All of Our ML Ideas Are Bad (and We Should Feel Bad)

Wednesday, 2019, October 2 - 16:0016:45

Todd Underwood, Google

Abstract: 

The vast majority of proposed production engineering uses of Machine Learning (ML) will never work. They are structurally unsuited to their intended purposes. There are many key problem domains where SREs want to apply ML but most of them do not have the right characteristics to be feasible in the way that we hope. After addressing the most common proposed uses of ML for production engineering and explaining why they won't work, several options will be considered, including approaches to evaluating proposed applications of ML for feasibility. ML cannot solve most of the problems most people want it to, but it can solve some problems. Probably.

Todd Underwood, Google

Todd Underwood is an SRE Director for Google in Pittsburgh and leads Machine Learning for SRE for Google. He's been working on making ML work better at Google since 2009. It's still not done. He last presented at SREcon in 2015.

Open Access Media

USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.

BibTeX
@conference {239524,
author = {Todd Underwood},
title = {All of Our {ML} Ideas Are Bad (and We Should Feel Bad)},
year = {2019},
address = {Dublin},
publisher = {{USENIX} Association},
month = oct,
}

Presentation Video