Evolutionary Search Procedure Research Articles

Learning to Learn Evolutionary Algorithm: A Learnable Differential Evolution

Research on evolutionary optimization has flourished for several decades. Now it has come to a turning point. With the advancement of artificial intelligence, especially deep learning and reinforcement learning, it is becoming appealing to rethink the design and development of evolutionary algorithm (EA). From our perspective, a new-generation EA should be <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">learned</i> rather than manually designed, based on learning from optimization experiences (such as obtained from optimizing a family of optimization problems), the deep understanding of the roles of recombination operators, and the usage of experiences extracted through history optimization trajectories, so as to intelligently decide the control parameters that can adapt to the problem landscape changes. This learning can be conducted by strongly coupling with reinforcement learning since an evolutionary search procedure can be modeled as a Markov Decision Process (MDP). In this paper we propose a framework for automatic learning of EA, and present an exemplar study on learning a differential evolution (DE). Experimental results show that the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">learned</i> adaptive DE is very competitive to some recent EAs on a commonly-used test suite, which indicates that the proposed learning framework has a great potential for the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">automatic</i> design of promising EAs.

Scatter search for capacitated minimum spanning tree

We develop a scatter search algorithm for the classical capacitated minimum spanning tree (CMST) problem. We address both the homogeneous and the more difficult heterogeneous variant of the problem. The CMST is central in a wide range of network design applications in industrial engineering, routing and logistics, and communication networks. Since the problem is NP-Complete, heuristic solution methods are often required to find high-quality solutions within practical time limits. Research efforts have been focused on complex algorithm designs combining advanced neighborhood search techniques with exact solution methods or hybrids of genetic algorithms with various alternative search techniques and likewise supplemented with optimization procedures. In this study we go back to the roots and offer an alternative algorithm that competes with the best algorithms in the literature and both avoids complicated artifacts and is simpler to implement. Key features of the proposed algorithm include an oscillation neighborhood search method characterized by two classes of neighborhoods and an evolutionary search procedure based on a scatter search framework.

Learning Adaptive Differential Evolution by Natural Evolution Strategies

Adaptive parameter control is critical in the design and application of evolutionary algorithm (EA), so does in differential evolution. In the past decade, many adaptive evolutionary algorithms have been proposed, in which online information collected until current generation during the evolutionary search procedure is used to determine the algorithmic parameters for the next generation. Recent studies often assume that the algorithmic parameters follow some distributions, while the distributions' parameters (called hyper-parameters) are updated by the collected information. Performances of these adaptive EAs depend highly on the hyper-parameters. Notice that the experiences obtained from optimizing some related problems could provide useful guidelines on how to adaptively control the distributions' parameters. However, few existing studies sufficiently used such experiences. To fill the gap, we propose a general framework for adaptive parameter control by modeling its evolution procedure as a Markov decision process. In the framework, a neural network is employed to act as the controller. The natural evolution strategies is applied to train the neural network. The proposed framework is applied on two well-known differential evolutions (DEs), namely JADE and LSHADE. By incorporating the learned controller, two DEs, named JADE/AC and LSHADE/AC, are formed. Experimental results on the CEC 2018 benchmark suite show that in general JADE/AC and LSHADE/AC perform significantly better than their counterparts. Moreover, in comparison with some well-known EAs including three suggested best DEs in a review paper (including LSHADE, cDE and CoBiDE), the championship algorithm in the CEC 2018 competitions, a recently-developed learnable DE and recently proposed DEs, our study shows that LSHADE/AC performs the best amongst them without sacrificing much computation time.

Neural Architecture Transfer.

Neural architecture search (NAS) has emerged as a promising avenue for automatically designing task-specific neural networks. Existing NAS approaches require one complete search for each deployment specification of hardware or objective. This is a computationally impractical endeavor given the potentially large number of application scenarios. In this paper, we propose Neural Architecture Transfer (NAT) to overcome this limitation. NAT is designed to efficiently generate task-specific custom models that are competitive under multiple conflicting objectives. To realize this goal we learn task-specific supernets from which specialized subnets can be sampled without any additional training. The key to our approach is an integrated online transfer learning and many-objective evolutionary search procedure. A pre-trained supernet is iteratively adapted while simultaneously searching for task-specific subnets. We demonstrate the efficacy of NAT on 11 benchmark image classification tasks ranging from large-scale multi-class to small-scale fine-grained datasets. In all cases, including ImageNet, NATNets improve upon the state-of-the-art under mobile settings ( ≤ 600M Multiply-Adds). Surprisingly, small-scale fine-grained datasets benefit the most from NAT. At the same time, the architecture search and transfer is orders of magnitude more efficient than existing NAS methods. Overall, experimental evaluation indicates that, across diverse image classification tasks and computational objectives, NAT is an appreciably more effective alternative to conventional transfer learning of fine-tuning weights of an existing network architecture learned on standard datasets. Code is available at https://github.com/human-analysis/neural-architecture-transfer.

Greedy-Based Non-Dominated Sorting Genetic Algorithm III for Optimizing Single-Machine Scheduling Problem With Interfering Jobs

Given the importance of production planning and control in the design of flexible services and manufacturing systems, scheduling problems with interfering jobs are much-needed optimization tools to respond to heterogeneous and fluctuating market demands in a timely fashion. This study contributes to the scheduling literature developing an effective multi-objective (M-O) metaheuristic to solve the Single-machine Scheduling Problems with Interfering Jobs (SSP-IJs). Integrating a local search-based mechanism into the evolutionary search procedure, a Greedy-based non-dominated sorting genetic algorithm III (GNSGA-III) is proposed that effectively explores multi-objective solution environments. Various performance indicators within extensive numerical tests are used to compare the performance of the GNSGA-III with that of the best-performing benchmark algorithm in the literature developed to solve the SSP-IJs. Statistical tests verify that the developed multi-objective optimization algorithm is superior with respect to various performance indicators. Applications of the developed solution approach are worthwhile topics to help advance multi-objective optimization problems.

Optimizing optogenetic stimulation protocols in auditory corticofugal neurons based on closed-loop spike feedback

Objective. Optogenetics provides a means to probe functional connections between brain areas. By activating a set of presynaptic neurons and recording the activity from a downstream brain area, one can establish the sign and strength of a feedforward connection. One challenge is that there are virtually limitless patterns that can be used to stimulate a presynaptic brain area. Functional influences on downstream brain areas can depend not just on whether presynaptic neurons were activated, but how they were activated. Corticofugal axons from the auditory cortex (ACtx) heavily innervate the auditory tectum, the inferior colliculus (IC). Here, we sought to determine whether different modes of corticocollicular activation could titrate the strength of feedforward modulation of sound processing in IC neurons. Approach. We used multi-channel electrophysiology and optogenetics to record from multiple regions of the IC in awake head-fixed mice while optogenetically stimulating ACtx neurons expressing Chronos, an ultra-fast channelrhodopsin. To identify cortical activation patterns associated with the strongest effects on IC firing rates, we employed a closed-loop evolutionary optimization procedure that tailored the voltage command signal sent to the laser based on spike feedback from single IC neurons. Main results. Within minutes, our evolutionary search procedure converged on ACtx stimulation configurations that produced more effective and widespread enhancement of IC unit activity than generic activation parameters. Cortical modulation of midbrain spiking was bi-directional, as the evolutionary search procedure could be programmed to converge on activation patterns that either suppressed or enhanced sound-evoked IC firing rate. Significance. This study introduces a closed-loop optimization procedure to probe functional connections between brain areas. Our findings demonstrate that the influence of descending feedback projections on subcortical sensory processing can vary both in sign and degree depending on how cortical neurons are activated in time.

Learning From a Stream of Nonstationary and Dependent Data in Multiobjective Evolutionary Optimization

Combining machine learning techniques has shown great potentials in evolutionary optimization since the domain knowledge of an optimization problem, if well learned, can be a great help for creating high-quality solutions. However, existing learning-based multiobjective evolutionary algorithms (MOEAs) spend too much computational overhead on learning. To address this problem, we propose a learning-based MOEA where an online learning algorithm is embedded within the evolutionary search procedure. The online learning algorithm takes the stream of sequentially generated solutions along the evolution as its training data. It is noted that the stream of solutions are temporal, dependent, nonstationary, and nonstatic. These data characteristics make existing online learning algorithm not suitable for the evolution data. We hence modify an existing online agglomerative clustering algorithm to accommodate these characteristics. The modified online clustering algorithm is applied to adaptively discover the structure of the Pareto optimal set; and the learned structure is used to guide new solution creation. Experimental results have shown significant improvement over four state-of-the-art MOEAs on a variety of benchmark problems.

Balancing exploration and exploitation in multiobjective evolutionary optimization

The tradeoff between exploration and exploitation is critical to the performance of an evolutionary algorithm. Different levels of exploration-exploitation tradeoff are required at different evolutionary stages for achieving a satisfactory performance of an evolutionary algorithm. In this paper, we propose a novel survival analysis method to intelligently guide the maintenance of the exploration-exploitation tradeoff in multiobjective evolutionary algorithms. The survival analysis stems from a deep understanding of the evolutionary search procedure. Through survival analysis, an indicator is derived, which is used to guide the adoption of appropriate recombination operators, based on the assumption that the roles of these operators in terms of their capabilities on exploration-exploitation can be asserted. In the developed algorithm, a differential evolution recombination operator and a new sampling strategy are hybridized. Empirical comparison with five well-known multiobjective evolutionary algorithms on a number of test instances with complex Pareto sets and Pareto fronts indicates the effectiveness and superiority of the developed algorithm in terms of commonly-used performance metrics on these test instances.

Evolutionary optimization of PAW data-sets for accurate high pressure simulations

We examine the challenge of performing accurate electronic structure calculations at high pressures by comparing the results of all-electron full potential linearized augmented-plane-wave calculations, as implemented in the WIEN2k code, with those of the projector augmented wave (PAW) method, as implemented in Quantum ESPRESSO or Abinit code. In particular, we focus on developing an automated and consistent way of generating transferable PAW data-sets that can closely produce the all electron equation of state defined from zero to arbitrary high pressures. The technique we propose is an evolutionary search procedure that exploits the ATOMPAW code to generate atomic data-sets and the Quantum ESPRESSO software suite for total energy calculations. We demonstrate different aspects of its workability by optimizing PAW basis functions of some elements relatively abundant in planetary interiors. In addition, we introduce a new measure of atomic data-set goodness by considering their performance uniformity over an extended pressure range.

Prediction of Concrete Compressive Strength by Evolutionary Artificial Neural Networks

Compressive strength of concrete has been predicted using evolutionary artificial neural networks (EANNs) as a combination of artificial neural network (ANN) and evolutionary search procedures, such as genetic algorithms (GA). In this paper for purpose of constructing models samples of cylindrical concrete parts with different characteristics have been used with 173 experimental data patterns. Water-cement ratio, maximum sand size, amount of gravel, cement, 3/4 sand, 3/8 sand, and coefficient of soft sand parameters were considered as inputs; and using the ANN models, the compressive strength of concrete is calculated. Moreover, using GA, the number of layers and nodes and weights are optimized in ANN models. In order to evaluate the accuracy of the model, the optimized ANN model is compared with the multiple linear regression (MLR) model. The results of simulation verify that the recommended ANN model enjoys more flexibility, capability, and accuracy in predicting the compressive strength of concrete.

Preference-based reinforcement learning: evolutionary direct policy search using a preference-based racing algorithm

We introduce a novel approach to preference-based reinforcement learning, namely a preference-based variant of a direct policy search method based on evolutionary optimization. The core of our approach is a preference-based racing algorithm that selects the best among a given set of candidate policies with high probability. To this end, the algorithm operates on a suitable ordinal preference structure and only uses pairwise comparisons between sample rollouts of the policies. Embedding the racing algorithm in a rank-based evolutionary search procedure, we show that approximations of the so-called Smith set of optimal policies can be produced with certain theoretical guarantees. Apart from a formal performance and complexity analysis, we present first experimental studies showing that our approach performs well in practice.

An evolutionary search procedure for optimizing time–cost performance of projects under multiple renewable resource constraints

This paper presents an effective procedure for optimizing time–cost performance of multi-mode resource constrained project scheduling problems in which activities are subject to finish-start precedence constraints under renewable limited resources. Associated with each execution mode of activities, there exists a direct cost, a processing time, and a set of required renewable resources. In optimizing time–cost performance, the procedure treats the cost as a non-renewable resource whose limit can affect the duration of the project and balances cost versus time through the notion of priority-rank. This is performed by the use of a module which handles multimode projects, and since the procedure has to call this module with different limits on the cost, the effectiveness of this module plays a key role in the overall efficiency. For this reason, an effective evolutionary search technique has been developed to create the basis of this module. For testing its effectiveness, this module has been tested on 552 largest multimode benchmark instances of the PSPLIB and the results are promising: For over 98% of instances, the module finds the best available solutions in the literature. The module also produces a solution for one of these benchmark instances that is better than all of the current solutions in the literature.

Red Chlorophylls in the Exciton Model of Photosystem I

Structural arrangement of pigment molecules of Photosystem I of photosynthetic cyanobacterium Synechococcus elongatus is used for theoretical modeling of the excitation energy spectrum. It is demonstrated that a straightforward application of the exciton theory with the assumption of the same molecular transition energy does not describe the red side of the absorption spectrum. Since the inhomogeneity in the molecular transition energies caused by a dispersive interaction with the molecular surrounding cannot be identified directly from the structural model, the evolutionary search procedure is used for fitting the low temperature absorption and circular dichroism spectra. As a result, one dimer, three trimers and one tetramer of chlorophyll molecules responsible for the red side of the absorption spectrum with their assignment to the spectroscopically established three bands at 708, 714 and 719 nm are determined. All of them are found to be situated not in the very close vicinity of the reaction center but are encircling it almost at the same distance. In order to explain the unusual broadening on the red side of the spectrum the exciton state mixing with the charge transfer (CT) states is considered. It is shown that two effects can be distinguished as caused by mixing of those states: (i) the oscillator strength borrowing by the CT state from the exciton transition and (ii) the borrowing of the high density of the CT state by the exciton state. The intermolecular vibrations between two counter-charged molecules determine the high density in the CT state. From the broad red absorption wing it is concluded that the CT state should be the lowest state in the complexes under consideration. Such mixing effect enables resolving the diversity in the molecular transition energies as determined by different theoretical approaches.

Ant colony optimization for the examination scheduling problem

Ant colony optimization is an evolutionary search procedure based on the way that ant colonies cooperate in locating shortest routes to food sources. Early implementations focussed on the travelling salesman and other routing problems but it is now being applied to an increasingly diverse range of combinatorial optimization problems. This paper is concerned with its application to the examination scheduling problem. It builds on an existing implementation for the graph colouring problem to produce clash-free timetables and goes on to consider the introduction of a number of additional practical constraints and objectives. A number of enhancements and modifications to the original algorithm are introduced and evaluated. Results based on real-examination scheduling problems including standard benchmark data (the Carter data set) show that the final implementation is able to compete effectively with the best-known solution approaches to the problem.

Solving nonlinear water management models using a combined genetic algorithm and linear programming approach

Gradient-based nonlinear programming (NLP) methods can solve problems with smooth nonlinear objectives and constraints. However, in large and highly nonlinear models, these algorithms can fail to find feasible solutions, or converge to local solutions which are not global. Evolutionary search procedures in general, and genetic algorithms (GAs) specifically, are less susceptible to the presence of local solutions. However, they often exhibit slow convergence, especially when there are many variables, and have problems finding feasible solutions in constrained problems with “narrow” feasible regions. In this paper, we describe strategies for solving large nonlinear water resources models management, which combine GAs with linear programming. The key idea is to identify a set of complicating variables in the model which, when fixed, render the problem linear in the remaining variables. The complicating variables are then varied by a GA. This GA&LP approach is applied to two nonlinear models: a reservoir operation model with nonlinear hydropower generation equations and nonlinear reservoir topologic equations, and a long-term dynamic river basin planning model with a large number of nonlinear relationships. For smaller instances of the reservoir model, the CONOPT2 nonlinear solver is more accurate and faster, but for larger instances, the GA&LP approach finds solutions with significantly better objective values. The multiperiod river basin model is much too large to be solved in its entirety. The complicating variables are chosen here so that, when they are fixed, each period's model is linear, and these models can be solved sequentially. This approach allows sufficient model detail to be retained so that long-term sustainability issues can be explored.

Towards the support of innovative conceptual design through interactive designer/evolutionary computing strategies

The paper discusses the requirements of the engineering designer during the higher, conceptual levels of the design process in terms of search within initial predefined design spaces and subsequent exploration of transformed design spaces. Such transformation is based upon information gathered from initial search using evolutionary techniques that rapidly identify high-performance regions of complex design spaces. It is proposed that off-line processing of such information and associated design team discussion in an iterative designer/evolutionary search procedure can strengthen the knowledge base relating to the problem at hand and initiate significant change in the design environment. Research relating to appropriate evolutionary search and exploratory procedures is described and illustrated with results from simple test functions and real-world design problems. These results support discussion related to the manner in which the techniques may best support innovative and creative design activity in an interactive design team/evolutionary search environment.

Ants can colour graphs

In the last few years researchers have shown how insect colonies can be seen as a natural model of collective problem solving. The analogy between the behaviour of ants looking for food and the well known travelling salesman problem has recently given rise to promising solution methods. We present in this paper an evolutionary search procedure for tackling assignment type problems. The algorithm repeatedly constructs feasible solutions of the problem under study by taking account of two complementary notions, namely the trace factor and the desirability factor. The use of such search principles will be illustrated for graph colouring problems. The results obtained have proven satisfactory when compared with those existing in the literature.

Pattern Categorization and Generalization with a Virtual Neuromolecular Architecture

A multilevel neuromolecular computing architecture has been developed that provides a rich platform for evolutionary learning. The architecture comprises a network of neuron-like modules with internal dynamics modeled by cellular automata. The dynamics are motivated by the hypothesis that molecular processes operative in real neurons (in particular processes connected with second messenger signals and cytoskeleton-membrane interactions) subserve a signal integrating function. The objective is to create a repertoire of special purpose dynamic pattern processors through an evolutionary search algorithm and then to use memory manipulation algorithms to select combinations of processors from the repertoire that are capable of performing coherent pattern recognition/neurocontrol tasks. The system consists of two layers of cytoskeletally controlled (enzymatic) neurons and two layers of memory access neurons (called reference neurons) divided into a collection of functionally comparable subnets. Evolutionary learning can occur at the intraneuronal level through variations in the cytoskeletal structures responsible for the integration of signals in space and time, through variations in the location of elements that represent readin or readout proteins, and through variations in the connectivity of the neurons. The memory manipulation algorithms that orchestrate the repertoire of neuronal processors also use evolutionary search procedures. The network is capable of performing complicated pattern categorization tasks and of doing so in a manner that balances specificity and generalization. Copyright © 1996 Elsevier Science Ltd.

Evolutionary artificial neural networks.

Evolutionary artificial neural networks (EANNs) can be considered as a combination of artificial neural networks (ANNs) and evolutionary search procedures such as genetic algorithms (GAs). This paper distinguishes among three levels of evolution in EANNs, i.e. the evolution of connection weights, architectures and learning rules. It first reviews each kind of evolution in detail and then analyses major issues related to each kind of evolution. It is shown in the paper that although there is a lot of work on the evolution of connection weights and architectures, research on the evolution of learning rules is still in its early stages. Interactions among different levels of evolution are far from being understood. It is argued in the paper that the evolution of learning rules and its interactions with other levels of evolution play a vital role in EANNs.

A review of evolutionary artificial neural networks

Research on potential interactions between connectionist learning systems, i.e., artificial neural networks (ANNs), and evolutionary search procedures, like genetic algorithms (GAs), has attracted a lot of attention recently. Evolutionary ANNs (EANNs) can be considered as the combination of ANNs and evolutionary search procedures. This article first distinguishes among three kinds of evolution in EANNs, i.e., the evolution of connection weights, of architectures, and of learning rules. Then it reviews each kind of evolution in detail and analyzes critical issues related to different evolutions. the review shows that although a lot of work has been done on the evolution of connection weights and architectures, few attempts have been made to understand the evolution of learning rules. Interactions among different evolutions are seldom mentioned in current research. However, the evolution of learning rules and its interactions with other kinds of evolution, play a vital role in EANNs. Finally, this article briefly describes a general framework for EANNs, which not only includes the aforementioned three kinds of evolution, but also considers interactions among them. © 1993 John Wiley & Sons, Inc.