Day 1 of Learning Machine Learning

So, I recently started Andrew Ng’smachine learning course. Here’s what I learned on day one.

Supervised Learning:

When we train the learning model by giving it problem/solution pairs so that it figures out the pattern and can then be used for future problems of a similar nature.

Some examples:

Giving it emails, showing it signs of spam: we get spam checker AI.

Giving it written text, showing it common grammatical errors, and what grammatically correct writing looks like: We get grammar-checking AI.

Giving it English scripts and showing it what the translated version looks like in other languages: We get language translator AI.

At its core, supervised learning is feeding the learning model data with correct labels.

A few more examples:

By giving it healthy plant images and showing it different diseases, we get an AI model that can check if a plant is healthy or diseased. The same type of model can be used for humans and animals with a different data set.

Fraud detection: showing the model different transactions and giving it different examples of common fraud tropes. The learning model can then be used to detect fraudulent transactions.

Supervised learning has two types: regression and classification. (Those are the ones that I am aware of at the moment.)

Regression refers to the method in which we ask the model to predict a number (value) from an infinite set of options. The more data we provide the model, the better its prediction will be.

Classification refers to the method by which we ask the model to categorize something into two or more categories.

That’s it for supervised learning, at least for now. Moving onto unsupervised learning.

Unsupervised Learning

In unsupervised learning, we don’t give the model any prior examples of our desired problem-solution pair. It’s created to find patterns in the data we give it.

So we don’t supervise the learning of the model in this method, but it itself takes note of interesting insights, patterns, and anomalies in the data.

Clustering is probably the most used type of unsupervised learning in the tech world. Google News is a good example of clustering. The machine learning model groups different news items that have the same keywords every day.

Another example of clustering, I guess, can be music recommendation systems that observe the type of music (genre, words, sentiment) you listen to and suggest more of the same. Instagram, Explore, and TikTok are also doing the same thing.

Another popular use case for unsupervised learning is anomaly detection and dimension reduction.

That’s it for today. I’m satisfied with the beginning. I will be back tomorrow with more learning.