Selected work

Representative Papers

Agentic AI Empowered Intent-Based Networking for 6G

Jan 10, 2026arXiv.org

This work addresses the challenge in existing intent-based networking (IBN) approaches of simultaneously achieving flexibility and interpretability in natural language understanding while strictly enforcing technical constraints—a key bottleneck for 6G autonomous orchestration. To bridge this gap, the authors propose a hierarchical multi-agent framework that integrates large language models (LLMs) with domain-expert agents. Leveraging the ReAct reasoning-action loop, the system collaboratively decomposes high-level natural language intents into network slice configurations compliant with RAN and core network constraints. This architecture represents the first approach to enable interpretable, constraint-aware, and iteratively reasoned automatic translation from intent to configuration. Experimental results demonstrate significant performance gains over rule-based systems and direct LLM prompting across diverse benchmark scenarios, validating its effectiveness in O-RAN deployments and highlighting the critical role of context-aware prompt engineering in network automation.

1 citationsRead paper

SimuGen: Multi-modal Agentic Framework for Constructing Block Diagram-Based Simulation Models

May 28, 2025arXiv.org

Current large language models (LLMs) lack sufficient Simulink-domain pretraining data, rendering them unreliable for generating complete, executable Simulink simulation models directly from natural-language requirements. To address this, we propose the first multimodal agent framework specifically designed for Simulink modeling. Our approach integrates graph-structured visual understanding of Simulink diagrams, a domain-specific knowledge base, and a modular role-based collaboration mechanism—featuring specialized agents such as an investigator and a debug locator—to enable interpretable and reproducible end-to-end model generation. Crucially, the framework jointly models the visual representation and symbolic logic of Simulink models, supporting automated generation, debugging, and formal verification of simulation models from textual specifications. Evaluated on representative control and signal processing tasks, our method achieves significant improvements in code generation accuracy and structural completeness, demonstrating both technical efficacy and engineering practicality.

1 citationsRead paper

Can We Trust AI in 6G? Verifiable and Auditable AI-Driven Trustworthy Wireless Networks

Jul 28, 2026

This work addresses the critical lack of verifiability and auditability of artificial intelligence (AI) in current 6G networks, which undermines confidence in the correctness and regulatory compliance of AI-driven decisions and impedes trustworthy deployment in mission-critical communication infrastructure. To bridge this gap, the paper proposes a mechanistic auditing framework that uniquely integrates mechanistic interpretability with 3GPP protocol specifications. By leveraging internal representation monitoring, causal analysis, and dedicated verification agents, the approach establishes an “audit-native” network architecture. This architecture enables both pre-deployment certification and continuous runtime auditing, operationalized through a principled three-step auditing methodology. The study thus provides a comprehensive foundation—including a theoretical framework, a system prototype, and a practical implementation pathway—for standardized validation of AI functionalities in 6G systems.

0 citationsRead paper

From Traditional Automation to Embodied Wireless Intelligence: Vision-Language-Action Empowered Physics-Aware Communication Networks

Jun 11, 2026

This work addresses the limited environmental awareness of existing wireless network automation systems, which optimize performance metrics without accounting for real-world propagation conditions. To bridge this gap, the paper introduces the embodied intelligence–enabled base station (eBS), pioneering the integration of embodied intelligence into wireless communications. The proposed system features a vision–language–action (VLA) pipeline that endows base stations with contextual perception, causal physical reasoning, and physics-aware action generation capabilities. It employs a two-layer asynchronous architecture: a semantic planner leverages state-of-the-art vision-language models to produce structured commands, while a tactical controller executes real-time adjustments. Experiments demonstrate that a single VLA pipeline—without fine-tuning—achieves zero-shot material reasoning, cross-view generalization, and dynamic event prediction prior to signal degradation, thereby advancing wireless networks from rule-driven paradigms toward embodied intelligence.

0 citationsRead paper

TWIST: Closed-Loop token Synchronization for Application-Aware Wireless Digital Twins

May 26, 2026

This work addresses the challenge of efficiently synchronizing the semantic state of a physical environment with its digital twin under limited and time-varying wireless resources, where conventional pixel-level or uniformly protected transmission proves suboptimal. To this end, the authors propose the TWIST framework, which encodes observations into semantic tokens grouped by task relevance and dynamically applies non-uniform error protection conditioned on channel states and application priorities. TWIST further incorporates a confidence-guided token erasure and semantic completion mechanism to enable closed-loop synchronization. Experimental results demonstrate that TWIST significantly improves traffic state inference accuracy and semantic synchronization performance in dynamic road scenarios while reducing average synchronization overhead.

0 citationsRead paper

Recent publications

Latest Papers

Can We Trust AI in 6G? Verifiable and Auditable AI-Driven Trustworthy Wireless Networks

Jul 28, 2026

This work addresses the critical lack of verifiability and auditability of artificial intelligence (AI) in current 6G networks, which undermines confidence in the correctness and regulatory compliance of AI-driven decisions and impedes trustworthy deployment in mission-critical communication infrastructure. To bridge this gap, the paper proposes a mechanistic auditing framework that uniquely integrates mechanistic interpretability with 3GPP protocol specifications. By leveraging internal representation monitoring, causal analysis, and dedicated verification agents, the approach establishes an “audit-native” network architecture. This architecture enables both pre-deployment certification and continuous runtime auditing, operationalized through a principled three-step auditing methodology. The study thus provides a comprehensive foundation—including a theoretical framework, a system prototype, and a practical implementation pathway—for standardized validation of AI functionalities in 6G systems.

0 citationsRead paper

From Traditional Automation to Embodied Wireless Intelligence: Vision-Language-Action Empowered Physics-Aware Communication Networks

Jun 11, 2026

This work addresses the limited environmental awareness of existing wireless network automation systems, which optimize performance metrics without accounting for real-world propagation conditions. To bridge this gap, the paper introduces the embodied intelligence–enabled base station (eBS), pioneering the integration of embodied intelligence into wireless communications. The proposed system features a vision–language–action (VLA) pipeline that endows base stations with contextual perception, causal physical reasoning, and physics-aware action generation capabilities. It employs a two-layer asynchronous architecture: a semantic planner leverages state-of-the-art vision-language models to produce structured commands, while a tactical controller executes real-time adjustments. Experiments demonstrate that a single VLA pipeline—without fine-tuning—achieves zero-shot material reasoning, cross-view generalization, and dynamic event prediction prior to signal degradation, thereby advancing wireless networks from rule-driven paradigms toward embodied intelligence.

0 citationsRead paper

TWIST: Closed-Loop token Synchronization for Application-Aware Wireless Digital Twins

May 26, 2026

This work addresses the challenge of efficiently synchronizing the semantic state of a physical environment with its digital twin under limited and time-varying wireless resources, where conventional pixel-level or uniformly protected transmission proves suboptimal. To this end, the authors propose the TWIST framework, which encodes observations into semantic tokens grouped by task relevance and dynamically applies non-uniform error protection conditioned on channel states and application priorities. TWIST further incorporates a confidence-guided token erasure and semantic completion mechanism to enable closed-loop synchronization. Experimental results demonstrate that TWIST significantly improves traffic state inference accuracy and semantic synchronization performance in dynamic road scenarios while reducing average synchronization overhead.

0 citationsRead paper

TONIC: Token-Centric Semantic Communication for Task-Oriented Wireless Systems

May 20, 2026

This work addresses the mismatch between conventional bit-fidelity-oriented wireless communication and the semantic units (e.g., tokens) required by downstream foundation models, which leads to inefficient resource usage and performance degradation. To bridge this gap, the authors propose a task-oriented semantic communication framework that treats tokens as the fundamental transmission unit. At the transmitter, utility-aware unequal error protection is applied based on task relevance, while at the receiver, a confidence-gated mechanism combined with a Transformer-based completion model converts detrimental errors into recoverable erasures. The architecture is modular and interpretable, underpinned by a utility-aware Bayesian risk theory that guides gate design. Experiments demonstrate that the proposed method significantly outperforms traditional separation-based schemes, pixel-domain DeepJSCC, and token-domain baselines in image classification tasks across AWGN, Rayleigh, and Rician fading channels.

0 citationsRead paper

GETA-3DGS: Automatic Joint Structured Pruning and Quantization for 3D Gaussian Splatting

May 03, 2026

This work addresses the limitations of existing 3D Gaussian Splatting (3DGS) compression methods, which process pruning and quantization in separate stages, rely on manual parameter tuning, and struggle to generalize under bitrate or quality constraints. To overcome these issues, we propose the first end-to-end framework that jointly optimizes structured pruning and quantization. Our approach introduces a 3DGS-aware Quantization-Aware Dependency Graph (QADG), rendering-aware saliency scoring, and heterogeneous mixed-precision allocation per attribute to automatically optimize rate-distortion performance. The method eliminates the need for scene-specific thresholds and achieves approximately 5× storage compression across multiple benchmarks, significantly outperforming uniform 6-bit quantization while remaining compatible with existing entropy coding schemes.

0 citationsRead paper