Introduction to ML Series (Part 1)
Im a Postgraduate student who is pursuing Data science and Artificial Intelligence program. I want to share what I learn program the program with you in Weekly basis. For me, I think it will broaden my knowledge and the understanding of the subject. Also later on, I can refer my notes, articles for myself when Im getting ready for my exams. Also for you (as a person who follows my series), you will get the relevant knowledge from beginning to end in a proper structural way. Hopefully at the end you will have same knowledge as me which you can confidently say “I have the relevant knowledge even though I didnt officially follows the MSc”.
Warning : Don’t expect senior or expert level of content explanations because I’m also a student who is passionate about AI/ML.
Also remember this is not a complete copy of the syllabus, learning materials of the subject which is a violate of their terms, therefore everything I include/written from this article to upcoming articles are my interpretation of the subject.
To continue this series you should have pre-requisites as below.
Above aren’t necessarily need for this article but surely will need in upcoming articles to understand the underground concepts.
What you’ll learn this series are below.
- Introduction to ML
- Nearest Neighbor Method
- Regression
- Probability Learning
- Support Vector Machine
- Reinforcement Learning
From this article I’ll discuss the topic “Introduction to ML”.
We hear about ML everywhere which it powers our Netflix recommendations, helps our phones understand our voice, and is paving the way for self-driving cars. But what is it, fundamentally? Let’s peel back the layers, core concepts that form the bedrock of this fascinating field.
Why Now? (Data Overload)
First, why is ML exploding now? The answer is simple “data”. We generate large amounts of it every single day. From social media interactions and online purchases to scientific experiments and sensor readings. Think terabytes upon terabytes, constantly flowing.
The challenge that Humans aren’t equipped to manually sift through this digital ocean to find meaningful patterns, especially when data has many dimensions (features). Trying to visualize data with more than three features is inherently difficult for us. Imagine trying to plot customer behaviour based on age, location, purchase history, time spent on site, device used, and browser type. It just quickly becomes impossible for our brains to grasp intuitively.
This is where ML shines. We can task computers, which don’t get bored or overwhelmed by high-dimensional complexity to learn from this data and uncover insights or make predictions that would otherwise remain hidden.
OK. Now What Exactly is ML? is it Beyond Traditional Programming?
Traditionally, we solve problems by writing explicit, step-by-step instructions for a computer (an algorithm). Need to sort a list of numbers? There are well-defined algorithms for that.
But what about tasks like identifying spam emails(very common example in the ML)? The “rules” for what constitutes spam are fuzzy, change over time, and can be personal. We can’t easily write a fixed set of instructions. Instead, we use ML.
Simply, ML gives computers the ability to learn from data without being explicitly programmed for every scenario.
Arthur Samuel, a pioneer in AI, defined it back in 1959 as the “field of study that gives computers the ability to learn without explicitly programmed.” A more operational definition comes from Tom Mitchell (1998): “A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E.”
Now break down that spam filter example using Mitchell’s definition.
- Task (T): Classifying emails as spam or not spam.
- Experience (E): Observing which emails you manually mark as spam or not spam.
- Performance (P): The percentage of emails correctly classified by the program.
As the program sees more examples (E), its accuracy (P) at the classification task (T) should improve. The core idea is generalization (the ability to perform well on new, unseen data based on patterns learned from the training data).
The Three Main Flavors of Machine Learning
ML isn’t monolithic; it comes in different types, primarily distinguished by the kind of data and feedback available during learning.
- Supervised Learning (Learning with a Teacher)
This is like learning with a teacher who provides the correct answers. The algorithm is trained on a dataset where each data point (input) is paired with a correct label (output). the Main goal is to learn a mapping function that can predict the output for new, unseen inputs.
2 main Sub types are,
- Classification: The output is a discrete category. Think assigning an email to “spam” or “not spam,” identifying handwritten digits (0–9), or diagnosing whether a medical image shows a specific condition. The algorithm learns “decision boundaries” to separate the classes.
- Regression: The output is a continuous numerical value. Examples include predicting the price of a house based on its features (size, location, age), forecasting future sales demand, or estimating a patient’s length of stay in a hospital.
2. Unsupervised Learning (Discovering Patterns)
Here, there’s no teacher and no correct answers provided in the training data. The algorithm explores the data on its own to find hidden structures or relationships. The main goal is to discover inherent patterns, groupings, or representations within the data.
2 main sub types are,
- Clustering: Grouping similar data points together. This is used for customer segmentation (finding groups of customers with similar buying habits), grouping related news articles, or identifying communities in social networks.
- Dimensionality Reduction: Simplifying data by reducing the number of features while preserving important information. This can be useful for visualization or as a pre-processing step for other ML algorithms.
3. Reinforcement Learning (Learning through Trial and Error)
This is like learning by doing, receiving feedback along the way. An “agent” interacts with an “environment” and learns to take actions that maximize a cumulative “reward.” It doesn’t get told the “right” action beforehand, but receives positive or negative feedback (rewards or penalties) for the actions it takes. The main goal is to learn an optimal strategy (policy) for decision-making in a potentially complex environment. Checkout below example so you can an Idea.
- Training an AI to play games (like Chess or Go), robotics (teaching a robot to walk or grasp objects), optimizing traffic light control systems, or developing recommendation systems that adapt to user feedback over time.
New research Directions
These core concepts power a vast array of applications, from the image recognition systems identifying objects in photos (like the famous ImageNet challenge) to the natural language processing behind virtual assistants. Cutting-edge research areas like Deep Learning (using complex neural networks) have driven massive progress, while fields like Explainable AI (XAI) aim to make ML models less like “black boxes,” and Generative Adversarial Networks (GANs) can create startlingly realistic synthetic data, images, or even music.
That’s the end of this article and I hope you got an idea about the outline and background of ML. Don’t foget to practice and refer the learning materials I provided in the article. As you know this is my first article in the ML series (will publish every week) and if you think this can be improved your feedbacks are highly welcome. Don’t forget to follow me then you’ll get notified when new article dropped. Stay Consistent 💪
Contact me via LinkedIn
