China 
Africa 
Explore chapters and articles related to this topic
Machine Learning
Published in Seyedeh Leili Mirtaheri, Reza Shahbazian, Machine Learning Theory to Applications, 2022
Seyedeh Leili Mirtaheri, Reza Shahbazian
The goal of mathematical optimization (alternatively spelled optimization) or mathematical programming is to choose the best element, based on some standards, from some set of available alternatives. According to optimization problems in all quantitative disciplines such as computer science, engineering, operations research, and economics, finding a mathematical solution has been considered for centuries. First, we must quantify the system’s performance using a single number called objective, which could be profit, time, potential energy, or any quantity or combination of quantities shown by a single number [25]. The primary system defines the objective attributes called variables or unknowns. These variables may be restricted or constrained because of the nature of the task. For example, the electron density in a molecule and the interest rate on a loan are always positive. Our goal in optimization is to specify the values of the variables that optimize the objective. Modeling that refers to the whole process of identifying objectives, variables, and constraints for a specific task is the first step of optimization. An appropriate model has the needed complexity for the problem. Simplicity in the model construction may lead to a lack of insights into the practical problem, while the model complexity may be made too difficult to solve. After model construction, an optimization algorithm can be used to find its solution. A computer is usually needed to implement this process due to the complexity of the algorithm and model. Instead of a universal optimization algorithm, there are several algorithms, each of which is suitable for a particular type of optimization problem. Algorithm selection depends on the task that the user tends to perform. This choice is an important one because the problem-solving speed and whether the solution is found depends on this selection.
Hybrid Deep Learning-based Models for Crop Yield Prediction
Published in Applied Artificial Intelligence, 2022
Alexandros Oikonomidis, Cagatay Catal, Ayalew Kassahun
Despite their importance and wide usage, there are several challenges in applying ML and DL models for crop yield prediction. The training of these models takes a lot of time, especially when the models contain many layers. Also, the performance of the models may be different depending on a number of circumstances. Furthermore, the most complex models may not always result in the best performance, making algorithm selection difficult. For instance, the XGBoost algorithm is preferred by many researchers due to its speed, efficiency, and requirement for less data manipulation. Convolutional Neural Networks (CNN) algorithm is well-known for its ability to learn features from data. Deep Neural Networks (DNN) algorithm holds the advantage of dealing with non-linear data successfully.
Comparative Study of AutoML Approach, Conventional Ensemble Learning Method, and KNearest Oracle-AutoML Model for Predicting Student Dropouts in Sub-Saharan African Countries
Published in Applied Artificial Intelligence, 2022
Yuda N Mnyawami, Hellen H Maziku, Joseph C Mushi
The studies by Mduma, Kalegele, and Machuve (2019), Lee and Chung (2019), Chareonrat (2016), Aguiar (2015), and Sara et al. (2015) have focused on establishing machine learning (ML) prediction models as measures to fight against student dropout in secondary schools, but the dropout problem still persists. The persisting dropout problem, especially in secondary schools, is attributed to a lack of proper identification of root causes and the unavailability of formal methods that can be used to project the severity of the problem. The difficulty stems from the fact that traditional machine learning algorithms suffer from feature processing and algorithm selection (Feurer et al. 2015). Moreover, no single machine learning algorithm/classifier or ensemble of classifiers can perfectly adapt to every test set (Azizi and Farah 2012). This is a significant drawback of most existing machine learning algorithms and ensemble learning techniques, as it compromises proper feature processing as well as the prediction accuracy of the machine learning models. However, automated machine learning (AutoML) methods provide better prediction results in different classification tasks. On the other hand, AutoML faces the challenge of selecting one of the optimal prediction models generated by the optimization methods for the various subsets of samples in the dataset (Tsiakmaki et al. 2020; Waring, Lindvall, and Umeton 2020). The challenge of selecting optimal prediction models among the pool of predictive methods is addressed by static and dynamic ensemble selection schemes (Ko, Sabourin, and Britto 2008). The static ensemble selection technique selects the optimal performing classifiers subsets for the whole test set (Azizi and Farah 2012). The method determines an ensemble of classifiers (EoC) for all test samples, not every selected classifier in the AutoML pool is an expert in classifying all known training samples (Ko, Sabourin, and Britto 2008). Dynamic ensemble techniques work by estimating the competence level of each classifier from a pool of classifiers (Vriesmann et al. 2012). The estimated competence of the ensemble of classifiers is based on a local region of the feature space where the query sample is located (Zhu, Wu, and Yang 2004). The local region can defined by different methods such as overall local accuracy, local class accuracy, A priori, A posteriori, and K-Nearest Oracle (KNORA) (Ko, Sabourin, and Britto 2008).
Metamodel-based dynamic algorithm configuration using artificial neural networks
Published in International Journal of General Systems, 2023
The algorithm selection problem was first formulated by Rice (1976) as the task of finding for a given problem instance the most suitable algorithm in terms of effectiveness and further related performance measures, e.g. computational time. However, this approach focuses on a per-instance selection of an algorithm. Over the decades, the theory was broadened in various directions as illustrated by the survey of Kerschke et al. (2019). To consider algorithm performance over a set of instances, algorithm configuration was devised as a methodology for providing the most favorable settings of an algorithm in terms of its parameters. As an extension, algorithm portfolios can be used both for selection and configuration to augment the set of candidate methods applicable to a specific instance or a set of instances, respectively. Depending on the domains of the search space, algorithm selection is further categorized into methods for continuous problems (Muñoz et al. 2015) and for combinatorial problems (Kotthoff 2016). In the former class, methods include general approaches such as line search, trust region procedures, random search. In the latter class, methods are tailored towards specific assumptions, requirements, and parameters of a combinatorial problem such as the traveling salesman problem (TSP), the propositional satisfiability problem (SAT), or mixed-integer programming (MIP) in general. Moreover, it is shown that the influence of parameters can be characterized through so-called algorithm configuration landscapes addressing the algorithm behavior upon variation of a single parameter (Pushak and Hoos 2018). When algorithm execution extends over a period of time, different parameter settings may be favorable during different stages of the runtime rather than having a fixed configuration over the entire runtime. Hamadi, Monfroy, and Saubion (2012) point towards the importance of online configuration of algorithms and provide references to dynamic algorithm configuration in metaheuristic algorithms. Available methods for switching between algorithms and/or parameter configurations during runtime in order to improve algorithm performance comprise reactive search, on-line algorithm control, adaptive operator selection techniques. Moreover, the value of static configuration schemes is seen as a baseline in order to facilitate online control and non-stationary algorithm parameter configurations in the first place. Details on different methods for implementing automated online tuning and dynamic adaptation of algorithm parameters are presented by Battiti, Brunato, and Mascia (2008) for the classes of model-based search, supervised learning (including ANNs), and reinforcement learning. Also Biedenkapp et al. (2020) emphasize the potential of a dynamic version of automated algorithm configuration and develop a reinforcement learning approach as a data-driven methodology which learns configuration policies over a set of given instances providing a white-box benchmark for the resulting dynamic configuration mechanism.