AI Concepts Series - 1) Machine Learning

Listen - Read Aloud

Folks.... I've decided to start a series of articles focussed on demystifying Artificial Intelligence and exploring high-level explanations of some of the core concepts within the field. I'll be using real-life analogies covering a collection of the following topics to help people and businesses get to grips with how AI actually works, what some of the technical principles of the wider tech are & ultimately help people understand how AI will change their day-to-day;

Machine Learning (ML): Foundational concept for many other AI technologies.
Deep Learning: Powering advancements in areas including Natural Language Processing, computer vision, and many other areas.
Natural Language Processing (NLP): Increasingly crucial for applications in data analysis, customer service, and more.
Computer Vision: Essential for numerous applications, from autonomous vehicles to medical imaging.
Convolutional Neural Networks (CNNs): Central to the progress in image recognition and computer vision.
Recurrent Neural Networks (RNNs): Key for processing sequential data, making them vital for speech recognition and language modelling.
Reinforcement Learning: Gaining importance in areas like robotics, gaming, and autonomous systems.
Supervised Learning: A fundamental holistic approach to Machine Learning that forms the basis for many practical applications in AI.
Unsupervised Learning: Important for discovering hidden patterns in data without structured, labeled responses.
Transfer Learning: Increasingly relevant for efficiently leveraging existing models on new tasks.
Generative Adversarial Networks (GANs): Gaining prominence in creative AI applications and data augmentation. Huge impact in OSINT with fake imagery and video.
Robotics: Integrates various AI technologies for practical applications in manufacturing, healthcare, and more.
Object Detection: A critical component in computer vision applications.
Chatbots and Virtual Assistants: Growing in relevance for customer interaction and automation.
Test Time Augmentation (TTA): Gaining attention for improving model robustness and performance.
Autonomous Vehicles: An emerging field that integrates multiple AI technologies.

Join me to gain a real understanding of how these technologies will help us and impact us over the coming years.

What is Machine Learning?

Imagine you're learning to cook. You start with a recipe, perhaps something simple like a spaghetti Bolognese. The first time you make it, you follow the recipe to the letter, but it turns out a bit too salty. The next time, you adjust the amount of salt. Maybe you also get adventurous and add a pinch of your favourite herbs. Over several attempts, your spaghetti Bolognese evolves from the basic recipe to a dish uniquely yours, improving each time based on the results of the last.

This process of trial, error, and improvement is remarkably similar to how machine learning works. At its core, machine learning is about teaching computers to learn from experience, much like how you learn to perfect a recipe.

The Recipe: Algorithms and Data

In machine learning, the recipe is an algorithm, a set of instructions a computer follows to perform a task. But an algorithm needs ingredients, and in this world, the ingredients are data. When you first try cooking, your ingredients are basic - perhaps just what the recipe explicitly calls for. In machine learning, this is like starting with a simple, clear dataset.

Taste Testing: Training and Feedback

Just like how you taste-test your dish and decide if it needs more seasoning, a machine learning algorithm uses data to learn. This is called training. The algorithm makes predictions or decisions based on the data it's given, and just like your taste buds give you feedback on your cooking, the algorithm gets feedback on its performance. This feedback helps the algorithm adjust and improve.

For example, a machine learning algorithm designed to identify spam emails is initially trained on a dataset of emails that are already marked as spam or not spam. It learns to detect patterns and other features common in spam emails - say, certain keywords or sender addresses. The first few times, it might mistakenly categorise a regular email as spam (akin to over-salting your dish), but with continuous feedback - when users mark an email as 'not spam' - it learns and refines its ability to differentiate better. It's super helpful!

The Secret Ingredient: Adaptation

The beauty of machine learning, much like cooking, lies in its adaptability. With every new piece of data, the algorithm adjusts, improving its accuracy and efficiency. This is akin to you refining your recipe with each attempt, learning from past mistakes, and adapting to new tastes and preferences.

In essence, machine learning is a computer's journey from following basic instructions to developing the ability to make decisions and predictions, much like how you evolve from a novice cook following a recipe to a proficient chef who can create a dish by instinct and experience. It's a journey of continuous learning and refinement, driven by the data it consumes and the feedback it receives.

Types of Machine Learning

Just as there are various ways we learn at school, machine learning can be categorised into different types, each with its unique method of learning from data. Let's explore these types through the lens of familiar school learning models.

Supervised Learning: The Classroom Setting

Think back to your school days, sitting in a classroom with a teacher guiding you through a lesson. The teacher provides you with specific information, asks questions, and corrects your answers. This process is akin to Supervised Learning in machine learning.

In supervised learning, the algorithm is provided with labeled data - this means the data comes with answers (like a teacher providing correct answers). For example, in a dataset for facial recognition, each image is labeled with the name of the person in the photo. The algorithm uses this data to learn and make predictions or decisions. When you train an algorithm to recognise faces, you are like the teacher providing it with a 'correct answer' for each image, which it then uses to learn and improve.

Unsupervised Learning: Self-Study and Exploration

Now, imagine you're given a research project or a puzzle to solve on your own. You're not given specific guidance but need to explore and find patterns or solutions yourself. This scenario resembles Unsupervised Learning.

In unsupervised learning, the algorithm is given data without any labels or specific instructions on what to look for. It must explore the data and identify patterns or structures on its own. A practical example is customer segmentation in marketing. The algorithm analyses customer data and groups (segments) customers with similar behaviours or traits, without being told what specific features to look for. It's like you grouping puzzle pieces based on color or shape without knowing the final picture, it's getting better every day and I imagine there are some key announcements around this tech on the horizon.

Reinforcement Learning: Learning a Sport or a Musical Instrument

Consider how you learn a sport or a musical instrument. It's not just about right or wrong answers; it's about practice, trial and error, and gradually improving based on feedback from your coach or the musical tones you produce. This is the essence of Reinforcement Learning.

In reinforcement learning, an algorithm learns by doing and receiving feedback in the form of rewards or penalties. It's like a game where the algorithm makes decisions, and based on the outcomes (positive or negative), it adjusts its strategy. A well-known example is AI playing and mastering video games. The AI 'practices' the game, receiving points (rewards) or losing lives (penalties), and over time, it learns the best strategies to maximise points.

Each type of machine learning offers a unique approach to learning from data, much like how we have different learning experiences at school. Whether it's the guided learning in a classroom, the exploratory nature of self-study, or the trial-and-error process in sports or music, hopefully these analogies help demystify the complex types of machine learning and make them more relatable to your everyday experiences.

How Machine Learning Works in Everyday Life

To truly grasp the impact of machine learning, let's look at something many of us enjoy as we approach Christmas: shopping online. Imagine having a personal shopping assistant who observes your style preferences, notes the brands you love, and even remembers the items you linger on but don't buy. This assistant then makes tailored recommendations just for you. This scenario is not just a luxury experience but a real-world application of machine learning in the form of recommendation systems.

The Personal Shopping Assistant: Recommendation Systems

When you shop online, whether it's for clothes, books, or groceries, machine learning algorithms work quietly in the background, playing the role of your personal shopping assistant. These algorithms analyse your past shopping behaviour, browsing history, and even items in your shopping cart or wish list to understand your preferences and interests.

Learning Your Style: Data Analysis and Pattern Recognition

Just as a human assistant would learn your style over time, machine learning algorithms identify patterns in your behaviour. They notice the colors you prefer, the brands you frequently purchase, and the price range you're comfortable with. This is achieved through analysing vast amounts of data - your shopping history - and identifying trends and preferences unique to you.

For instance, if you've been browsing a lot of science fiction books lately like me, the algorithm takes this as a cue to recommend similar books. Or, if you frequently purchase eco-friendly products, it might start highlighting more environmentally sustainable options in your future searches. The worry is when large companies start to point you down one track or another.

Making Recommendations: Predictive Analytics

Once the algorithm understands your preferences, it uses predictive analytics to recommend products you might like. This is similar to your shopping assistant suggesting a new book by your favourite author or a dress in your preferred style and colour. The algorithm uses the data it has gathered about you to make educated guesses about what you might want to buy next.

In practice, this means when you log in to your favourite shopping site, you see a 'Recommended for You' section filled with items that align with your past behaviour and preferences. It's like walking into a store where everything on display is tailored to your taste.

The Feedback Loop: Continuous Improvement

The most intriguing part of this process is the feedback loop. Every time you interact with these recommendations - clicking, ignoring, or purchasing - you provide the algorithm with more data. This feedback helps the algorithm refine its understanding of your preferences, much like how a personal assistant learns from your reactions to their suggestions.

For example, if you consistently ignore or remove a certain type of item from your recommendations, the algorithm learns to avoid suggesting similar items in the future. This continuous learning process ensures that the recommendations become more accurate and personalised over time.

I hope this has helped you get a better idea of how Machine Learning works in principle and you're able to see the similarities between how we train the machine and how we learn ourselves!