作者:
Chen, JinzhuoXu, YongnanSun, WeizeHuang, LeiShenzhen Univ
Coll Elect & Informat Engn Guangdong Key Lab Intelligent Informat Proc Shenzhen Guangdong Peoples R China Shenzhen Univ
Coll Elect & Informat Engn Shenzhen Key Lab Adv Nav Technol Coll Elect & Inf Shenzhen Guangdong Peoples R China
Over the pass decade, deep neural network (DNN) has been widely applied in various applications. To alleviate the storage and computation requirement of the complicated DNNs, network compression methods are developed....
详细信息
Over the pass decade, deep neural network (DNN) has been widely applied in various applications. To alleviate the storage and computation requirement of the complicated DNNs, network compression methods are developed. The sparse structure learning methods based on multi-objective optimization have been proven to be valid to balance the sparsity of the network model and network performance. However, when multiple applications are deployed on one single platform simultaneously, these methods become inefficient because each network model for each application needs to be trained and optimized individually. In this article, a multi-objective, multi-application sparse learning model is proposed to optimize multiple targets from a set of applications together. The joint network structure is first proposed. After a pre-training of the network model, a joint multi-objective evolutionary algorithm is derived to solve the optimization problems. Note that an improved initialization method for parent model generation is also developed. Finally, based on the joint loss between the objectives, fine tuning is used to compute the final models with good performance. The proposed method is evaluated under different datasets with a comparison to the state-of-the-art approaches, and experimental results demonstrate that the multi-application optimization model can give much better performance than the single-application optimization ones, especially in the case that different datasets are involved simultaneously.
Many existing deep learning models are vulnerable to adversarial examples that are imperceptible to humans. To address this issue, various methods have been proposed to design network architectures that are robust to ...
详细信息
Many existing deep learning models are vulnerable to adversarial examples that are imperceptible to humans. To address this issue, various methods have been proposed to design network architectures that are robust to one particular type of adversarial attacks. It is practically impossible, however, to predict beforehand which type of attacks a machine learn model may suffer from. To address this challenge, we propose to search for deep neural architectures that are robust to five types of well-known adversarial attacks using a multi-objective evolutionary algorithm. To reduce the computational cost, a normalized error rate of a randomly chosen attack is calculated as the robustness for each newly generated neural architecture at each generation. All non-dominated network architectures obtained by the proposed method are then fully trained against randomly chosen adversarial attacks and tested on two widely used datasets. Our experimental results demonstrate the superiority of optimized neural architectures found by the proposed approach over state-of-the-art networks that are widely used in the literature in terms of the classification accuracy under different adversarial attacks. (c) 2021 Elsevier B.V. All rights reserved.
Energy consumption, production cost, and efficiency are highly concerned by decision makers of energyintensive and high -cost industrial production systems. Intelligent production scheduling is a necessary means to ac...
详细信息
Energy consumption, production cost, and efficiency are highly concerned by decision makers of energyintensive and high -cost industrial production systems. Intelligent production scheduling is a necessary means to achieve their optimization. This work delves into a novel multi -objective production scheduling problem arising from a steel hot -rolling process, which is a representative energy -intensive and high -cost industrial process. The challenge of the problem involves scheduling customized production jobs subject to intricate process constraints with the goal to minimize three objective functions, i.e., energy consumption, setup cost, and the number of tardy jobs. A mixed integer linear programming model is formulated for the problem. In order to solve it, an improved multi -objectiveevolutionaryalgorithm based on decomposition is presented. The algorithm incorporates problem -specific encoding and model -based decoding mechanisms, rendering it wellsuited for addressing the concerned multi -constrained multi -objective optimization problem. The introduced modified Tchebycheff approach mitigates the impact of objective functions with varying value ranges on the algorithm's convergence. Additionally, a Metropolis acceptance criterion is integrated to facilitate the escape from local optimal solutions, enhancing the algorithm's global optimization capability. Numerous experiments are conducted to verify the effectiveness of the improvements and to compare the performance of the presented algorithm against its competitive peers. The results demonstrate its high performance, suggesting its significant potential for its application to steel hot -rolling systems.
Planning under complex uncertainty often asks for plans that can adapt to changing future conditions. To inform plan development during this process, exploration methods have been used to explore the performance of ca...
详细信息
Planning under complex uncertainty often asks for plans that can adapt to changing future conditions. To inform plan development during this process, exploration methods have been used to explore the performance of candidate policies given uncertainties. Nevertheless, these methods hardly enable adaptation by themselves, so extra efforts are required to develop the final adaptive plans, hence compromising the overall decisionmaking efficiency. This paper introduces Reinforcement Learning (RL) that employs closed -loop control as a new exploration method that enables automated adaptive policy -making for planning under uncertainty. To investigate its performance, we compare RL with a widely -used exploration method, multi -objectiveevolutionaryalgorithm (MOEA), in two hypothetical problems via computational experiments. Our results indicate the complementarity of the two methods. RL makes better use of its exploration history, hence always providing higher efficiency and providing better policy robustness in the presence of parameter uncertainty. MOEA quantifies objective uncertainty in a more intuitive way, hence providing better robustness to objective uncertainty. These findings will help researchers choose appropriate methods in different applications.
The product form evolutionary design based on multi-objective optimization can satisfy the complex emotional needs of consumers for product form, but most relevant literatures mainly focus on single-objective optimiza...
详细信息
The product form evolutionary design based on multi-objective optimization can satisfy the complex emotional needs of consumers for product form, but most relevant literatures mainly focus on single-objective optimization or convert multiple-objective optimization into the single objective by weighting method. In order to explore the optimal product form design, we propose a hybrid product form design method based on back propagation neural networks (BP-NN) and non-dominated sorting genetic algorithm-II (NSGA-II) algorithms from the perspective of multi-objective optimization. First, the product form is deconstructed and encoded by morphological analysis method, and then the semantic difference method is used to enable consumers to evaluate product samples under a series of perceptual image vocabularies. Then, the nonlinear complex functional relation between the consumers' perceptual image and the morphological elements is fitted with the BP-NN. Finally, the trained BP-NN is embedded into the NSGA-II multi-objective evolutionary algorithm to derive the Pareto optimal solution. Based on the hybrid BP-NN and NSGA-II algorithms, a multi-objective optimization based product form evolutionary design system is developed with the electric motorcycle as a case. The system is proved to be feasible and effective, providing theoretical reference and method guidance for the multi-image product form design.
Purpose-Accurately reproducing physiological and time-varying variables in cardiac bioreactors is a difficult task for conventional control methods. This paper presents a new controller based on a genetic algorithm fo...
详细信息
Purpose-Accurately reproducing physiological and time-varying variables in cardiac bioreactors is a difficult task for conventional control methods. This paper presents a new controller based on a genetic algorithm for the control of a cardiac bioreactor dedicated to the study and conditioning of heart valve substitutes. Methods-A multi-objective genetic algorithm was designed to obtain an accurate simultaneous reproduction of physiological periodic time functions of the three most relevant variables characterizing the blood flow in the aortic valve. These three controlled variables are the flow rate and the pressures upstream and downstream of the aortic valve. Results-Experimental results obtained with this new algorithm showed an accurate dynamic reproduction of these three controlled variables. Moreover, the controller can react and adapt continuously to changes happening over time in the cardiac bioreactor, which is a major advantage when working with living biological valve substitutes. Conclusion-The strong non-linear interaction that exists between the three controlled variables makes it difficult to obtain a precise control of any of these, let alone all three simultaneously. However, the results showed that this new control algorithm can efficiently overcome such difficulties. In the particular field of bioreactors reproducing the cardiovascular environment, such a flexible, versatile and accurate reproduction of these three interdependent controlled variables is unprecedented.
In this paper, a cooperative-competitive multi-objectiveevolutionary fuzzy system called E2PAMEA is presented for the extraction of emerging patterns in big data environments. E2PAMEA follows an adaptive schema to au...
详细信息
In this paper, a cooperative-competitive multi-objectiveevolutionary fuzzy system called E2PAMEA is presented for the extraction of emerging patterns in big data environments. E2PAMEA follows an adaptive schema to automatically employ different genetic operators according to the learning needs, which avoid the tuning of some parameters. It also employs a token-competition-based procedure for updating an elite population where the best set of patterns found so far is stored. In addition, a novel MapReduce procedure for an efficient computation of the evaluation function employed for guiding the search process is proposed. The method, called Bit-LUT employs a pre-evaluation stage where data is represented as a look-up table made of bit sets. This look-up table can be employed later in the chromosome evaluation by means of bitwise operations, reducing the computational complexity of the process. The experimental study carried out shows that E2PAMEA is a promising alternative for the extraction of high-quality emerging patterns in big data. In addition, the proposed Bit-LUT evaluation shows a significant improvement on efficiency with a great scalability capacity on both dimensions of data, which enables the processing of massive datasets faster than other alternatives. (C) 2020 Elsevier B.V. All rights reserved.
A gradient-based search method (GBSM) is developed to solve multi-objective optimiza-tion problems. It uses the multi-objective gradient information to construct descent direc-tions, i.e., Pareto descent directions (P...
详细信息
A gradient-based search method (GBSM) is developed to solve multi-objective optimiza-tion problems. It uses the multi-objective gradient information to construct descent direc-tions, i.e., Pareto descent directions (PDDs), to accelerate the convergence. In addition, a multi-objective evolutionary algorithm based on decomposition is adopted to improve the diversity. The comparisons between GBSM with several selected multi-objective evolu-tionary algorithms and gradient based algorithms on benchmark functions indicate that the proposed method performs competitively and effectively. (c) 2021 Elsevier Inc. All rights reserved.
The multi-offspring method has been recognized as an efficient approach to enhance the performance of multi-objective evolutionary algorithms. However, some pre-screening strategies should be used when a multi-offspri...
详细信息
The multi-offspring method has been recognized as an efficient approach to enhance the performance of multi-objective evolutionary algorithms. However, some pre-screening strategies should be used when a multi-offspring-assisted multi-objective evolutionary algorithm is used to solve computationally expensive problems. So far, there is no any reported comprehensive study that compares the effects of different pre-screening strategies on the performance of the multi-offspring-assisted multi-objective evolutionary algorithms. In this paper, four pre-screening strategies (convergence-based, maximin distance-based expected improvement matrix (EIM-based), diversity-based and random-based strategies) for the multi-offspring-assisted multi-objective evolutionary algorithm are compared. The convergence-based strategy gives more priority to non-dominated solutions, and it is vital for exploiting the current promising areas. The diversity-based strategy gives more priority to solutions with greater uncertainties, and it is important for exploring the sparse areas. The EIM-based strategy considers the exploration and exploitation simultaneously, and the random-based strategy gives no priority to any solution. A series of benchmark problems whose dimensions vary from 8 to 30 and a reactive power optimization problem are used to test the multi-offspring-assisted multi-objective evolutionary algorithm under the four pre-screening strategies. The experimental results show that the convergence-based strategy performs best on most of the simple problems, while the EIM-based strategy performs best on most of the complex problems. The diversity-based strategy can produce positive effects on some problems, while the random-based strategy cannot improve the performance of its basic algorithm.
Football in Uruguay is the most popular sport, representing an important part of the country's culture. It is not surprising that the Uruguayan football championship serves as entertainment for a large part of the...
详细信息
ISBN:
(数字)9781665488587
ISBN:
(纸本)9781665488587
Football in Uruguay is the most popular sport, representing an important part of the country's culture. It is not surprising that the Uruguayan football championship serves as entertainment for a large part of the population. The process of designing a concrete fixture has many implications, for instance: is the tournament sportingly fair? is it attractive for the fans? is it economically profitable? In this work, we tackle the scheduling of fixtures for both Uruguayan football and basketball leagues to improve the fairness of the competitions. We address this problem following a multi-objective approach and use the wellknown NSGA-II, as well as an Integer Programming solver for considering each objective independently. The results obtained show the effectiveness of both algorithms studied to find highquality solutions. The resulting fixtures are far superior to the ones that are actually being used in the leagues.
暂无评论