Machine Learning “Demystified”!

Photo by Pietro Jeng on Unsplash

Machine Learning — The “Basics” Demystified!

The goal of this article is to introduce the reader to the concept of machine learning.

Definition

“Machine Learning is the science of getting computers to learn and act like humans do, and improve their learning over time in autonomous fashion, by feeding them data and information in the form of observations and real-world interactions.” — Emerj

There are hundreds of definitions for Machine Learning out there, but the one that resonated with me the most was from Emerj. We will understand the definition of Machine Learning by breaking this down and understanding it in detail.

Machine Learning is the science of getting computers to learn and act as humans do…

So how do humans learn?

1. We first start by observing what we intend to learn and storing it in our memory. Storing data in our minds is the first step towards advanced reasoning.
2. We then assign meaning to this stored data to create an abstract meaning, which is formed based on observations and identifying patterns.
3. We then generalize this abstracted knowledge to be utilized for any future action on similar tasks. Note that the tasks need not be identical. For instance, if you know how to play cricket, the mind uses the same constructs to play a new game, such as squash.
4. To master what we have learned, we evaluate the outcomes. For instance, one might work on physical strength to get better at a sport or work on agility and concentration.

Machine Learning uses the same steps as mentioned above for learning.

1. The first step in creating any machine learning algorithm is to provide a large enough data set for the system to learn.
2. This raw data is summarized using models which best describe the patterns within the data.
3. There could be several patterns in the underlying data. The algorithm employs a process called heuristics, which are educated guesses to find the most useful inferences. This is synonymous with generalization.
4. The algorithm is then evaluated with test data to check its success. There will be additional training and tuning as needed.

PS: Traditional programming involved writing a computer program to generate output for a given set of inputs. However, with machine learning, the inputs and the outputs are fed to the system to create a program. The program is used for further prediction.

..improve their learning over time in an autonomous fashion

To understand this statement, let us look at the most common types of Machine Learning algorithms.

1. Supervised Learning: In supervised learning, the model is provided with labeled training data. Labeled training data means both the inputs and expected outputs are known. Consider we are building an algorithm to identify different vegetables. We start the process by providing a large sample set of vegetable photos along with their names. Features such as shape, size, color, etc., are specified along with its name (label). Through training, the relationship between features and labels is identified, incrementally. The relationship is referred to as a model. The model is used to then predict values automatically for any vegetable, including what it has not seen previously.
2. Unsupervised Learning: In unsupervised learning, the machine needs to identify patterns in an unlabeled data set. In other words, the model has no hints on how to categorize each piece of data and must infer its own rules for doing so automatically. Once the system identifies patterns, it can predict values further.

As you can see in both the examples, with training, the machine learning models improve autonomously.

by feeding them data and information in the form of observations and real-world interactions

It must be apparent by now that the most fundamental step to creating a robust machine learning algorithm is to feed the system with appropriate and diverse training data. The Quality of the training data determines the effectiveness of the eventual model.

I hope you now have a clear understanding of machine learning. In the next post, we will explore how to go from a machine learning problem to a solution.