Jason Antic

1/ I had a question about successfully handling high traffic on computationally expensive deep learning models (like DeOldify). I'm pretty new to this myself and therefore may be very wrong but I do have a take and I'm wondering what others think. Here's what I said:

2/ "As far as scaling goes- this is something that we actually wound up “outsourcing” for DeOldify. In that we’re just licensing the model and letting the licensing companies (like MyHeritage) figure out the hard part of scaling. "

3/ "Because we knew that would be hard and quite frankly not something we’re comfortable with taking on as a two person team. Before we decided to go down this route, we were originally going to go with an app that strictly ran on iPhone hardware (6s+)."

4/ "Yes,this actually worked, and quite well (noticeably better than open source). As you can imagine- getting rid of servers like this solves the scaling problem :) We’ve also considered doing a desktop app (haven’t ruled it out, just not right now). "

5/ "Again- deferring huge, expensive computation to the users, hence avoiding a lot of complication on our end. And we’d argue it’s better for the users."

6/ "My hot take on “the cloud” approaches is that they tend to be the default approach to a fault and traditional desktop/local deployments tend to get overlooked even if they make the most sense."

Very thought-provoking. You're saying take advantage of designing with constraints in mind from the beginning? (make what you're making work on a 6s) Instead of defaulting to cloud's unlimited offerings?