What is a neural network? Many small weighted votes
Big idea: a neural network is many small weighted votes. Whoever changes the weights changes the answer.
The story in short
Every Saturday the Silverline Heights society argued about Sunday’s cricket match. Sky? Ground soggy? Gowda thatha’s aching knee? Exams? Enough players? Bunny and Riya turned the argument into a paper “neuron”: each clue gets a number and a weight (sky +3, ground +4, knee −1, exam −5, players +2). Multiply, add up, and if the total is more than 5, play. This Saturday’s total was 3.6, so no match. Then Happy secretly added a star card weighted +100, and the neuron said “play” no matter what.
The idea
- Inputs: numbers going into a neuron, like how clear the sky is.
- Weights: how much each input counts. Negative weights push towards “no”.
- Weighted sum: each input times its weight, all added up.
- Threshold (activation): the line the total must cross to say “yes”.
- Neural network: many neurons in layers, each layer passing its answers to the next.
- Training data: Prakash uncle’s three-year notebook of Sundays: examples to learn good weights from.
- Tampering: secretly changing weights or inputs to force an answer.
Try the maths
Sky 1 × 3 + ground 1 × 4 + exam 1 × (−5) = 2. Not more than 5, so: no match.
Remember: many small voices, weighed well, make a wise choice. But guard the weights.
Check yourself
A neuron has inputs ground = 1 (weight +4), players = 1 (weight +2) and knee = 1 (weight −1). What is the weighted sum?
With a threshold of "more than 5", does a weighted sum of exactly 5 say "play"?
In a neural network, what is a layer?
More for you: 🎒 Students · 🧑🏫 Parents and teachers