AI glossary
Neural network
A model made of layers of simple connected units whose connection strengths (weights) are adjusted during training.
A neural network is loosely inspired by the brain, but it is really a large mathematical function. Each unit multiplies its inputs by weights, adds them up and passes the result on. Training adjusts billions of these weights so the network’s outputs get closer to the right answers.
When you hear that a model has “70 billion parameters”, those parameters are mostly the network’s weights.
Example: To recognise handwritten digits, a network takes the image’s pixels, passes them through several layers and returns a probability for each number from 0 to 9. When it is wrong, training nudges the weights a little.
In practice
- It is not a brain: it does not understand like a person but learns correlations from data.
- Large networks need lots of data and computing power; small ones fit on a phone.
- The 3Blue1Brown channel has a very clear visual series on them; it is in videos and podcasts.