

4.4. Intelligent Traffic Management

  • 54.42% lower communication overhead than traditional source encoder/decoder; the vehicle-count model also improved MAE and MSE over GRU, LSTM, and FCN-rLSTM baselines.On TRANCOS, MAE fell 90.71%, 73.03%, and 19.1%, respectively, and MSE fell 103.91%, 77.74%, and 13.45%, respectively. The CNN-LSTM semantic encoder transmits vehicle-density information to a central traffic controller for prediction.Source: §4.4 · p. 38
  • Under low-SNR conditions, SMSC-FIR improved ReID, color-classification, and type-classification performance at 0 dB over sequential and random feature selection.Compared with sequential and random selection, ReID improved 40.0% and 212.2%, color classification 11.0% and 9.8%, and type classification 6.9% and 7.0%, respectively. Feature-importance ranking and dynamic coding-rate adjustment prioritize critical features.Source: §4.4 · p. 38
  • More than 25% mIoU improvement under low-SNR conditions and an average 14 dB reduction in CSI-estimation NMSE were reported for DMCE in multiuser traffic-scene communication.DMCE learns channel-effect data distributions to suppress noise in CSI estimation; experiments used 1,569 RGB-IR urban traffic image pairs.Source: §4.4 · p. 39
  • Semantic encoding and decoding of emergency-vehicle data enables traffic infrastructure to adjust signal durations, improving emergency response while minimizing disruption to other traffic flows.Scaling this scheduling approach to complex, high-traffic urban networks remains a challenge.Source: §4.4 · p. 39
  • The reviewed studies report reduced overhead and improved prediction accuracy for CNN-LSTM, low-SNR multitask performance for SMSC-FIR, and more accurate semantic recovery for DMCE [48].The CNN-LSTM model was trained and tested on TRANCOS, while SMSC-FIR experiments used VeRi-776.Source: §4.4 · p. 39
  • Semantic extraction, reliable communication architecture, and resource-management strategies support rapid event identification and real-time, accurate traffic-management decisions.The integrated technologies are presented as supporting traffic-system efficiency, safety, and sustainable urban traffic development.Source: §4.4 · p. 39
  • City- or region-scale traffic management must address data storage, processing, and analysis demands, inter-platform coordination, and possible policy and regulatory barriers.The passage calls for cooperation with government and regulatory authorities to update regulations for new technologies.Source: §4.4 · p. 40

4.5. Lessons Learned

  • Lessons learned emphasize task-oriented semantic frameworks for low-SNR perception and scheduling, while identifying complex-environment generalization, scalability, and privacy as continuing concerns.Edge computing and distributed knowledge sharing are identified as important for scalability, and federated learning as a privacy-preserving approach for sensitive applications such as remote diagnostics.Source: §4.5 · p. 40

5. Challenges and Future Research Directions

The survey identifies challenges in knowledge-base updates, semantic understanding, real-time and reliable communication, security and privacy, and standardization. It outlines technology and collaboration directions for addressing these challenges, while noting that IoV semantic communication still faces unresolved issues.

5.1. Knowledge Base Creation and Updating

  • Heterogeneous sources and complex traffic knowledge complicate knowledge-base construction, while current update mechanisms can miss IoV real-time demands.The survey suggests large models, federated learning, and blockchain for knowledge integration, updating, privacy-preserving collaboration, and reliable update records.Source: §5.1 · pp. 40–41

5.2. Semantic Understanding and Ambiguity

  • Heterogeneous data and ambiguous instructions challenge semantic understanding; proposed directions include unified models, context-aware fusion, knowledge graphs, and edge processing.Examples include sensor and infrastructure semantic mismatches and the underspecified instruction “slow down at the upcoming intersection.”Source: §5.2 · p. 41

5.3. Real-Time and Reliability Requirements

  • High-speed motion and simultaneous links threaten timely encoding and decoding; CT-MAC improves broadcast reliability and channel utilization through frame-length optimization and capture effect.The survey also calls for low-latency semantic encoding and decoding and error-control, multipath, or redundancy mechanisms for reliable delivery.Source: §5.3 · pp. 41–42

5.4. Security and Privacy Protection

  • Sensitive vehicle information and forged or tampered messages threaten privacy and safety; encryption can add latency, and detectors may generalize poorly to unknown attacks.Suggested directions include semantic-specific encryption, lightweight edge-based key distribution, and online adaptive detection.Source: §5.4 · pp. 42–43

5.5. Standardization and Regulation

  • Different semantic definitions and encoding methods across organizations and industries hinder seamless communication, motivating unified semantic standards and international alignment.The survey calls for collaboration among manufacturers, communication companies, and research institutions.Source: §5.5 · p. 43

5.6. Lessons Learned

  • Cross-cutting priorities include adaptive low-latency models, AI-driven knowledge fusion, lightweight encryption, unified ontologies, global standards, and collaboration across academia, industry, and policymakers.The survey highlights unresolved challenges in security, real-time knowledge-base updates, and standardization.Source: §5.6 · p. 43

6. Conclusions

The paper reviews semantic communication technologies in the IoV, from technical foundations and applications to challenges and future directions. It concludes that semantic communication offers broad prospects, while further research and collaboration are needed to support practical implementation.

6. Conclusions

  • The survey covers IoV semantic communication foundations, applications, challenges, and future directions.It reviews technologies intended to extract, transmit, and understand semantic information.Source: §6 · p. 43
  • Semantic communication can reduce redundant transmissions and improve communication efficiency, data processing, and intelligent decision support in the IoV.The paper frames these capabilities as technical support for safer, more efficient, and smarter transportation.Source: §6 · p. 43
  • Practical implementation still requires solutions to IoV semantic communication challenges, including through AI, edge computing, blockchain, and interdisciplinary collaboration.The conclusion highlights closer ties among academia, industry, and application stakeholders.Source: §6 · p. 44
Loading 2503.03767v2…