I add junk data on purpose when training neural networks

Most folks clean their data spotless, but I mix in some random noise. It helps the model handle real-world messes better.

3 comments

Log in to join the discussion

Log In

3 Comments

wesley8087mo ago

So you're basically training your models on a digital hoarder's garage? That's one way to make them tough I guess. Just hope they learn to ignore the junk and not start collecting it. Next you'll be feeding them blurry pictures on purpose to fix their vision. Where does it end?

4

phoenix_burns266mo ago

Exactly! Gotta build that mental filter.

5

anna_hall7mo ago

Honestly that "real-world messes" part is key. Tbh a lot of real data has hidden patterns in the noise, not just pure randomness. Adding structured junk might teach the model to ignore actual useful signals.

1