Neural networks — AIF-C01

What it is

A neural network is a method in artificial intelligence that teaches computers to process data in a way that is inspired by the human brain. It uses interconnected nodes — artificial neurons — arranged in a layered structure to recognize patterns in data.

Every connection between nodes carries a weight — a positive number if one node excites another, or a negative number if one node suppresses another. The network learns by adjusting these weights through a process called backpropagation: paths that produce correct outputs gain higher weights; paths that produce incorrect outputs are down-weighted.

Mental model

Think of a neural network as a relay of decision filters. Raw data enters on the left. Each hidden layer asks a slightly more abstract question about that data. By the time information reaches the output layer, the network has distilled millions of small pattern-checks into a single answer. No single neuron "knows" the answer — the knowledge lives in the collective pattern of weights across the connections.

When to use it

Neural networks are appropriate when the relationship between inputs and outputs is too complex to express as hand-crafted rules, and when enough labeled training data is available to tune the weights.

Situation	Neural network	Simpler ML model
High-dimensional, unstructured data (images, audio, text)	Strong fit	Typically insufficient
Small, structured tabular dataset	Often overkill	Frequently a better starting point
Pattern too complex for manual feature engineering	Strong fit	Requires explicit feature design
Interpretability is the primary requirement	Harder to explain	Easier to inspect

The key insight the exam tests: neural networks are not the universal default. Simpler machine learning models are still the right tool when data is limited or interpretability is required.

Common misconception

Misconception: a neural network and deep learning are the same thing.

They are not. A neural network is the broader structure — the layered arrangement of artificial neurons and weighted connections. Deep learning is a specific type of neural network that contains several hidden layers, often with millions of artificial neurons linked together. Deep learning networks also require substantially more training data — millions of examples, rather than perhaps the hundreds or thousands that a simpler network might need.

In other words: all deep learning models are neural networks, but not all neural networks are deep learning models. Conflating the two leads to misidentifying which tool fits a given problem, which is exactly the kind of distinction the exam probes.

How it shows up on the exam

The cognitive target for this concept is recognition and differentiation: can you identify what a neural network is, name its structural components, and distinguish it from related terms like deep learning and machine learning?

Watch for scenarios that describe a system learning from examples using layers of interconnected nodes — that phrasing points to neural networks. A common candidate confusion is treating "neural network," "deep learning," and "machine learning" as interchangeable synonyms. The official AWS documentation draws clear boundaries: machine learning is the parent discipline, neural networks are one method within it, and deep learning is a specific subtype of neural network defined by its many hidden layers and large data requirements.

Grounded signal phrases to recognize:

• "interconnected nodes" or "neurons in layers" → neural network
• "several hidden layers" or "millions of parameters" → deep learning, a subtype
• "learns from examples without explicit rules" → machine learning (the broader category)

Related concepts

• Artificial intelligence overview
• Deep learning
• Machine learning basics