The ML research community has long been driven by the need to publish, which results in a stark, sometimes ridiculous bias towards complexity. Remember to ask: "can we do this with k-means and logistic regression?"
Show this thread
-
Can we do 83% accuracy on CIFAR10 with k-means and logistic regression? Yes (Coates 2011).
Show this thread
I think we should have a "simple, strong baselines" workshop at major ML conferences. This would help focus ML research on real progress.
-
-
a paper is brewing on this topic, stay tuned.
-
What's your angle with the paper?
End of conversation
New conversation -
-
-
I would love a "simple, strong baselines" workshop. Even a
@Kaggle competition where the metrics highlight simplicity and/or training speed.@Kaggle scripts already encourage this (nudge@benhamner). Personal take on this:http://smerity.com/articles/2017/baselines_need_love.html … -
Reminder of the worst case that can happen when baselines are ignored or underoptimized -
@frankmcsherry will come along and demolish your 128 core distributed setup with his laptop and a single threaded program ;) http://www.frankmcsherry.org/graph/scalability/cost/2015/02/04/COST2.html …pic.twitter.com/0r9JSaIuPG
End of conversation
New conversation -
-
-
I think it's valuable to look at ML solutions as a Pareto front of complexity vs. performance. Allow end-users to choose trade-off between those two criteria.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Here is an example of such a study in NLP from 2013. It ended up being reasonably influential (400 citations), but I'm not aware of follow-up work that similarly evaluates baselines on other tasks (translation, Q&A) https://www.aclweb.org/anthology/P12-2018 …
-
I personally really like this one on Reading Comprehension: https://arxiv.org/pdf/1606.02858v2.pdf … Their linear, feature-based model beats most previous NN approaches.
- Show replies
New conversation -
-
-
all your base belong to us.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Totally agree. Academia what it is, you need to have something novel to publish. I tried „nice and small“ several times and only got „has been done before“ and „all parts are already known“ reviews...
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I think the distinction is whether your goal is to build or explore. Put some of these thoughts down here:http://blog.mikiobraun.de/2012/07/academia-industry-explore-build.html …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.