🧠 Neural Networks

A neural network is an ML model inspired by the structure of the human brain. It consists of layers of interconnected nodes (called neurons or perceptrons) that process data through mathematical transformations. Data flows in through the input layer, gets transformed by one or more hidden layers, and produces a prediction at the output layer.

Think of it like a factory assembly line. Raw materials (input data) enter at one end. Each station on the line (hidden layer) processes and transforms the materials in some way. By the end of the line, a finished product (prediction) comes out. The magic is that the factory learns how to configure each station by looking at thousands of examples of what the finished product should look like.

How a Single Neuron Works

Each neuron receives inputs, multiplies each input by a weight (a number that indicates how important that input is), adds them together, adds a bias (a baseline value), and passes the result through an activation function (which decides how strongly this neuron "fires"). If the combined value is above a threshold, the neuron activates and passes its signal to the next layer. If not, it stays quiet.

This is loosely analogous to how biological neurons work: they receive signals from many other neurons, and if the combined signal is strong enough, they fire and send a signal to the next neuron.

How Neural Networks Learn (Backpropagation)

Training a neural network involves showing it thousands of examples and adjusting the weights to reduce errors. The process is called backpropagation: (1) Feed an input through the network and get a prediction. (2) Compare the prediction to the correct answer and calculate the error. (3) Work backwards through the network, adjusting each weight slightly to reduce the error. (4) Repeat for every training example. Over thousands of iterations (called epochs), the weights gradually converge on values that produce accurate predictions.

Deep Learning: Neural Networks with Many Layers

When a neural network has many hidden layers (typically 10 or more), it is called a deep neural network, and training it is called deep learning. Each additional layer allows the network to learn more abstract, complex features. In image recognition: the first layer detects edges, the second detects shapes, the third detects object parts, the fourth detects whole objects. This hierarchical feature extraction is what makes deep learning so powerful for complex tasks like image recognition, speech processing, and language understanding.

💡 Strengths & Limitations

Strengths: Can learn extremely complex patterns (images, language, audio). Automatically discovers relevant features from raw data. Scales well with more data and compute power. Powers the most impressive AI achievements (ChatGPT, Face ID, AlphaGo, DALL-E).

Limitations: Requires massive amounts of data and compute power. Acts as a "black box" — hard to explain WHY it made a specific prediction. Prone to overfitting without regularisation. Training can take hours, days, or weeks on expensive GPU hardware.

Used for: Image recognition (Face ID, medical imaging), natural language processing (ChatGPT, Google Translate), speech recognition (Siri, Alexa), game playing (AlphaGo), self-driving cars (Tesla), art generation (DALL-E, Midjourney).

✏️ Describe what a neural network is and its key features. (2–3 sentences)