Today Machine Learning is everywhere. From video recommendations, spam filters, voice recognition to self-driving cars, Machine Learning is the mastermind behind those features. Machine Learning algorithms utilize a large volume of data to find patterns inside the data and apply those patterns for making predictions and decisions. According to Machine Learning by Tom M. Mitchell (1997):
A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks T, as measured by P, improves with experience E.
To put in another way, machine learning requires experience E in the forms of data by doing tasks T (analyzing and modelling the data, approximating mapping function using Machine Learning algorithms) with result of P (error margins, results comparison, etc.) to ‘learn’ or finding patterns in data.
Generally, Machine Learning methods fall into three main classes: Supervised Learning, Unsupervised Learning, and Semi-Supervised Learning.
Supervised Learning is a method to determine a mapping function of the data. This method uses a ‘labeled’ dataset, meaning both input variables and output variables are given.
Hence, the machine iteratively learns to approximate output variables by adjusting parameters of the algorithms such that it could predict the output if a new input data is given. Classification and Regression algorithms are two common examples of supervised learning algorithms.
Unsupervised Learning is a method to analyze the data to determine its structure and distribution with the goal of gaining insight into the data. The dataset for unsupervised learning is ‘unlabeled’, meaning only input variables are available.
Unsupervised Learning solves Clustering and Associations problems where Clustering, a problem with the goal of determining groupings within the data, and Association, a problem with the purpose of finding rules within the data.
Semi-Supervised Learning is the combination of Supervised and Unsupervised Learning. The given dataset contains both labeled and unlabeled, hence both techniques from Supervised and Unsupervised Learning, depending on the goal of learning, are necessary to discover its pattern.
Choosing between these methods depends on what dataset is available, the goal of applying the machine learning itself, which appropriate algorithms for learning the given dataset, and the performance measures for the algorithms.
Comentarios