1 Introduction Although deep neural networks(DNNs)have made groundbreaking progress in various machine learning domains,their practical implementation is still significantly impeded by adversarial vulnerability[1].Adv...1 Introduction Although deep neural networks(DNNs)have made groundbreaking progress in various machine learning domains,their practical implementation is still significantly impeded by adversarial vulnerability[1].Adversarial training,the primary approach to enhance the adversarial robustness of DNNs,augments the training set with adversarial examples and applies adversarial regularization loss to improve robustness[2].However,finding models that achieve a reasonable trade-off between accuracy and robustness remains an unresolved challenge.In this paper,we propose the adoption of global probability constraints to stabilize model decision-making.Our contributions can be summarized as follows.展开更多
This paper proposes a data-and model-driven col-laborative resource scheduling method to maximize the spec-tral efficiency(SE)of cell-free(CF)downlink multiuser multiple-input multiple-output(MIMO)systems,subject to d...This paper proposes a data-and model-driven col-laborative resource scheduling method to maximize the spec-tral efficiency(SE)of cell-free(CF)downlink multiuser multiple-input multiple-output(MIMO)systems,subject to delay violation probability and power constraints.The method integrates the weighted minimum mean square error(WMMSE)algorithm within the safety reinforcement learn-ing(Safety-RL)framework.The original optimization problem is decomposed into two coupled subproblems.The Safety-RL algorithm leverages state features to determine user priority weights and allocate bandwidths,while the WMMSE algorithm calculates the precoding matrix and fur-ther schedules resources based on user priority weights to obtain the reward and costs of Safety-RL.Considering dy-namic user access in CF systems,a distributed algorithm with user scalability is also proposed.Simulation results demonstrate that the proposed approach improves the SE while meeting the different delay violation probability con-straints of users.Furthermore,the distributed algorithm of-fers comparable performance to the fully centralized method while considerably reducing model training overhead,par-ticularly as users dynamically access the system.展开更多
基金partially supported by the National Natural Science Foundation of China(Grant Nos.61772284 and 62476137)the Jiangsu Province Excellent Postdoctoral Program,and the Natural Science Research Start-up Foundation of Recruiting Talents of Nanjing University of Posts and Telecommunications(No.NY223213).
文摘1 Introduction Although deep neural networks(DNNs)have made groundbreaking progress in various machine learning domains,their practical implementation is still significantly impeded by adversarial vulnerability[1].Adversarial training,the primary approach to enhance the adversarial robustness of DNNs,augments the training set with adversarial examples and applies adversarial regularization loss to improve robustness[2].However,finding models that achieve a reasonable trade-off between accuracy and robustness remains an unresolved challenge.In this paper,we propose the adoption of global probability constraints to stabilize model decision-making.Our contributions can be summarized as follows.
基金The National Natural Science Foundation of China (No. 62271140, 62225107)the Natural Science Foundation of Jiangsu Province (No. BK20240174)+1 种基金the Fundamental Research Funds for the Central Universities (No. 2242022k60002)the Fund of Jiangsu Provincial Scientific Research Center of Applied Mathematics (No. BK20233002)。
文摘This paper proposes a data-and model-driven col-laborative resource scheduling method to maximize the spec-tral efficiency(SE)of cell-free(CF)downlink multiuser multiple-input multiple-output(MIMO)systems,subject to delay violation probability and power constraints.The method integrates the weighted minimum mean square error(WMMSE)algorithm within the safety reinforcement learn-ing(Safety-RL)framework.The original optimization problem is decomposed into two coupled subproblems.The Safety-RL algorithm leverages state features to determine user priority weights and allocate bandwidths,while the WMMSE algorithm calculates the precoding matrix and fur-ther schedules resources based on user priority weights to obtain the reward and costs of Safety-RL.Considering dy-namic user access in CF systems,a distributed algorithm with user scalability is also proposed.Simulation results demonstrate that the proposed approach improves the SE while meeting the different delay violation probability con-straints of users.Furthermore,the distributed algorithm of-fers comparable performance to the fully centralized method while considerably reducing model training overhead,par-ticularly as users dynamically access the system.
