Machine Learning (I)

Machine Learning (ML) is all about algorithms that can tell you something about a set of data without you having to write any custom code specific to the problem; ie., instead of writing code, you feed data to the generic algorithm and it builds its own logic based on the data.

ML can be understood as a sub-field of Artificial Intelligence (AI), a branch of computer science that works to build machines that perform tasks that are characteristic of human intelligence (term "AI" was coined by John McCarthy in 1956). Applications of AI are capable of performing tasks previously done by a person but often faster and at a larger scale and conceptually contemplates planning, understanding language, recognizing objects and sounds, learning, and problem solving. AI can be roughly categorized in two 'groups', general (general AI would have all of the characteristics of human intelligence, including the capacities mentioned above), and narrow (Narrow AI exhibits some facets of human intelligence, and can do that facet extremely well, but lacking in other areas — ex: recognizing images, but nothing else beyond that).

ML is the main basis behind Predictive Modeling, an increasing popular practice of constructing computer-based models capable of making predictions. Typically, such models include an ML algorithm (or set of algorithms) that learn certain properties from a training set of data (dataset) in order to make those predictions. Arthur Samuel coined the phrase "Machine Learning" in 1959, defining it as, “the ability to learn without being explicitly programmed.”

Technically speaking, ML intersects statistical, probabilistic, linear algebra and algorithmic aspects arising from learning iteratively from data and finding hidden insights which can be used to build intelligent applications.

So, let's describe ML in simple math terms:

learning a target function (f) that best maps input variables (X) to an output variable (Y): Y = f(X).

This is a general learning task where we would like to make predictions in the future (Y) given new examples of input variables (X). We don’t know what the function (f) looks like or its form (in case we did, we would use it directly and we would not need to learn it from data using machine learning algorithms).

Most common ML techniques include Regression and (pattern) Classification.

Regression models are based on the analysis of relationships between variables and trends in order to make predictions about continuous variables, for example — the prediction of the maximum temperatures during next week days. Linear Regression is its most popular form.

Conversely, Classification goal is to assign discrete class labels to particular observations as outcomes of a prediction. Leveraging above example: A pattern classification task in weather forecasting could be the prediction of a sunny, rainy, or snowy day.

More generally, ML regression and classification algorithms can be developed under two broad approaches: Supervised and Unsupervised learning.

Let’s take the classification task under Supervised learning approach as a first illustration. The class labels in the dataset, which is used to build the classification model, are known (for instance — a dataset for email spam filtering would contain spam messages as well as “ham” [ie.,not-spam] messages). In a supervised learning problem, we would know which message in the training set is spam or ham, and we’d use this information to train our model in order to classify new unseen messages.

In contrast, Unsupervised learning task deal with unlabeled instances, and the classes have to be inferred from the unstructured dataset. When talking about classification, unsupervised learning employs a clustering technique in order to group the unlabeled samples based on certain similarity (or distance) measures.

Technically, there’s a third class of learning algorithms described by the term Reinforcement learning. In this case, the model is learned from a series of actions by maximizing a “reward function”. The reward function can either be maximized by penalizing “bad actions” and/or rewarding “good actions”. A well known example of reinforcement learning would be the training of self-driving car using feedback from the environment.

In summary: