Machine Learning & Neural Networks
1. AI vs. Machine Learning vs. Deep Learning
AI is the broad concept of machines acting "smart." Machine Learning (ML) is a subset where the computer "learns" from data without being explicitly programmed.
Supervised Learning
The model is trained on a labeled dataset (Inputs + Correct Answers). It learns to predict the label for new data.
Example: An email filter trained on "Spam" vs "Not Spam" emails.
Unsupervised Learning
The model finds hidden patterns in unlabeled data. It groups similar items together.
Example: Netflix recommending movies based on "users like you."
2. Graphs in AI (Dijkstra’s & A*)
In the syllabus, AI often uses Graphs (Nodes and Edges) to solve problems. A search algorithm like A* Search uses a Heuristic (an educated guess) to find the shortest path efficiently.
3. Artificial Neural Networks (ANN)
ANNs are inspired by the human brain. They consist of layers of "neurons" that pass signals to each other. Each connection has a Weight that changes as the model learns.
- Input Layer: Takes the raw data (e.g., pixels of an image).
- Hidden Layers: Where the complex math and "pattern finding" happens.
- Output Layer: The final decision (e.g., "This is a Cat, 98% probability").
Process: Backpropagation is the method used to adjust weights based on the error of the output.
4. Real-World AI Products (What Students Use)
Students today interact with AI in almost every app. Here are the syllabus concepts mapped to real life:
- Large Language Models (LLMs): ChatGPT Gemini Claude
These use Deep Learning and Transformers to predict the next token in a sentence. - Computer Vision: FaceID Self-Driving Teslas Snapchat Filters
These use Convolutional Neural Networks (CNNs) to identify features like eyes, lanes, or obstacles. - Generative AI: Midjourney Suno GitHub Copilot
Models trained to create new data (images, music, or code) rather than just classify it.
5. Ethics and Social Impact
A-Level exams often ask about the "dark side" of AI. Key concepts include:
- Bias: If the training data is biased, the AI will be biased (e.g., a recruitment AI that only picks men because it was trained on historical data).
- Accountability: If a self-driving car crashes, who is responsible? The programmer? The owner? The AI?
- Jobs: Automation of repetitive tasks (Coding, Data Entry) vs. the creation of new AI-maintenance roles.
An older form of AI. It uses a Knowledge Base (facts) and an Inference Engine (IF-THEN rules) to give advice, like a digital doctor. Unlike Neural Networks, it cannot "learn" on its own—it only knows what humans tell it.