Machine Learning 101

If you live anywhere on this earth you have probably been hearing the word Machine Learning being dropped a lot lately. And on some occasions, you might even find yourself in a conversation about ML and AI where you hear terms that mean nothing to you. In this article, I am going to demistify the ML jargon for you, in simple words.

What is Machine Learning?

If you google the term “Machine Learning” you will stumble upon definitions that sounds pretty much like the following:

Machine learning (ML) refers to a system’s ability to acquire, and integrate knowledge through large-scale observations, and to improve, and extend itself by learning new knowledge rather than by being programmed with that knowledge.

Now let’s translate this to english!

Similar to a growing child that learns by observing the world around it, a machine learning model learns from the data provided to it. Unlike traditional systems that follow a set of rigid rules, it adapts and changes in response to the data it encounters. In other words, the same ML algorithm will yield different results when applied to different datasets. This makes ML the perfect approach to solve a variety of everyday problems where the result is not always static, such as demand forecasting, disease diagnosis and speech recognition.

Now let’s learn some basic terms to be able to understand the rest of the article:

• Machine Learning Algorithm: It’s a step-by-step set of instructions, mainly based on mathematical and statistical theory, that the model follows to learn from the provided data. Think of it as the way a baby’s brain processes infromation to realte a picture of a cat to the word “Cat”.

• Training Data: Simply put, it’s the set of examples you feed the algorithm to learn from. It can consist of input-output pairs, or just inputs, where the inputs are the features or attributes, and the outputs are the corresponding labels or targets. It’s like showing a baby (the algorithm) pictures of different animals while saying their names and hoping they make the connection.

• Machine Learning Model: A machine learning algorithm uses training data to produce a model capable of making predictions or decisions, we call this a machine learning model. In our example, the model will be the baby that is now capable of telling the animal’s name based on their picture.

💡 In summary: ML Algorithm + Training Data = ML Model

Machine Learning Types

Now, ML is categorized into 3 types:

• Supervised Learning: Which I like to define as “when you know what you are looking for”, or in other words, it’s when your dataset is labeled. In this case your model will learn from the dataset you feed to it and will be able to predict the label for unseen data. Now, be careful, despite what the name might suggest, a label is not always a category or a class. It can also be a number, such as predicting a movie score based on the cast.

• Unsupervised Learning: This type is like wandering in the dark without a flashlight—you don’t know what you’re looking for. In unsupervised learning, the dataset isn’t labeled, meaning the algorithm must figure out patterns or groupings on its own. Imagine trying to identify different species of flowers based on petal colors and shapes without knowing their names. Clustering algorithms, like K-means, are often used here to group similar data points together, helping you discover hidden structures within the data.

• Reinforcement Learning: Think of this as training a puppy with treats. In reinforcement learning, an agent learns to make decisions by performing actions in an environment and receiving feedback in the form of rewards or penalties. For instance, if our puppy (the agent) sits on command (the action), it gets a treat (the reward). The goal here is for the agent to maximize its cumulative reward over time, effectively learning the best strategies for different situations. This approach is commonly used in game playing, robotics, and navigation tasks.

Key Concepts in Machine Learning

Now that we’ve covered the basics, let’s dive into some key concepts that will further enhance our understanding of machine learning:

• Overfitting: Imagine if the baby we talked about earlier memorizes every single detail about the pictures instead of understanding the general characteristics of each animal. That’s overfitting! It occurs when a model learns the training data too well, including noise and outliers, resulting in poor performance on new, unseen data. To combat overfitting, techniques like cross-validation, regularization, and using simpler models are employed.

• Underfitting: On the flip side, underfitting is when the model is too simplistic to capture the underlying patterns of the data. It’s like a baby who can only recognize a dog if it’s wearing a specific hat—if the hat changes, it doesn’t know what to do. This happens when a model is not complex enough or when there’s not enough training data. To improve underfitting, you can use more complex models or provide additional features to the model.

• Feature Engineering: This is the process of selecting, modifying, or creating new features (attributes) from raw data to improve model performance. Good feature engineering is often key to the success of a machine learning project. Think of it as providing the baby with better toys or tools to learn more efficiently—if the toys are too basic, the baby won’t learn much!

Why Should You Care?

You might be wondering why you should even bother learning about machine learning. Well, ML is transforming industries from healthcare to finance, and understanding its fundamental concepts can empower you in your career and daily life. Whether you’re looking to enhance your skills, dive into data science, or simply keep up with tech trends, grasping the basics of machine learning is a valuable investment in your future.

As we continue this journey through the world of machine learning, we’ll explore specific algorithms, tools, and practical applications in the next sections. So stay tuned!