Machine Learning explores the construction and study of learning algorithms. It can be viewed as a union of unsupervised learning methods, more commonly referred to as data mining, and supervised learning methods deeply rooted in mathematical theory, specifically statistics, combinatorics, and optimization. A third branch is now forming called reinforcement learning, where goal performance is earned but not specifically teacher recognized – driving a vehicle for example.
Programming machines to quickly learn from queries and adapt to changing data sets led to a completely new field within Big Data referred to as machine learning.89 Processes run, and results are stored that are then used in subsequent runs to iteratively inform the process and refine the results.
Machine Learning explores the construction and study of learning algorithms. These algorithms fall into three types:
- Supervised Learning: Based on generalized rules; for example, separating SPAM from non-SPAM email
- Unsupervised Learning: Based on identifying hidden patterns (i.e., data mining)
- Reinforcement Learning: Based on achieving a goal (e.g., beating an opponent at chess)
Statistical modeling and machine learning have been employed to automate otherwise costly research and development projects, by performing several trial and error passes on a vast set of data, repeating trials with the results collected, analyzed, and errors corrected. This approach can reduce time to answer dramatically and guide organizational initiatives with insights based on cost effective repeatable processes.
While Machine Learning taps into data in new ways, machine learning has ethical implications, especially with respect to the principle of transparency. Evidence shows that deep learning neural networks (DLNN) work. They learn things. However, it is not always clear how they learn. As the algorithms that drive these processes become more complex, they also become more opaque, functioning as ‘black boxes’. As they account for a greater number of variables and as those variables themselves are more abstract, algorithms test the limits of human ability to interpret the machine. The need for transparency – the ability to see how decisions are made – will likely increase as this functionality evolves and is put to use in a wider array of situations.
