In recent years, the increasing complexity of deep network structures has hindered their application in small resource constrained hardware. Therefore, we urgently need to compress and accelerate deep network models. Channel pruning is an effective method to compress deep neural networks. However, most existing channel pruning methods are prone to falling into local optima. In this paper, we propose a channel pruning method via Improved Grey Wolf Optimizer Pruner which called IGWO-Pruner to prune redundant channels of convolutional neural networks. It identifies pruning ratio of each layer by using Improved Grey Wolf algorithm, and then fine-tuning the new pruned network model. In experimental section, we evaluate the proposed method in CIFAR datasets and ILSVRC-2012 with several classical networks, including VGGNet, GoogLeNet and ResNet-18/34/56/152, and experimental results demonstrate the proposed method is able to prune a large number of redundant channels and parameters with rare performance loss.

The selection of mutation strategy greatly affects the performance of differential evolution algorithm (DE). For different types of optimization problems, different mutation strategies should be selected. How to choose a suitable mutation strategy for different problems is a challenging task. To deal with this challenge, this paper proposes a novel DE algorithm based on local fitness landscape, called FLIDE. In the proposed method, fitness landscape information is obtained to guide the selection of mutation operators. In this way, different problems can be solved with proper evolutionary mechanisms. Moreover, a population adjustment method is used to balance the search ability and population diversity. On one hand, the diversity of the population in the early stage is enhanced with a relative large population. One the other hand, the computational cost is reduced in the later stage with a relative small population. The evolutionary information is utilized as much as possible to guide the search direction. The proposed method is compared with five popular algorithms on 30 test functions with different characteristics. Experimental results show that the proposed FLIDE is more effective on problems with high dimensions.

In 2019, a new selection method, named fitness-distance balance (FDB), was proposed. FDB has been proved to have a significant effect on improving the search capability for evolutionary algorithms. But it still suffers from poor flexibility when encountering various optimization problems. To address this issue, we propose a functional weights-enhanced FDB (FW). These functional weights change the original weights in FDB from fixed values to randomly generated ones by a distribution function, thereby enabling the algorithm to select more suitable individuals during the search. As a case study, FW is incorporated into the spherical search algorithm. Experimental results based on various IEEE CEC2017 benchmark functions demonstrate the effectiveness of FW.

Rapid development of modern communications has initiated essential requirements for providing efficient algorithms that can solve the routing and wavelength assignment (RWA) problem in satellite optical networks. In this paper, the bee colony algorithm optimization based on link cost for RWA (BCO-LCRWA) is tailored for satellite networks composed of intersatellite laser links. In BCO-LCRWA, a cost model of intersatellite laser links is established based on metrics of network transmission performance namely delay and wavelengths utilization, with constraints of Doppler wavelength drift, transmission delay, wavelength consistency and continuity. Specifically, the fitness function of bee colony exploited in the proposed algorithm takes wavelength resources utilization and communication hops into account to implement effective utilization of wavelengths, to avoid unnecessary over-detouring and ensure bit error rate (BER) performance of the system. The simulation results corroborate the improved performance of the proposed algorithm compared with the existing alternatives.

The probability associated with population fitness in a Genetic Algorithm (GA) is studied using the concept of average Euclidean distance. Based on the probability derived from population fitness, the GA can effectively terminate its evolution operations to mitigate the total computational load. Simulation results verify the feasibility of the derived probability used for the GA's termination strategy.

We discuss a new design methodology of polynomial neural networks (PNN) in the framework of genetic algorithm (GA). The PNN is based on the ideas of group method of data handling (GMDH). Each node in the network is very flexible and can carry out polynomial type mapping between input and output variables. But the performances of PNN depend strongly on the number of input variables available to the model, the number of input variables, and the type (order) of the polynomials to each node. In this paper, GA is implemented to better use the optimal inputs and the order of polynomial in each node of PNN. The appropriate inputs and order are evolved accordingly and are tuned gradually throughout the GA iterations. We employ a binary coding for encoding key factors of the PNN into the chromosomes. The chromosomes are made of three sub-chromosomes which represent the order, number of inputs, and input candidates for modeling. To construct model by using significant approximation and generalization, we introduce the fitness function with a weighting factor. Comparisons with other modeling methods and conventional PNN show that the proposed design method offers encouraging advantages and better performance.

In the present paper, we propose a method for reconstructing the surfaces of objects from stereo data. Both the fitness of stereo data to surfaces and interrelation between the surfaces are defined in the framework of a three-dimensional (3-D) Markov Random Field (MRF) model. The surface reconstruction is accomplished by searching for the most likely state of the MRF model. Three experimental results are shown for synthetic and real stereo data.

An efficient clustering strategy for estimation of distribution algorithms (EDAs) is presented. It is used for properly fitting probabilistic models that play an important role in guiding search direction. To this end, a fitness-aided ordering scheme is devised for deciding the input sequence of samples (i.e., individuals) for clustering. It can effectively categorise the individuals by using the (available) information about fitness landscape. Moreover, a virtual leader is introduced for providing a reliable reference for measuring the distance from samples to its own cluster. The proposed algorithm incorporates them within the framework of random the leader algorithm (RLA). Experimental results demonstrate that the proposed approach is more effective than the existing ones with regard to probabilistic model fitting.

In this work we give an extension of Kauffman's NK-Landscapes to multiobjective MNK-Landscapes in order to study the effects of epistasis on the performance of multiobjective evolutionary algorithms (MOEAs). This paper focuses on the development of multiobjective random one-bit climbers (moRBCs). We incrementally build several moRBCs and analyze basic working principles of state of the art MOEAs on landscapes of increased epistatic complexity and number of objectives. We specially study the effects of Pareto dominance, non-dominance, and the use of memory and a population to influence the search. We choose an elitist non-dominated sorting multiobjective genetic algorithm (NSGA-II) as a representative of the latest generation of MOEAs and include its results for comparison. We detail the behavior of the climbers and show that population based moRBCs outperform NSGA-II for all values of M and K.

NK-Landscapes are stochastically generated fitness functions on bit strings, parameterized with N bits and K epistatic interactions between bits. The term epistasis describes nonlinearities in fitness functions due to changes in the values of interacting bits. Empirical studies have shown that the overall performance of random bit climbers on NK-Landscapes is superior to the performance of some simple and enhanced genetic algorithms (GAs). Analytical studies have also lead to suggest that NK-Landscapes may not be appropriate for testing the performance of GAs. In this work we study the effect of selection, drift, mutation, and recombination on NK-Landscapes for N = 96. We take a model of generational parallel varying mutation GA (GA-SRM) and switch on and off its major components to emphasize each of the four processes mentioned above. We observe that using an appropriate selection pressure and postponing drift make GAs quite robust on NK-Landscapes; different to previous studies, even simple GAs with these two features perform better than a random bit climber (RBC+) for a broad range of classes of problems (K 4). We also observe that the interaction of parallel varying mutation with crossover improves further the reliability of the GA, especially for 12 < K < 32. Contrary to intuition, we find that for small K a mutation only evolutionary algorithm (EA) is very effective and crossover may be omitted; but the relative importance of crossover interacting with varying mutation increases with K performing better than mutation alone (K > 12). This work indicates that NK-Landscapes are useful for testing each one of the major processes involved in a GA and for assessing the overall behavior of a GA on complex non-linear problems. This study also gives valuable guidance to practitioners applying GAs to real world problems of how to configure the GA to achieve better results as the non-linearity and complexity of the problem increases.

Feature subset selection is an important preprocessing task for pattern recognition, machine learning or data mining applications. A Genetic Algorithm (GA) with a fuzzy fitness function has been proposed here for finding out the optimal subset of features from a large set of features. Genetic algorithms are robust but time consuming, specially GA with neural classifiers takes a long time for reasonable solution. To reduce the time, a fuzzy measure for evaluation of the quality of a feature subset is used here as the fitness function instead of classifier error rate. The computationally light fuzzy fitness function lowers the computation time of the traditional GA based algorithm with classifier accuracy as the fitness function. Simulation over two data sets shows that the proposed algorithm is efficient for selection of near optimal solution in practical problems specially in case of large feature set problems.

In computer-aided drafting and design, interactive graphics is used to design components, systems, layouts, and structures. There are several approaches for using automated graphical layout tools currently. Our approach employs a genetic algorithm to implement a tool for automated 3D graphical layout design and presentation. The effective use of a genetic algorithm in automated graphical layout design relies on defining a fitness function that reflects user preferences. In this paper, we describe a method to define fitness functions and chromosome structures of selected objects. A learning mechanism is employed to adjust the fitness values of the objects in the selected layout chosen by the user. In our approach, the fitness functions can be changed adaptively reflecting user preferences. Experimental results revealed good performance of the adaptive fitness functions in our proposed mechanism.

A number of flexible control mechanisms used in buffer management, congestion control and bandwidth allocation have been proposed to improve the performance of ATM networks by introducing parameters, such as threshold, push-out probability and incremental bandwidth size of a virtual path, which are adjustable by network providers. However, it is difficult to adaptively adjust these parameters, since the traffic in ATM networks is further complicated by accommodating various kinds of services. To overcome the problem, we propose in this paper a control scheme based on the genetic algorithms and the neural estimator. The neural estimator forecasts the future QOS values for each candidate parameter set, and the genetic algorithms select the best one to control the real network. An example of buffer management in an ATM switch is examined in this paper. Simulation results show the effectiveness of the proposed control scheme in adaptively adjusting the parameter set even when the traffic environment and the QOS requirements are dynamically changing.