Recent advancements in pretrained foundation models have shown considerable promise across various machine learning tasks. However, their application in the broader IoT industry remains limited, particularly in IoT physical-layer authentication. In this domain, the presence of domain shift between training and deployment environments, combined with privacy and data security concerns, renders traditional domain adaptation methods that rely on source domain data impractical. Motivated by these challenges, source-free unsupervised domain adaptation (SFUDA) presents a more feasible solution. In this paper, we propose a novel SFUDA framework that leverages a pretrained generative foundation model to augment target domain data without requiring access to source domain information. Additionally, we integrate an uncertainty-aware pseudo-labeling strategy along with consistency regularization to further enhance the adaptation process. Experimental results validate that our approach significantly outperforms conventional techniques, providing an effective and robust solution for IoT physical-layer authentication under realistic constraints.

The Industrial Internet of Things(IIoT) deepens the collaborative computing between different devices and control layers. IIoT, as a dynamic, time-varying and complex environment, digital twin(DT) drives the real-time dynamic mapping between physical devices and twins to achieve global control of industrial production lines. This paper reviews the relevant research on the application and deployment of DT in the IIoT, which provides technical support for cross-production line data sharing and accurate decision-making. In particular, the related work of the multi-layer collaborative computing architecture built by federated learning(FL) in DT-IIoT is reviewed. We systematically explore the key technical breakthroughs of DT and FL in the IIoT, such as data security collaboration, dynamic resource scheduling, computing efficiency improvement, and cross-domain collaborative computing, from four technologies: edge computing, blockchain, deep reinforcement learning, and personalized learning. On this basis, we have explored and experimented in detail with fully decentralized SL in IIoT. SL-knowledge distillation(KD), it can solve the challenge of reducing the reliability of global model decisions in SL due to multi-source heterogeneous data in complex industrial scenarios. Experimental results show that SLKD outperforms other baselines by an average of 0.024 and 0.060 in recall and accuracy. In addition, we discuss the current challenges and future research directions.

Radio-frequency (RF) signal processing has seen significant advancements with the advent of deep learning, providing more accurate and efficient solutions for tasks such as signal classification and generation. However, most existing methods are heavily dependent on large labeled datasets, which are often scarce and costly to obtain in real-world RF environments. Furthermore, these approaches tend to be task-specific, limiting their ability to generalize across various RF applications. To address these challenges, this paper proposes RF-MAE, a self-supervised adaptive frequency masked autoencoder. RF-MAE leverages self-supervised learning (SSL) to capture intrinsic patterns from large-scale unlabeled RF data. Central to RF-MAE is a novel Adaptive Frequency Masked (AFM) strategy, which dynamically masks frequency components based on their energy distribution. Supported by a robust theoretical foundation, AFM ensures the model focuses on the most informative signal components, thereby enhancing generalization across RF tasks. By pretraining on unlabeled data and fine-tuning on specific tasks, RF-MAE significantly reduces the reliance on labeled datasets while improving adaptability across diverse RF signal processing tasks. Experimental results demonstrate that RF-MAE consistently outperforms traditional models, underscoring its potential to generalize across tasks and deliver superior performance in a wide range of RF signal applications.

Radio frequency fingerprint identification (RFFI) aims to identify subtle impairments in hardware devices, which play an important role in the mobile environment security community. To identify various mobile devices in the complex electromagnetic environment, deep learning methods such as convolutional neural networks (CNN) and recurrent neural networks (RNN) have been adopted to extract device hardware-related features. However, the single network structure has difficulty in comprehensive feature extraction, as many factors can introduce hardware impairments. In this paper, we propose a hybrid model termed switching dynamical deep network (SDDN) for RFFI tasks, which can jointly extract both coarse-grained radio frequency fingerprints (RFFs) and fine-grained RFFs. Additionally, the proposed hybrid model consists of a probabilistic part and a deterministic part. Specifically, in the probabilistic part, the switching linear dynamical systems (SLDS) are incorporated to establish the correspondence between the signal slice and the feature extraction network (FEN). In the deterministic part, multiple independent FENs are established to extract the RFFs. Moreover, to automatically determine the suitable number of FENs, a Bayesian nonparametric prior distribution is placed over the probabilistic part. Finally, an end-to-end parameter optimization method that is based on variational inference and stochastic gradient descent is proposed. Experiments on a real-life Wi-Fi dataset demonstrate the superiority of the proposed method over existing methods.

Source-free unsupervised domain adaptation (SFUDA) aims to improve performance in unlabeled target domain data without accessing source domain data. This is crucial in scenarios with data-sharing restrictions due to privacy or compliance constraints. Existing SFUDA approaches often rely on pseudo-labeling techniques based on entropy or confidence metrics. These often overlook fine-grained data features, resulting in noisy pseudo-labels that degrade model performance. To overcome this limitation, we develop a new method called fine-grained pseudo-labeling and feature alignment (FGPLFA) to enhance SFUDA's performance. FGPLFA starts with a gradient-based metric that integrates insights from both model knowledge and data features, creating a more reliable sample metric. To enhance fine granularity, the fine-grained pseudo-labeling (FGPL) module was introduced. This module clusters data based on the magnitude and direction of gradients, allowing for dataset partitioning into subsets at the sample level. The subsets are pseudo-labeled with category-specificity and domain specificity, establishing a multilevel granularity structure that reduces noisy pseudo-labels. Subsequently, the mean-covariance adjustment feature alignment (MCAFA) method was introduced. Features from the subsets are aligned in a specified sequence, enhancing model adaptability in the target domain. Extensive experiments conducted across multiple datasets validate the superiority of FGPLFA.

The quality of edge computing microservices is significantly influenced by their ability to perceive the spatiotemporal dynamics of user locations. Traditional approaches to microservice deployment in edge environments often rely on manual adjustments based on user position and base station load, which introduces substantial complexity and inefficiency. To address these challenges, we propose a novel methodology for spatiotemporal-aware microservice pre-deployment utilizing large language models (SAMPLE). By leveraging the predictive capabilities of spatiotemporal large language models, our approach enhances the microservice’s spatiotemporal awareness through trajectory forecasting. Additionally, we introduce an automated framework for generating optimal microservice deployment strategies based on the spatiotemporal relationships between users and services. Experimental results demonstrate that the proposed method significantly improves service quality by autonomously sensing user movement and dynamically adjusting deployment strategies, enhancing both the efficiency and responsiveness of edge services. The implementation code and datasets are available at https://github.com/ssea-lab/SAMPLE.

Satellite-terrestrial edge computing network (STECN) is emerging as a novel computation paradigm that enables a fashion of on-orbit computation, accommodating real-time processing of various types of computation tasks within a wider area. However, STECNs are incapable of meeting the diverse Quality-of-Service (QoS) requirements of various computational tasks without deploying dedicated infrastructures tailored to each type of task. Therefore, we studied the problem of joint network slicing and task offloading, which is modeled as a problem with a two-layer structure. At the higher level, slice tenants optimize their profits by determining resource allocations from the STECN, while at the lower layer, the users within each network slice maximize their utilities by deciding their offloading policies. In light of the complexity and the intricate interplay of these two problems, a multileader-disjoint-follower bi-level game is proposed. We show that both the leader’s game and the followers’ game admit at least one Nash equilibrium (NE). We then proposed two distributed algorithms that could prove to converge to the NE of these two games, respectively, without revealing any private information of all stakeholders. We evaluate the performance of our algorithms through extensive simulations, and the results demonstrate that our algorithms converge to the NE rapidly and can achieve superior performance gains compared with some known benchmarks.

Driven by the pursuit of enhanced performance, deep learning has recently seen rapid developments in the scaling of network architectures and parameters. However, this advancement has led to extremely high computational costs, undesirable in real-time and resource-limited scenarios. To address these challenges, we propose SwiftNet, a cost-efficient deep learning framework. Our novelty lies in SwiftNet's innovative multidimensional early-exit strategy that integrates seamlessly with existing neural network architectures. The framework includes additional branch classifiers concatenated to the backbone network, allowing high-confidence samples to exit early, thereby, reducing computational load. Unlike traditional methods, SwiftNet dynamically assesses confidence levels, ensuring only low-confidence samples proceed to subsequent classifiers or the final layer, optimizing resource usage without compromising accuracy. We have validated SwiftNet on multiple neural network models and datasets, demonstrating its ability to significantly reduce the computational cost of models while maintaining neural network performance.

With the proliferation of low-cost mobile edge devices, security and reliability have become crucial for mobile edge computing networks, especially for high-stakes applications. To facilitate it, Radio Frequency Fingerprint Identification (RFFI) has emerged as a promising physical layer security paradigm, offering a non-cryptographic and lightweight solution. However, most existing Deep Learning (DL) based RFFI methods operate under a closed-set assumption, limiting their ability to recognize devices not seen during training and posing a risk of misclassification. Addressing the open-set RFFI problem is critical for real-world deployments, where the system must handle both known and unknown devices, ensuring robust security in dynamic environments. In this paper, we propose OpenRFI, a novel test-time fine-tuning-based RFFI framework, consisting of two sequential stages: pre-training and test-time fine-tuning. During the pre-training stage, we design a data augmentation module, a feature extraction module, and an efficient hybrid loss function to minimize intra-class feature distances and tighten decision boundaries, enhancing the model's ability to distinguish between different classes. In the test-time fine-tuning stage, we introduce a fine-tuning dataset construction module and a full-parameter fine-tuning module to dynamically adapt to the test environment and capture information from unknown samples, further improving open-set recognition. We theoretically establish the performance boundary of the fine-tuning dataset construction method, providing insights into its robustness and scalability. Extensive numerical results based on an open source dataset demonstrate the effectiveness of the proposed OpenRFI framework in comparison with existing baselines.

One of the key challenges faced by 5G-Advanced and the forthcoming 6G networks is the provisioning of delay-critical services. Recently, the integration of network slicing and Multi-access Edge Computing (MEC) is regarded as a promising solution for this challenge. However, existing proposals for the integration are far from perfect with various drawbacks, such as rigid slicing, privacy leakage, and high signaling costs. In this paper, we investigate the problem of joint network slicing and computation offloading, which is formulated as a multi-leader-disjoint-follower Stackelberg game. We prove that the game is a potential game which has at least one global Nash Equilibrium (gNE). Then we propose a distributed algorithm that provably converges to a gNE of the game without revealing any private information of all stakeholders. The performance of our proposed algorithm is evaluated through simulations, which demonstrate that the algorithm converges to the gNE rapidly and outperforms a number of benchmark algorithms.

With the widespread adoption of 5G and Internet of Things (IoT) technologies, the low latency provided by edge computing has great importance for real-time processing. However, managing numerous simultaneous service requests poses a significant challenge to maintaining low latency. Current edge server task scheduling methods often fail to balance multiple optimization goals effectively. This paper introduces a novel task-scheduling approach based on Evolutionary Computing (EC) theory and heuristic algorithms. We model service requests as task sequences and evaluate various scheduling schemes during each evolutionary process using Large Language Models (LLMs) services. Experimental results show that our task-scheduling algorithm outperforms existing heuristic and traditional reinforcement learning methods. Additionally, we investigate the effects of different heuristic strategies and compare the evolutionary outcomes across various LLM services.

A prevailing assumption in existing deep learning research posits that data across source and target domains adhere to the independent and identically distributed (i.i.d.) assumption. However, this assumption often proves inadequate in real-world scenarios, leading to significant performance degradation when models encounter data with divergent distributions. To address this challenge, a novel unsupervised domain adaptation (UDA) algorithm, decoupling feature alignment (DFA), is introduced. The approach begins with the establishment of a robust theoretical framework, serving as the foundation for the mean-covariance adjustment feature alignment (MCAFA) algorithm. Simultaneously, a data decoupling (DD) module is introduced, effectively segregating target domain data into two subsets: one that mirrors the source domain and another that diverges markedly. Furthermore, a multidimensional alignment module is employed, leveraging the MCAFA algorithm and the DD module to align target data with source data across various layers and categories. Comprehensive evaluations on multiple data sets underscore the superiority of DFA.

Decentralized federated learning (DFL) has emerged as a conducive paradigm, facilitating a distributed privacy-preserving data collaboration mode in mobile edge computing (MEC) systems to bolster the expansion of artificial intelligence applications. Nevertheless, the dynamic wireless environment and the heterogeneity among collaborating nodes, characterized by skewed datasets and uneven capabilities, present substantial challenges for efficient DFL model training in MEC systems. Consequently, the design of an efficient collaboration strategy becomes essential to facilitate practical distributed knowledge sharing and cost reduction for MEC. In this paper, we propose an adaptive decentralized federated learning framework that enables heterogeneous nodes to learn tailored collaboration strategies, thereby maximizing the efficiency of the DFL training process in collaborative MEC systems. Specifically, we present an effective option critic-based collaboration strategy learning (OCSL) mechanism by decomposing the collaboration strategy model into two sub-strategies: local training strategy and resource scheduling strategy. In addressing inherent issues such as large-scale action space and overestimation in collaboration strategy learning, we introduce the option framework and a dual critic network-based approximation method within the OCSL design. We theoretically prove that the learned collaboration strategy achieves the Nash equilibrium. Extensive numerical results demonstrate the effectiveness of the proposed method in comparison with existing baselines.

In line with the increasing machine learning model inference accuracy, deep learning (DL) models have been increasingly applied to structured data for a wide spectrum of real-world applications, including product recommendations, online advertisement, healthcare analytics and risk analysis. However, unlike unstructured data, structured data is high-dimensional and sparse and therefore engenders a large number of parameters in DL, making DL models more prone to overfitting. To alleviate the overfitting problem, various regularization methods have been designed to constrain the model parameters as a means to control the model complexity. Unfortunately, these methods are often restricted to regularizing the parameter values directly without considering the intrinsic correlations and dependencies between attribute fields of structured data which is however key to effective structured data modeling. In this paper, we re-examine DL for structured data from a new perspective of attribute interactions. In particular, we seek to explicitly model and regularize the pairwise relationships between attribute fields of structured data, in a field-adaptive manner, via a proposed attentive and interpretable framework called ATT-Reg. Specifically, in this framework, a set of attentive weight matrices are introduced to each attribute field for modeling obviously different relationships with its neighboring attribute fields. Further, we derive from the Bayesian viewpoint a novel Attentive Regularization method for imposing adaptive regularization strengths on different pairs of attribute fields, based on the informativeness of their relationship, which is calculated using both data-driven information and functional dependency (FD) knowledge. Such adaptive regularization facilitates each attribute field to learn discriminative and diversified representations for more effective predictive analytics. We also develop a feature attribution method for supporting more interpretable predictions We validate the effectiveness of our ATT-Reg on six real-world datasets. Extensive experimental results show that ATT-Reg achieves significant improvement over state-of-the-art graph models, attentive models as well as regularization methods and supports an excellent degree of interpretation.

Data collaboration enables multiple parties to pool data for deriving meaningful data insights. However, data misuse and unlawful data collection have led to precautionary measures being imposed by individual organizations to guide against data leakage and abuse. As a response, decentralized federated learning (DFL) has emerged as an attractive paradigm to facilitate data collaboration while being amenable to privacy-preserving data and knowledge sharing, cost reduction, and prediction accuracy improvement. Unfortunately, the participating parties in DFL tend to be heterogeneous with skew datasets and uneven capabilities. Inevitably, training and transmission costs, and the presence of free-riders pose challenges to the adoption and participation of DFL. The absence of centralized parameter servers further exacerbates the problem of evaluating the contribution of each individual party. Therefore, an effective incentive mechanism is essential to promote data collaboration. In this paper, we propose a novel Incentive-aware Decentralized fEderated leArning (IDEA) framework for facilitating data collaboration. Specifically, we first design a customizable reward scheme for heterogeneous parties to optimize their respective objectives such as higher model accuracy, communication efficiency, and computational efficiency. To reward fairly to deserving parties while offering flexibility, we propose a novel multi-agent reinforcement learning (MARL) incentive mechanism, which enables heterogeneous parties to learn their own optimal collaboration policy. We then design an efficient decentralized data collaboration algorithm that supports the customizable reward scheme based on individual objective-specific collaboration policy. We theoretically prove that the algorithm achieves a Nash equilibrium, which ensures the fairness of the corresponding rewards for parties. We conduct extensive experiments to evaluate the performance of our proposed framework against four baselines on five real-world datasets. The results show that IDEA outperforms state-of-the-art methods in terms of effectiveness, efficiency, and accumulated reward.

Unmanned aerial vehicle (UAV) assisted wireless network has been recognized as an effective technology to facilitate the formation of a super flexible low-altitude platform for relieving the strain on traditional ground cellular systems. However, the on-demand deployment of the UAV-assisted wireless networks (OWN) becomes an essential yet challenging issue, as the constraints of UAVs’ location, resource provisioning, and demand distribution should be jointly considered. In this work, we investigate the OWN problem by proposing an autonomous learning framework (ALF) consisting of three sequential stages: demand prediction, proactive deployment, and resource allocation fine-tuning, which can be capable of autonomous network planning without reliance on manual operations in an extremely dynamic environment. In the demand prediction stage, we first design a dual transformer network (DTN) to capture the temporal and spatial dependencies of wireless traffic. We further reduce the computational complexity of DTN from quadratic time complexity to log-linear time complexity. In the proactive deployment stage, we jointly optimize the UAVs’ location and resource provisioning by proposing a modified general benders decomposition algorithm with a$\Gamma $-optimal convergence, where a learning-based discerning module is designed to accelerate the algorithm. In the resource allocation fine-tuning stage, we propose a simulated annealing-based algorithm to minimize the transmission rate degradation of users to reduce the bias caused by traffic demand prediction. Extensive numerical results based on an open source dataset demonstrate the effectiveness of the proposed methods in comparison with existing baselines.

Unmanned aerial vehicle (UAV) assisted wireless network is emerging as a promising technology to address the extremely high and dynamic traffic demands in future communication systems. In this paper, we investigate the on-demand deployment of UAV assisted wireless networks (OWN) problem. We propose an efficient autonomous learning framework (ALF), for learning a proactive and optimal on-demand deployment policy to complement terrestrial networks. In ALF, the OWN problem is solved in two co-related stages: the demand prediction stage and the proactive deployment stage. We first design a dual transformer network (DTN) to forecast the wireless traffic in the demand prediction stage. To decrease the complexity of DTN, we employ a patch embedding method and a modified self-attention scheme to improve the efficiency. With the predicted traffic demands, we jointly optimize the UAVs' location and wireless resource allocation by formulating it as a non-convex mixed integer nonlinear programming (MINLP) problem in the proactive deployment stage. To provide an efficient guaranteed solution to the MINLP problem, a multi-cut general benders decomposition algorithm is proposed to decompose the optimization problem into two subproblems. We theoretically prove that the proposed algorithm can achieve a T-optimal convergence. Extensive simulation results show the proposed solution outperforms existing baselines.

Radio Access Network (RAN) slicing is a promising architectural technology to address extremely diversified service demands and provide profitable business models for future mobile networks. In RAN slicing architecture, self-healing is an important functional module to minimize the impact of network failings on the performance of RAN slices. However, self-healing of burgeoning sliced RAN is vastly different from that of traditional RAN and has been rarely investigated. In this paper, we address the Self-healing of RAN Slice (SRANS) problem by modeling it as a Pareto optimization problem with the aim of maximizing the self-healing utilities of individual RAN slices. To deal with the weakness of diversity maintenance in traditional Pareto optimization methods, we propose a Generative Adversarial Network (GAN) based Pareto Optimization (GPO) framework. Specifically, we employ self-conditioned GANs to replace the offspring reproduction module in the traditional Evolutionary Algorithm (EA), where the insufficiency of diversity maintenance in EAs is effectively overcome. Furthermore, we theoretically prove that GPO framework is guaranteed to converge to the optimal Pareto solution set. Numerical results demonstrate that the convergence of proposed GPO framework can be expedited by enhancing the diversity of solution sets in solving the SRANS problem. Compared with traditional schemes, GPO can achieve significant performance gain in terms of the utilities and isolation level of repaired RAN slices.

Radio Access Network (RAN) slicing is a promising architectural technology to address extremely diversified service demands for future mobile networks. As an essential requirement for RAN slicing, self-healing is to provide services with certain quality requirements by minimizing the impact of mobile network failings. In this paper, we propose a Multi-objective Pareto Optimization based Self-healing (MPOS) scheme to solve the SRANS problem. We model the SRANS problem as a multi-objective optimization problem with aim of maximizing the self-healing profits of individual RAN slices and demonstrate the NP-hardness. In proposed MPOS scheme, we employ self-conditioned GANs to replace the offspring reproduction module in the traditional Multi-Objective Evolutionary Algorithm (MOEA), where the insufficiency of diversity maintenance in MOEA is effectively overcome. Furthermore, we theoretically prove that MPOS framework is guaranteed to converge to the optimal Pareto solution set with probability 1. Numerical results demonstrate that our MPOS scheme is effective in reducing the inverted generational distance of optimal Pareto solutions and achieving high profit and isolation level of RAN slices.

It is widely acknowledged that the agile reconfiguration of network slice according to traffic demand is of vital importance in 5G-and-beyond systems. Existing relevant works make reconfiguration decisions based either on point prediction of the uncertain demand, which lacks indications on how accurate it is, or on handcrafted uncertainty set with robust optimization, which may lead to resource over-provisioning due to the lack of prediction mechanism. To overcome these drawbacks, in this paper, we propose a predictor-optimizer framework that intelligently performs inter-slice reconfiguration with the aim of minimizing the energy consumption of serving these slices. Specifically, the predictor produces a prediction interval comprised of lower and upper bounds that bracket the future traffic demands with a prespecified probability. Then by regarding the prediction interval as the uncertainty set, we formulate the network slice reconfiguration problem as a Robust Mixed Integer Programming (RMIP). We solve this RMIP by using linearization technique and robust optimization. Numerical results demonstrate that the proposed framework outperforms traditional methods in terms of robustness and energy consumption. Meanwhile, the tradeoff between robustness and the energy consumption can be automatically adjusted according to the type of slice and traffic demands.

Due to the explosive growth of data traffic and poor indoor coverage, ultra-dense network has been introduced as a fundamental architectural technology for the 5G-and-beyond systems. As the telecom operator is shifting to the plug-and-play manner in mobile networks, network planning and optimization become difficult, especially in residential small-cell base stations (SBSs) deployment. Under this circumstance, severe inter-cell interference becomes inevitable which deteriorates network performance and the quality of service (QoS) of user equipments (UEs). In this paper, we propose a fully distributed self-learning interference mitigation (SLIM) scheme for autonomous networks under a model-free multi-agent reinforcement learning (MARL) framework. In SLIM, SBSs autonomously perceive surrounding interferences and determine downlink transmit power without necessity of signaling interaction between SBSs for mitigating interferences. To tackle the dimensional disaster of joint action in MARL model, we employ the Mean Field Theory to approximate the action value function, thus to greatly decrease the computational complexity. Simulation results based on 3GPP dual-stripe urban model demonstrate that SLIM outperforms conventional interference coordination schemes in mitigating interference while guaranteeing UEs'QoS.

It is widely acknowledged that network slicing can tackle the diverse usage scenarios and connectivity services that the 5G-and-beyond systems need to support. To guarantee performance isolation while maximizing network resource utilization under traffic uncertainty, network slice needs to be reconfigured adaptively. However, it is commonly believed that the fine-grained resource reconfiguration problem is intractable due to the extremely high computational complexity caused by the numerous variables. In this paper, we investigate network slice reconfiguration with aim of minimizing long-term resource consumption by exploiting Deep Reinforcement Learning (DRL). To address the curse of dimensionality of the problem, we propose to incorporate the Branching Dueling Q-network (BDQ) into DRL, to avoid some unnecessary calculations of Q-value by separating the Q-network into a shared value branch and a number of distributed advantage branches. Furthermore, the value branch and the advantage branch of each dimension are aggregated to derive the corresponding dimension's sub-Q-value. Then the best reconfiguration action is composed of the subactions in individual dimensions which are selected by €-greedy policy. Finally, we design an intelligent online network slice reconfiguration policy based on BDQ and extensive simulation experiments are conducted to validate the effectiveness of the proposed slice reconfiguration policy.

It is widely acknowledged that network slicing can tackle the diverse usage scenarios and connectivity services that the 5G-and-beyond system needs to support. To guarantee performance isolation while maximizing network resource utilization under dynamic traffic load, network slice needs to be reconfigured adaptively. However, it is commonly believed that the fine-grained resource reconfiguration problem is intractable due to the extremely high computational complexity caused by numerous variables. In this article, we investigate the reconfiguration within a core network slice with aim of minimizing long-term resource consumption by exploiting Deep Reinforcement Learning (DRL). This problem is also intractable by using conventional Deep Q Network (DQN), as it has a multi-dimensional discrete action space which is difficult to explore efficiently. To address the curse of dimensionality, we propose to exploit Branching Dueling Q-network which incorporates the action branching architecture into DQN to drastically decrease the number of estimated actions. Based on the discrete BDQ network, we develop an intelligent network slice reconfiguration algorithm (INSRA). Extensive simulation experiments are conducted to evaluate the performance of INSRA and the numerical results reveal that INSRA can minimize the long-term resource consumption and achieve high resource efficiency compared with several benchmark algorithms.