Welcome to the first article of Introduction to Machine Learning Series. You can continue reading about the Perceptron Algorithm after this one.
What are Cognitive Algorithms?
An algorithm is:
a process or set of rules to be followed in calculations or other problem-solving operations, especially by a computer.
We can perform calculations, automate tasks or process data with algorithms.
What is Cognition?
derived from cognoscere (latin) : to recognize
In simple terms; cognition is using the existing knowledge to generate new knowledge.
Basics of ML
Each problem, both in math and real life, has its own set of inputs.
In machine learning, it is crucial to choose features wisely, while solving a real life problem. For example, while forecasting weather, we would need the following inputs to achieve a accurate solution:
- Last 10 days
- Weather data for last 10 years in that day
- Air Pressure
There is no computer powerful enough or no algorithm so genius, that can solve a problem with insufficient data. But we should also keep in mind, that unnecessary data would:
- increase data volume
- increase noise of data, therefore worsen the performance
- increase the time needed to train our model
This process is called Feature extraction and corresponding set is the Feature Space of our problem. There are numerous algorithms and techniques on extracting feature, modifying feature space size, dimension, stability etc. , and we will talk about those in the next articles.
Each data input is a vector in our feature space X
Supervised vs Unsupervised Problems
In an unsupervised problem,
we are given a bunch of (usually at least >10k) data points in the feature space and the algorithm should find an underlying pattern in data
Whereas a supervised problem uses the given labels in the data set, to predict a label for a new data point.
Classification and Regression
Now that we not what supervised learning algorithms do, there is a one last point left to make.
A Classification problem has a limited label domain. The Label Set Y can have only 2 distinct elements for a binary classification problem, and corresponding number of elements for a multiple classification problem.
Whereas the Domain of Y can be Real numbers, when working with Regression.
So, a regression algorithm predicts the exact label value in real numbers given a new point, while classification algorithms, as the name suggests, assigns the point with the correct class.