Machine Learning (ML)

Machine Learning (ML) systems are computational systems designed to learn patterns, make decisions, or predict outcomes from data without being explicitly programmed. They adapt and improve their performance over time through experience and exposure to more data.

Key Components:

Data:

• The foundation of any ML system.
• Includes training data (to teach the model) and testing data (to evaluate the model’s performance).

Algorithms:

• Define how the system learns from the data.
• Examples: Linear regression, decision trees, neural networks and clustering.

Model:

• The output of the training process.
• Represents the learned patterns or relationships within the data.

Training Process:

• The process of feeding data to the algorithm to create a model.
• Adjusts parameters to minimise errors or maximise accuracy.

Inference:

• The phase where the trained model is used to make predictions or decisions based on new data.

Types of Machine Learning:

Supervised Learning:

1. Uses labelled data (input-output pairs) for training.
2. Example: Predicting house prices based on features like size and location.

Unsupervised Learning:

1. Works with unlabelled data to find hidden patterns or groupings.
2. Example: Customer segmentation in marketing.

Reinforcement Learning:

1. An agent learns by interacting with an environment and receiving rewards or penalties.
2. Example: Teaching a robot to navigate a maze.

Semi-Supervised Learning:

1. Combines small amounts of labelled data with large amounts of unlabelled data.
2. Example: Speech recognition systems.

Applications:

• Healthcare: Disease diagnosis, personalised medicine, drug discovery.
• Finance: Fraud detection, algorithmic trading, credit scoring.
• Retail: Recommendation systems, inventory management, customer sentiment analysis.
• Autonomous Systems: Self-driving cars, robotics, drones.
• Natural Language Processing (NLP): Chatbots, language translation, sentiment analysis.
• Computer Vision: Image recognition, facial recognition, medical imaging.

Advantages:

• Automation: Reduces the need for manual programming of rules.
• Scalability: Can process vast amounts of data efficiently.
• Adaptability: Improves with more data and experience.

Challenges:

• Data Quality: Requires clean, representative data for effective learning.
• Bias and Fairness: Susceptible to biases in training data.
• Explainability: Complex models (e.g., deep learning) can be hard to interpret.
• Resource Intensive: Demands significant computational power and storage.