Ryan Drapeau: Battling Fraud with ML at Stripe
The Gradient: Perspectives on AI
00:00
The Importance of Retirable Networks in ML Engineering
We were looking at like training times for our model on the order of like 15 to 20 hours to train the production model and it was largely dominated by xg boost here. So, they became a really strong desire to speed up our iteration time by removing this component, the xg boost component in the landing on what we'll call like yeah pure DNN only type architecture. We ended up with a model that was quite a bit more performant than the wide and deep model that we were coming from but even more importantly, then the performance in my opinion, were kind of like two major pieces of impact.
Transcript
Play full episode
