AI Foundations
ReLU
Rectified linear unit
A calculation that keeps positive numbers unchanged and turns negative numbers into zero.
Example
An AI stage changes a negative number to zero while leaving a positive number as it is.
Why people use it
It gives networks a simple rule that helps them learn more than straight-line relationships.
What you'll hear
“Use this simple rule inside the network.”
What this means for you
Treat this as one small part of the system, not a measure of its overall ability.
Can you control it?
Developer-only
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Can this calculation lose information?
- Yes. Different negative numbers all become zero, so their original differences are no longer visible afterward.
- What happens to a negative value?
- ReLU turns it into zero rather than passing the negative value onward.
- Can a unit become inactive?
- Yes. If it keeps receiving negative values that become zero, it can stop contributing useful learning.