Update paper.md

paper/paper.md

@@ -22,9 +22,9 @@ bibliography: paper.bib
 
 # Summary
 
-The field of Automated Machine Learning (AutoML) has been developed to automate data preprocessing and search for optimal algorithms together with their hyperparameters in order to discover the best possible ML pipeline for an input dataset [@bookHutter]. AutoML can be modeled as a continuous optimization problem with several potential optimization methods considered. Stochastic population-based nature-inspired algorithms [@yang2014; @engelbrecht2007computational] are a popular class of tools for dealing with such continuous optimization problems. These algorithms are inspired mainly by the biological principles and phenomena of the behavior of various species living in nature [@fister2013brief]. Such algorithms are composed of a population of individuals that undergo different variation operations during the evolution process which results in new populations. The Python framework we have developed, NiaAML, incorporates these stochastic algorithms to search for the most suitable classification pipeline in a dataset [@Fister2020Continuous].
+The field of Automated Machine Learning (AutoML) has been developed to automate data preprocessing and search for optimal algorithms together with their hyperparameters in order to discover the best possible ML pipeline for an input dataset [@bookHutter]. AutoML can be modeled as a continuous optimization problem with several potential optimization methods considered. Stochastic population-based nature-inspired algorithms [@yang2014; @engelbrecht2007computational] are a popular class of tools for dealing with such continuous optimization problems. These algorithms are inspired mainly by the biological behavior of various species living in nature [@fister2013brief]. Such algorithms are composed of a population of individuals that undergo different variation operations during the evolution process which results in new populations. The Python framework we have developed, NiaAML, incorporates these stochastic algorithms to search for the most suitable classification pipeline in a dataset [@Fister2020Continuous].
 
-The framework is developed in a layer style layout architecture, consisting of several components, i.e. Feature Selection algorithms, Feature Transformation algorithms and classifiers. Its task is to find a perfect combination of components with proper classifier hyperparameter settings to build an efficient, yet customizable classification pipeline with the help of a popular collection of nature-inspired algorithms, named NiaPy [@Vrbančič2018]. Two types of optimizations in the NiaAML follow: (1) The first to find the optimal set of components for the pipeline, and (2) The second to tune the hyperparameters. Users can choose the ML components to be included into the optimization process freely, as well as select suitable fitness functions to be used for evaluation of candidate pipelines. Input data can be in the form of numerical and categorical features, as well as missing attributes, while pipelines are exported and imported as binary files for post-hoc use. Further, they can be exported as user-friendly text files that contain all of the relevant information about the pipeline and its components. A graphical outline of the NiaAML method is presented in \autoref{fig:NiaAMLflow}.
+The framework is developed in a layer style layout architecture consisting of several components, including feature selection algorithms, feature transformation algorithms and classifiers. Its task is to find a perfect combination of components with proper classifier hyperparameter settings to build an efficient, yet customizable classification pipeline with the help of a popular collection of nature-inspired algorithms, named NiaPy [@Vrbančič2018]. NiaAML incorporates two types of optimizations, the first involves finding the optimal set of components for the pipeline, and the second involves tuning the hyperparameters. Users can freely choose the ML components to be included into the optimization process, as well as select suitable fitness functions to be used for evaluation of candidate pipelines. Input data can be in the form of numerical and categorical features, as well as missing attributes, while pipelines are exported and imported as binary files for post-hoc use. Further, they can be exported as user-friendly text files that contain all of the relevant information about the pipeline and its components. A graphical outline of the NiaAML method is presented in \autoref{fig:NiaAMLflow}.
 
 ![NiaAML flow.\label{fig:NiaAMLflow}](niaamlFlow.png)