Deep Learning Fundamentals
Neural networks and layers
The story: A fruit sorting line in a factory has three rows of workers. The first row only looks at each fruit and calls out simple facts: "round", "red", "shiny". The middle rows combine those facts: "round and red and shiny sounds like an apple". The last row makes the final call and drops the fruit in the "apple" bin.
In AI/AWS terms: The workers are nodes. The first row is the input layer, the middle rows are the hidden layers (there can be one or many), and the last row is the output layer. Together they make an artificial neural network, loosely modeled on how the brain passes signals between neurons.
For the exam: A neural network has an input layer, one or more hidden layers, and an output layer, all made of nodes.
How the network learns
The story: On the first day, the workers get a lot wrong. Each time a supervisor corrects them, they change how much they listen to each other. "I'll trust the person who checks color more, and the person who checks size less." After thousands of fruits, they sort fruit they've never seen before correctly.
In AI/AWS terms: How much each worker listens to another is a weight, the strength of a connection between nodes. Training means adjusting those weights as the network sees examples, until it recognizes patterns in data it has never seen.
For the exam: Neural networks learn by adjusting the weights of the connections between nodes.
Computer vision
The story: A security guard watches camera feeds. They can say "this is a parking lot" (what the whole picture is), "there's a car here and a person there" (where things are), and trace the exact outline of the car (which pixels belong to it).
In AI/AWS terms: Computer vision uses deep learning to understand images and video. Naming the whole picture is image classification. Finding and boxing each thing is object detection. Tracing the exact outline is image segmentation.
For the exam: Computer vision covers image classification, object detection, and image segmentation.
Natural language processing
The story: A busy receptionist sorts the mail into "bills" and "letters", can tell from a customer's email that they're angry, translates a note from a Japanese visitor, and writes replies.
In AI/AWS terms: Natural language processing (NLP) uses deep learning on human language. Sorting mail is text classification. Reading the mood is sentiment analysis. Translating is machine translation. Writing replies is language generation.
For the exam: NLP covers text classification, sentiment analysis, machine translation, and language generation.