SysML 18: Michael Jordan, Perspectives and Challenges

[@WilliamWardlaw]

This is an excellent and under-represented perspective on the field.

[@skeselj]

I'm shocked this has less than 20k views. These are such valuable ideas. I think in the coming years more people will embrace them.

[@philipghuPride]

sound needs to be fixed...no right channel

[@marcusklaas4088]

Yes. Very annoying.

[@bhi5hmaraj]

Even I thought so , but if you adjust the headphone jack I was able to get it on both the channels

[@Jarrar100]

use 7.1

[@Sumta555]

Let's fix the basic sound problem first and then we have many miles to go building a perfect AI machine. :)

[@jordanott365]

Any references for what he mentions at 10:30 regarding imagenet?

[@jpratt8676]

Jordan Ott would love to see more information on that. I struggle to see how imagenet could be doing that poorly.

[@Rizhiy13]

Here is a similar paper on CIFAR-10: arxiv.org/abs/1806.00451. If your error doubles when you just change the dataset a little bit, but leave all constraints the same, it is very plausible that it will increase significanlty when you remove all constraints on input.

[@TheHyperion11]

37:05 Every speakers nightmare questioner right here.

[@antonisminaskrasakis9164]

you do realize thats blockchain, right? heh

[@greyreynyn]

it annoys me too MJ, it annoys me too :(

[@benanderson6108]

lol, AI fanboys btfo, nothing is happening there until you get the underlying philosophy of mind right and that's not just a bunch of numbers, you have to actually think about things more involved than just a number line lmao