Deep neural network(DNN)models have been widely used in e-commerce,games,auto-mobiles,manufacturing,and so on.Improper structure,parameters,activation function,or incorrect loss function of the DNN models may cause de...Deep neural network(DNN)models have been widely used in e-commerce,games,auto-mobiles,manufacturing,and so on.Improper structure,parameters,activation function,or incorrect loss function of the DNN models may cause defects in performance or secu-rity.As a result,there are some researches that focus on repairing DNN such as MODE and Apricot.However,the cost of repairing is high or the repair may lead to overfit-ting.In order to solve this problem,we propose GenMuNN,which is a Mutation-Based Approach to Repair Deep Neural Network Models.First,it analyzes the importance of the weights of the neurons in each layer of the DNN model to the correctness of the final prediction results,and ranks the weights according to the influence on the prediction results of the DNN model.Second,mutation is performed to generate mutants based on the rank of weights,and genetic algorithms are used to select mutants for the next round of mutation until the stop condition is touched.Experiments are carried on a set of DNN models which are trained with the MNIST dataset.The experimental results show that GenMuNN can improve the accuracy of the DNN models.展开更多
As the system becomes more intelligent and embedded,the operating environment is gradually changed from closed,static,and controllable to more open,dynamic,and difficult to control.As a result,the design of complex so...As the system becomes more intelligent and embedded,the operating environment is gradually changed from closed,static,and controllable to more open,dynamic,and difficult to control.As a result,the design of complex software systems is faced with many challenges arising from the uncertainty of the environment(UoE).On the one hand,ignoring the UoE to manually describe requirements is not only a tough job,but it can also hinder the discovery of potential requirements;on the other hand,it is a challenge to integrate the representation of and reasoning of UoE into the process of modeling complex systems.Based on the analysis of the characteristics of complex systems engineering,this paper takes solving the UoE caused by stakeholders prefer-ences and complex environment context as the entry point and designs a fuzzy con-trol decision-making framework.Specifically,the framework contributes to the spiral of complex systems while solving the UoE by constructing a closed-loop intelligent sys-tem based on automatic data flow between information space and physical space for environment sensing,uncertainty analysis,requirements mining,fuzzy reasoning,deci-sion execution as well as feedback optimization.Finally,the framework is validated with a concrete example of an adaptive treadmill system based on the support tools developed.展开更多
基金This work was supported by the Beijing Information Science and Technology Uni-versity“Qin-Xin Talent”Cultivation Project(No.QXTCP C201906)the Bei-jing Information Science and Technology University Research Level Improvement Project(No.2020KYNH214)the Science and Technology Project of the Bei-jing Municipal Education Commission(No.KM201811232016)。
文摘Deep neural network(DNN)models have been widely used in e-commerce,games,auto-mobiles,manufacturing,and so on.Improper structure,parameters,activation function,or incorrect loss function of the DNN models may cause defects in performance or secu-rity.As a result,there are some researches that focus on repairing DNN such as MODE and Apricot.However,the cost of repairing is high or the repair may lead to overfit-ting.In order to solve this problem,we propose GenMuNN,which is a Mutation-Based Approach to Repair Deep Neural Network Models.First,it analyzes the importance of the weights of the neurons in each layer of the DNN model to the correctness of the final prediction results,and ranks the weights according to the influence on the prediction results of the DNN model.Second,mutation is performed to generate mutants based on the rank of weights,and genetic algorithms are used to select mutants for the next round of mutation until the stop condition is touched.Experiments are carried on a set of DNN models which are trained with the MNIST dataset.The experimental results show that GenMuNN can improve the accuracy of the DNN models.
文摘As the system becomes more intelligent and embedded,the operating environment is gradually changed from closed,static,and controllable to more open,dynamic,and difficult to control.As a result,the design of complex software systems is faced with many challenges arising from the uncertainty of the environment(UoE).On the one hand,ignoring the UoE to manually describe requirements is not only a tough job,but it can also hinder the discovery of potential requirements;on the other hand,it is a challenge to integrate the representation of and reasoning of UoE into the process of modeling complex systems.Based on the analysis of the characteristics of complex systems engineering,this paper takes solving the UoE caused by stakeholders prefer-ences and complex environment context as the entry point and designs a fuzzy con-trol decision-making framework.Specifically,the framework contributes to the spiral of complex systems while solving the UoE by constructing a closed-loop intelligent sys-tem based on automatic data flow between information space and physical space for environment sensing,uncertainty analysis,requirements mining,fuzzy reasoning,deci-sion execution as well as feedback optimization.Finally,the framework is validated with a concrete example of an adaptive treadmill system based on the support tools developed.