Foundations of AI: Module 2 – Introducing Machine Learning

AI and Machine Learning (ML)

In my Introducing Artificial Intelligence post, I mentioned that AI encompasses a spectrum of technologies to perform different tasks, including the ability to replicate human intelligence. This includes the capability to learn from experiences, unlike the traditional way where systems and machines were given a set of pre-defined rules to follow.

But how can machines learn?

It’s like us human beings. Whatever we do, whether reading a book, studying, playing, or interacting with other human beings, our mind continuously collects and processes data. 

AI does the same thing…

Machine Learning, a branch/subset of AI, is what allows ‘machines’ to learn. They use algorithms and statistical techniques to learn from data, allowing them to draw patterns. Through these patterns, the machines can make predictions and/or decisions without explicit programming. A typical example of this behaviour is the ‘recommendation’ system used by several applications and websites, including Netflix, Amazon and your smartphones! 

How does learning work?

First, through a Decision Process. As discussed earlier, it tries to make guesses/predictions through data patterns.

The second step is through what is called the “Error Function”. This is a mathematical measure of how well the predictions were when compared to the ground truth, therefore, a measure of how accurate the model is.

Finally, through a Model Optimisation (training and learning) process. This is where efforts are made to increase the model’s accuracy by fine-tuning the model’s parameters/weights, reducing the gap between the predicted and known values.

Let’s take an example.

A user logs into Netflix for the first time and provides information about the type of movies s/he likes. The ‘recommendation system’ built into Netflix will use that information, and through an algorithm, it provides the user several movies that s/he might enjoy, based on parameters such as actors and genre. This all falls under the Decision Process.

Following this, the system will then measure how well its predictions were. This is done by monitoring how many of its recommended movies were watched by the user. If none, the gap between the predicted and the ground truth – i.e., what the user wanted – is significant. This is what is called the Error Function phase.

The Model Optimisation phase will then kick-off, where the built-in algorithm will adjust the weights of each parameter with the intent to reduce this gap. This means that the system’s accuracy will continue improving with each run as it teaches itself from the data it analyses. 

The fascinating thing is that this iterative nature of learning is done autonomously without human intervention!

What are the Machine Learning methods?

Machine learning happens in four primary different ways:

1. Supervised Learning – in simple terms, the model is trained on what is called “labelled data”, where it is fed with a set of data inputs and its corresponding correct output. By continuously adjusting its accuracy through the error function, it can establish the right algorithm/pattern for prediction when fed with a new set of data input.  
2. Unsupervised Learning – this works the opposite of Supervised Learning, where the model is fed with “unlabelled data”. This means the model tries to find patterns, relationships, and clusters within the provided data without indicating how the output should look. This is done without any human intervention.
3. Semi-Supervised Learning – as the name suggests, this is a mix of Supervised and Unsupervised learning where the model is trained on a data set containing labelled and unlabelled data. The labelled data guides the learning process, while the unlabelled data helps the model uncover patterns and relationships, allowing the model to learn from both types of data sets. This type of learning is beneficial when it is very costly or time-consuming to obtain a large amount of labelled data.
4. Reinforcement Learning – in this type of learning, the algorithm learns through interaction with the external environment. Simply put, it receives positive and negative feedback for each good and bad thing it performs, allowing the system to learn through trial and error. This means the algorithm isn’t trained on sampling data but learns as it goes by. 

Machine Learning has drastically changed the way we work. We are no longer instructing computers to do specific activities for us, but they can now learn from us, adapt to our needs, come up with their suggestions, and sometimes, act on our behalf.

As I delve more into machine learning algorithms, we will witness the power of data and how the convergence of data and advancements in technology and computational techniques will allow machines to uncover insights that would remain hidden from the human eye.

As the objective of this post is to offer newcomers in the field of Technology a foundational understanding, I have intentionally simplified certain aspects of the content. If you seek more comprehensive and precise information, please don’t hesitate to reach out.