How Do You Train the Negatives for the Training Data?

We use the positives, but we also use the negatives to see, oh, maybe this item didn't belong in the first or second row. And again, we're doing things with like this human in a loop. We don't rely on just what the seller describes it as. We can actually use what we think the image is. Another interesting way that were useing this is home decore and trying to identify style. Our current canidate said, is reaching above 70 % precision to identify these types of styles. So all of this has enabled just better and better candit selection, heavy use of machine learning and this persilization in real time.

Play episode from 20:34

Transcript

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!

Get the app