machine-learning

Neural networks & deep learning

What is machine learning (ML)?

• Machine learning is a type of artificial intelligence (AI) that allows computers to:

  • Learn patterns from data

  • Improve performance without being explicitly programmed

• Instead of following fixed rules, machine learning systems:

  • Analyse large amounts of data

  • Identify patterns or trends

  • Make predictions or decisions based on that data

• ML is used in: spam filters, voice assistants, recommendation systems, fraud detection

What are artificial neural networks (ANNs)?

• Artificial neural networks (ANNs) are algorithms inspired by the structure of the human brain

  • A neural network is made up of layers of nodes (neurons) connected by weighted links

  • Each neuron receives input, processes it, and passes the result to the next layer

  • The network adjusts the weights based on errors in output (using algorithms like backpropagation)

How ANNs help machine learning

• ANNs are powerful because they:

  • Automatically learn from experience, even with complex or unstructured data

  • Improve their accuracy with more data and training

  • Can solve problems that are too complex for rule-based programming

• Examples: handwriting recognition, speech recognition, image classification, language translation

Deep learning (DL)

• Deep learning is a subfield of machine learning that uses deep (multi-layered) neural networks

  • The more layers a network has, the more complex patterns it can learn

  • Especially effective at learning abstract features in images, text, audio, and video

• Example: facial recognition apps learn features like eyes, noses, and then full faces through many layers

Reinforcement learning (RL)

• Reinforcement learning is another type of machine learning where an agent learns by interacting with an environment

  • It receives rewards for good actions and penalties for poor actions

  • Over time, the agent learns an optimal strategy (called a policy) to maximise rewards

• Used in: robotics, self-driving cars, game-playing AI (like AlphaGo), industrial automation

Summary: why use these methods?

Categories of machine learning

• Machine learning algorithms are categorised based on how they learn from data

• The two main categories are:

Supervised learning

• Supervised learning is when the algorithm is trained on a labelled dataset

• Meaning the input data has known outputs (answers)

  • The goal is to learn a function that maps inputs to the correct output

  • Once trained, the model can make predictions on unseen data

• Example:

  • Input: Email text

  • Output: Spam or Not Spam

  • The model learns from thousands of pre-labelled emails

Used for:

• Classification (e.g. face recognition, spam detection)

• Regression (e.g. predicting house prices)

Unsupervised learning

• Unsupervised learning is when the algorithm is given unlabelled data, and it must find patterns or groupings on its own

  • There are no correct answers, the system explores the data structure

  • It identifies clusters, trends, or anomalies without any guidance

• Example:

  • Input: Customer purchase data

  • Output: Groups of customers with similar buying habits

Used for:

• Clustering (e.g. customer segmentation, social network analysis)

• Dimensionality reduction (e.g. simplifying complex data)

Back propagation & regression

What is back propagation?

• Back propagation is a training method used in artificial neural networks to improve accuracy by adjusting the weights of connections

• It is a key part of the supervised learning process and works as follows:

1. Forward pass:

   1. Input data passes through the network layer by layer

   2. The network produces an output (a prediction)

2. Error calculation

   1. The output is compared to the actual target value

   2. The difference is called the error

3. Backward pass

   1. The error is propagated backwards through the network

   2. Each layer calculates its contribution to the error

4. Weight adjustment

   1. The weights of the connections are updated using an algorithm (e.g. gradient descent) to reduce future errors

Why use back propagation?

• Helps the neural network learn from mistakes

• Makes the model more accurate over time

• Allows multi-layer networks to fine-tune all layers, not just the output

• Common in: image recognition, speech recognition, language translation

What is regression?

• Regression is a type of supervised learning used to predict continuous values (rather than categories)

• The aim is to find the relationship between input features and a numerical output

Types of regression:

Example: linear regression

• Predicting house prices based on square footage:

  • Input: size of house

  • Output: estimated price

  • The model learns a line that best fits the data points

Why use regression methods?

• Useful for forecasting, trend analysis, and risk prediction

• Interpretable, often used in real-world problems like:

  • Business analytics

  • Finance

  • Healthcare