What is machine learning?
Most of us are very much aware of the demand for ML across industries. This has brought a unique change in the recruitment of candidates. An ML aspirant must be well versed with the basics of linear algebra, probability and statistics, optimization theory, and many more. This leaves a unique opportunity for aspirants from various fields to have equal opportunities to compete for jobs in industry irrespective of their background. However, we find people from different backgrounds approaching distinctively for the same set of problems. In this blog, I try to elaborate on the “learning process” in ML which is core and universal.
Machine Learning is defined in many ways from different people from different backgrounds. The essence of machine learning can be pinned down to three main parts:
- We have the dataset.
- A pattern must exist in the dataset.
- We cannot pin down the pattern existing in the dataset mathematically.
Dataset
ML can be applied to the problems of core engineering disciplines, pure and applied sciences, and many fields of computer science like computer vision, NLP, robotics, and bioinformatics, etc. However, fundamentally we need datasets to apply ML. Because data is the core of any learning process.
Pattern and its aesthetics
The second most important thing to apply ML is there must be a pattern within the dataset. Be it prediction or clustering or forecasting we need to create models, that can learn that pattern. Also, it is highly advisable to have a pattern that cannot be pinned down mathematically because if we want to kill a scourge of mosquitoes it is not optimal to kill them with a bomb when a simply poisonous gas can do it.
Let us clearly understand this with the help of some examples:
The above plot is a complex function that can be modeled by advanced ML models. Unlike simple functions like log(X) that can be modeled using simple mathematical and statistical models ML models can model both simple and complex functions.
Let’s look at another example:
In radio communications, an evolved antenna is an antenna designed fully or substantially by an automatic computer design program that uses an evolutionary algorithm that mimics Darwinian evolution. This procedure has been used in recent years to design a few antennas for mission-critical applications involving stringent, conflicting, or unusual design requirements, such as unusual radiation patterns, for which none of the many existing antenna types are adequate.[1] If we closely observe the antenna, we can infer that these kinds of design making it highly impossible for humans to imagine.
