Agentic AI networking (AgentNet) is a novel AI-native networking paradigm in which a large number of specialized AI agents collaborate to perform autonomous decisions, dynamic environmental adaptation, and complex missions. AgentNet has the potential to facilitate real-time network management and optimization functions, including self-configuration, self-optimization, and self-adaptation across diverse and complex environments, laying the foundation for fully autonomous networking systems. Despite its promise, AgentNet is still in the early stages of development and still lacks an effective networking framework to support automatic goal discovery, multi-agent self-orchestration, and task assignment. This paper proposes SANet, a novel semantic-aware AgentNet architecture for wireless networks. SANet can infer the semantic goal of the user and automatically assign agents associated with different layers of the network stack to fulfill the inferred goal. Motivated by the fact that AgentNet is a decentralized framework in which collaborating agents may generally have different and even conflicting objectives, we formulate the decentralized optimization of SANet as a multi-agent multi-objective problem, and focus on finding the Pareto-optimal solution for agents with distinct and potentially conflicting objectives. We propose three novel metrics for evaluating SANet: (the agents' objective) optimization error, (dynamic environment) generalization error, and (multi-objective) conflicting error. Furthermore, we develop a model partition and sharing (MoPS) framework in which large models, e.g., deep learning models, of different agents can be partitioned into shared and agent-specific parts that are jointly constructed and deployed according to agents' local computational resources. Two decentralized optimization algorithms, static-weighting and dynamic-weighting algorithms, are introduced to optimize the above three metrics. A bandwidth-adaptive compression framework is also proposed to enable different agents to perform in situ compression of their intermediate embeddings, dynamically adjusting to localized resource constraints and task requirements. We derive theoretical bounds for all these performance metrics and prove that there exists a three-way tradeoff among optimization, generalization, and conflicting errors. Finally, to validate our theoretical results, we develop an open-source Radio Access Network (RAN) and core network-based hardware prototype that implements three Transformer-based time-series prediction agents to interact with three different layers of the network. Experimental results show that the proposed MoPS framework achieves performance gains of up to$14.61\%$while requiring only$44.37\%$of the Floating-Point Operations (FLOPs) for inference at each agent compared to state-of-the-art algorithms. Also, compared to the static-weighting algorithm, the dynamic-weighting algorithm achieves up to$83.81\%$reduction in training errors caused by conflicting objectives.

Agentic AI networking (AgentNet) has attracted significant interest due to its promising potential to move traditional AI-based networking solutions beyond closed-loop and passive learning to proactive interaction and goal-driven action, offering a path to self-learning and generally intelligent networking systems. Despite its promise, ensuring the security and trustworthiness of such systems presents significant challenges, particularly concerning identity management, agent capability verification, and data integrity during collaborative learning. To address these issues, this paper proposes TrustAgentNet, a novel consortium blockchain-based framework for unified and trusted agent identification, traceable skillset and tag descriptions, and secure on-chain collaborative learning in AgentNet. In TrustAgentNet, a chain of skillset (CoS) is introduced, consisting of a skillset chain to distributedly store all the verified skillsets and associated tags, and a dedicated training chain for each distinct skillset can be jointly constructed and maintained by the authorized agents using a collaborative learning-based approach. Theoretical analysis suggests that there exists a three-way trade-off among the security level, skillset performance, and resource cost. This tradeoff is also empirically validated by the experimental results obtained from a hardware prototype implemented based on a Hyperledger Fabric-based consortium blockchain. To verify the practical performance of TrustAgentNet, we consider a real-world scenario of multi-agent collaborative learning under malicious attack. Experimental results suggest that TrustAgentNet can effectively guarantee the security of skillset training and enable rapid response and recovery from potential attacks within seconds.

Agentic AI networking (AgentNet) has attracted significant interest recently due to its promising potential in supporting proactive learning and seamless collaboration among distributed task-oriented agents in various environments. However, existing solutions face inherent dilemmas. On the one hand, developing a single globalized model that can generalize well for diverse agents incurs inconsistent and unreliable performance due to the neglect of their distinct deployment environments. On the other hand, constructing personalized models that are tailored according to the individual needs of each agent often suffers from resource inefficiency and under-utilization of shared knowledge among agents. To overcome these challenges, we propose MAN, a novel Meta learning-based AgentNet architecture that optimally balances generalization and personalization for all the agents when performing different tasks in dynamic environments. Specifically, MAN adopts a bi-level optimization framework to develop foundation meta-models as the shared initialization of all agents, and each agent can then fine-tune the meta model to enable personalized deployment. We derive theoretical bounds on both global generalization and local personalization errors, demonstrating that the fundamental tradeoff between these two can only be optimized but cannot be fully eliminated. Extensive experiments on real-world datasets validate the performance of the proposed MAN and confirm the generality of our theoretical insights.

Data-driven AI models, particularly deep learning-based wireless channel estimation solutions, have exhibited promising potentials in modeling the intricate, non-linear relationships between environmental conditions and wireless channel characteristics. However, multi-environment model training requires high computational and communication resource costs. Also, each AI model has limited generality and cannot be directly adopted to new and unknown environments that have different channel conditions, compared to its training dataset. To address these limitations, this paper proposes LoRAT, a low-rank model adaptation and transfer framework for efficient multi-environment channel estimation model development. In LoRAT, a number of foundation models are first trained for a limited number of known environments, and then a simple low-rank model transfer approach is proposed to enable quick and efficient transfer of these foundation models to new unknown environments. We consider the uplink-based downink channel state information (CSI) estimation problem for an FDD wireless system as an example to describe the implementation details and evaluate performance of LoRAT. In this case, we train a set of attention-based diffusion models as foundation models for downlink CSI estimations in the known environments and we show that transferring the low-rank parts of these foundation models is capable of constructing environment-specific CSI estimation models for any new unknown environment. Our proposed LoRAT does not require any labeled datasets in the unknown environments and, since the model transfer only requires to calculate the low-rank part of the model parameters, the computational and communication cost is much lower than the traditional AI model-based solutions, especially for wireless systems across a large number of different environments. Extensive experiments have been conducted to compare the performance of LoRAT with state-of-the-art solutions. Our experimental results suggest that LoRAT provides up to 15% and 265% improvement in accuracy of multi-environment channel estimation, compared to FIRE and codebook-based solution, respectively.

Agentic AI networking (AgentNet) is a novel AI-native networking paradigm that relies on a large number of specialized AI agents to collaborate and coordinate for autonomous decision-making, dynamic environmental adaptation, and complex goal achievement. It has the potential to facilitate real-time network management alongside capabilities for self-configuration, self-optimization, and self-adaptation across diverse and complex networking environments, laying the foundation for fully autonomous networking systems in the future. Despite its promise, AgentNet is still in the early stage of development, and there still lacks an effective networking framework to support automatic goal discovery and multi-agent self-orchestration and task assignment. This paper proposes SANNet, a novel semantic-aware agentic AI networking architecture that can infer the semantic goal of the user and automatically assign agents associated with different layers of a mobile system to fulfill the inferred goal. Motivated by the fact that one of the major challenges in AgentNet is that different agents may have different and even conflicting objectives when collaborating for certain goals, we introduce a dynamic weighting-based conflict-resolving mechanism to address this issue. We prove that SANNet can provide theoretical guarantee in both conflict-resolving and model generalization performance for multi-agent collaboration in dynamic environment. We develop a hardware prototype of SANNet based on the open RAN and 5GS core platform. Our experimental results show that SANNet can significantly improve the performance of multi-agent networking systems, even when agents with conflicting objectives are selected to collaborate for the same goal.

There is still lacking a simple and comprehensive framework to model and optimize the overall energy consumption of an FEI network, especially in heterogeneous scenarios. This paper proposes a comprehensive framework to characterize the overall energy consumption of FEI networks. The computation and communication overhead as well as the number of coordination rounds required to train a satisfactory model are analytically modeled and evaluated. We investigate and compare the energy consumption of FEI networks with two popular distributed algorithmic implementations: FedAvg and FedMeta. We observe that although FedMeta consumes more energy than FedAvg in each single coordination round, the overall energy consumption of FedMeta is much lower than that of FedAvg. Finally, we evaluate the energy consumption of both algorithms based on a hardware prototype. Numerical results show that the overall energy consumption of FedMeta is 77.9% less than that of FedAvg.

This paper focused on the illumination model of machined surface based on micro-image. According to micro-image forming condition, the theory that the image brightness is related to the microfacet topography and surface reflection characteristics is presented. The distribution rule of micro-topography and reflection characteristics of sample surface is analyzed according to the measured data. An illumination mode of machined surface is established based on the analysis result, and the model parameters are obtained by using the simulated annealing algorithm. Verification results show that this proposed model can improve the simulation accuracy significantly and describe the lighting effect of machined surface. The research will provide a new idea and method for the 3D reconstruction.