Lecture 1: What is Machine Learning (ML)

Supervised Learning

Learning form a " labeled " data.

What is the goal of Artificial Intelligence ?

Artificial intelligence (AI) emerged with the goal of making machines do what intelligent beings do. You can say that artificial intelligence is about machines that make conclusions.

What is the difference between Classification & Clustering ?

In Clustering, clusters and their numbers are unknown. In Classification, Classes and their numbers are previously known.

Machine Learning Problems

Machine learning is mainly focused on finding and formalizing different regularities based on the given examples.

Machine Learning (Arthur Samuel)

The field of study that gives computers the ability to learnwithout being straightforward programmed.

How a machine works to predict a result based on input data ?

The machine trained on diverse data and big datasets observes more precedents. Therefore, it easier finds patterns and provides more accurate results.

When can we call a Machine Intelligent ?

When a machine performs non trivial tasks, it demonstrates the intelligence.

Why we should extract important features ?

When there are many predictors, the model is learning slower, and its performance decreases. Therefore, the model is less efficient.

A test data set

a test data set is a data set used to evaluate the model efficiency.

A training data set

a training data set is a data set used to train the model.

Difference between Bagging and Boosting

boosting draws subsets of input data, but sampling isn't entirely random.

Boosting

involves incremental training of algorithms in such a way that each new algorithm emphasizes errors of the previous algorithms.

A Classification problem

is a problem of assigning an object to one of the predefined classes. ( used in supervised learning )

Stacking

is based on the following principle. Several algorithms are trained, combined, and fed into the last algorithm that makes a final decision.

Bagging

trains each algorithm in the ensemble using a randomly drawn subset of the input data. Then, the obtained outputs are averaged.

Reinforcement Learning (RL)

1. The machine is given a positive reward for 'good' actions, a neutral reward if nothing changes, and it receives a negative reward for 'bad' actions. 2. No prior coding for learning. 3. Told to explore by itself on Trial and Error basis.

Association Rule Learning

A Problem of discovering the relationships between interconnected items (objects or events).

Machine Learning (Tom Mitchell)

A computer program is said to learn from experience with respect to some class of tasks and performance measure.

Dimensionality Reduction

A problem of finding the transformation data, which allows reducing the number of used variables (Possibly new).

Clustering

A problem of grouping a set of objects into previously unknown disjoint sets(clusters) is called a clustering problem

Ensemble Learning

Different methods are combined to complement each other. When one method fails or makes many mistakes, another method eliminates them.

Unsupervised Learning

In unsupervised learning, we give gigabytes of input data to a machine and ask it something.

Predictors (independent variables)

Input Variables that describe objects of interest designated by X1,X2,...,Xp

Unsupervised Learning

Learning from " unlabeled " data.

Response (dependent variables)

Output variables based on some predictor values, it is designated by Y.

Regression

Regression is the problem of predicting a value. offers easy-to-understand graphical models.

Ensemble Methods

Stacking, Bagging & Boosting.

Branches of machine learning

Supervised Learning, Unsupervised Learning & Reinforcement Learning.