Model Training & Adaptation
Contrastive learning
AI learning useful differences by comparing examples treated as similar or different.
Example
AI learns to connect two differently cropped versions of the same picture.
Why people use it
It teaches AI which examples should be treated as alike or different.
What you'll hear
“These two views should count as the same object.”
What this means for you
Check that examples treated as similar really should be similar for the intended job.
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
- Must similar examples look identical?
- No. Different crops or lighting can show the same object and still be treated as related examples.
- Who decides which examples are similar?
- The training setup defines that relationship, sometimes using people and sometimes using automatically created pairs.
- Can a bad pairing teach the wrong lesson?
- Yes. If different things are treated as alike, useful distinctions can be lost.