MLConf 2014

Just spent a day at MLConf where I was talking about how we do music recommendations. There was a whole range of great speakers (actually almost 2/3 women which was pretty cool in itself).

Here are my slides:

Justin Basilico from Netflix talked about how they deliver a personalized start page using lots of ranking

Claudia Perlich also had a great presentation about ad targeting. A few things were really interesting: (a) using transfer learning to learn from web site visiting and apply it to add targeting (b) causality tests to figure out if a campaign actually made an impact (c) how they intentionally introduce bias in their classifiers by skewing subsampling. I think she also mentioned throwing in artificial negative data, which is something we do too.

Josh Wills had a hilarious slide about “ML as a tool” companies. His analogy was waiting two hours in a line for a roller coaster ride, and then someone jumps in front of you, rides the roller coaster, and gives you a video of the ride. His point was that ML is the 1% fun part, and 99% plumbing is the boring part. If you want to start a company, focus on the boring stuff no one else wants to deal with.