Clicked Gallery

What is a Neural Network?

Highlighted from a real engineering doc. Explained by Clicked.

Used in a sentence

The team replaced its hand-written fraud rules with a neural network trained on five years of transactions.

The reader highlighted one word in the docs. Clicked made the technical term “neural network” easy to understand:

Explained in three depths

●○○

Overview

▾

A neural network is a program that learns from examples instead of following rules anyone wrote. Show it millions of examples and it gradually adjusts itself until the right answers come out, building skills nobody could write down. It is the structure all of modern AI is built from.

●●○

Detail

▾

Learning from examples instead of from written rules is the whole idea: nobody has to write rules, because the network works them out from millions of past examples. Banks once wrote fraud rules by hand, flag large purchases abroad, and thieves slipped between them; a network learns its own patterns from labeled history, including combinations nobody thought to write. The machinery is layers of simple calculators with adjustable dials, and training sets the dials. Fresh dials are random, so the guesses start as garbage. Training shows millions of past transactions labeled fraud or fine, compares each guess to the truth, and nudges every dial the direction that would have improved it, a procedure called backpropagation. Once the dials settle, a new purchase enters as raw numbers, amount, time, place, and each layer turns them into bigger clues, odd hour, first-time merchant, wrong city, until the last layer gives the verdict: fraud or fine. Written rules were sometimes just clumsier; for driving's endless edge cases they are impossible, and that is why example-learning, at billions of dials, became modern AI, transformers included.

●●●

Analogy

▾

How you recognize a friend from a distance just by their walk. Nobody handed you rules for it and there was no lesson; you saw them walk a thousand times, and something inside quietly tuned itself until you know them from a hundred meters, before you can even see a face. Ask yourself what exactly gives them away and you have no answer. The rules exist and they work, but they were learned from examples, never written down, which is exactly a neural network's kind of knowledge.

AI explanations may contain errors · Not professional advice

Same facts, different vibe — Slang mode 😎

●○○

Overview

▾

A neural network is math that learns from examples instead of following written rules: guess, check the miss, adjust the dials, repeat a few million times. Layers of simple units, tuned until the answers come out right. It is the machinery under basically all modern AI. 😎

●●○

Detail

▾

The whole idea is learning from examples so nobody has to write rules, and for recognizing a friend's voice in a noisy bar, nobody could write them anyway. The machinery is layers of little units, each with dials deciding how much weight to give every number coming from the layer below. Training sets the dials, and it is a grind. Fresh dials are random, so the guesses start as garbage; feed it a few million labeled clips, Dave, not-Dave, score each miss, and nudge every dial the direction that shrinks the miss, that nudge is called backpropagation. When the dials settle, showtime: a new recording enters as raw numbers, the sound wave measured thousands of times a second, and the tuned layers stack the clues, wobbles into pitches, pitches into syllables, syllables into a voice. Top layer calls it: Dave. Billions of dials later, same recipe, and you have modern AI, transformers included. 😎

●●●

Analogy

▾

Nobody learns free throws from a physics textbook. You shoot, miss left, adjust, shoot, come up short, adjust again, and a million tiny corrections later your arms just know. Ask for the rules and you have nothing to say; the knowledge lives in the tuning, not in words. A neural network trains exactly like that: guess, check the miss, nudge, repeat until the misses get small.

AI explanations may contain errors · Not professional advice

Formal definition — The same term, explained the usual way

A neural network is a parametric function composed of layered units applying weighted sums and nonlinear activations, trained by gradient-based optimization, typically backpropagation, to minimize error over example data. Depth enables hierarchical feature learning, and scaled members of this family, including transformers, form the basis of contemporary machine learning systems.

Want Clicked to explain terms like “neural network” directly in your browser — including on PDFs?

Add to Chrome — Free

50 free Explanations · No credit card required

More from the gallery

What is Backpropagation?

How one final error becomes a precise adjustment for millions of settings — Goal Seek at scale.

What is a Transformer (in AI)?

The design behind the T in GPT: every word attends to every other word, all at once.

What is Stochastic Gradient Descent (SGD)?

Millions of dials, one wrongness score — how models actually learn, tiny step by tiny step.

Overfitting vs. Underfitting: What's the Difference?

Memorized the answers vs. never opened the book — and the score gap that exposes both.

What is the Bias-Variance Tradeoff?

Too rigid vs. too impressionable — and why total error is lowest in between.

What is Quantization?

Storing a model's numbers less precisely — half the memory, barely any loss.