Conversation

Notices

1. sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 18:35:32 UTC sazius on Mayfirst
   Some photos from my doctoral defence: https://media.saz.im/u/sazius/collection/doctoral-defence/
   • Temporary Marjolein (mk@oracle.skilledtests.com)'s status on Sunday, 14-Dec-2014 18:46:52 UTC Temporary Marjolein

     in reply to
     @sazius nice - what's the role of that dog-in-a-box (if any)?
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 18:49:14 UTC sazius on Mayfirst

     in reply to

     • Temporary Marjolein
     @mk It was just a random Flickr picture I used to illustrate what humans can understand in a picture and contrast with how hard it is for computers to do the same.
   • Temporary Marjolein (mk@oracle.skilledtests.com)'s status on Sunday, 14-Dec-2014 18:50:49 UTC Temporary Marjolein

     in reply to
     @sazius ah, so it did indeed play a role! :)
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 18:53:43 UTC sazius on Mayfirst

     in reply to

     • Temporary Marjolein
     @mk it has the added benefit that a human can even deduce things which are not visible in the picture, like that the dogs owner is pulling a rope :-)
   • Temporary Marjolein (mk@oracle.skilledtests.com)'s status on Sunday, 14-Dec-2014 19:39:49 UTC Temporary Marjolein

     in reply to
     @sazius not necessarily; the rope could be tied to a static object (tree) or to a floating object (boat) ;-) IOW, the human isn't necessarily even present at the scene (& given the angle it must be another person than the photographer IF the human is indeed pulling)
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 19:41:57 UTC sazius on Mayfirst

     in reply to

     • Temporary Marjolein
     @mk That's true, but at least we can make some probable guesses about things outside of the picture, which requires a lot of prior knowledge about the world (that the computer doesn't have, and that is really hard to encode and reason about). (Actually I've verified later that there is indeed a person pulling the rope :-)
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 19:51:34 UTC sazius on Mayfirst

     • Luke
     @luke You have to try to look smart when you are defending your thesis :-)
   • Temporary Marjolein (mk@oracle.skilledtests.com)'s status on Sunday, 14-Dec-2014 20:03:19 UTC Temporary Marjolein

     in reply to
     @sazius yes, but even very small children (babies) have an innate sense of "agency": things have causes, people do things. Which leads to the "wired for religion" hypothesis: if no human agent is visible or at least plausible, then there must be a "super human" agent. build a computer with a "wired" sense of "agency" and it might do better at deducing these beyond the edge contexts.
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 20:09:23 UTC sazius on Mayfirst

     in reply to

     • Temporary Marjolein
     @mk Sure, but my main point in the talk wasn't even that (that was just a bonus problem :-). My main point was even recognising that there is a dog in the image is very hard for a computer, let alone much harder issues like that.
   • Temporary Marjolein (mk@oracle.skilledtests.com)'s status on Sunday, 14-Dec-2014 20:12:42 UTC Temporary Marjolein

     in reply to
     @sazius Google seems to make some headway with computer-recognition of common photo subjects in pictures - although it sometimes makes hilarious mistakes
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 20:18:11 UTC sazius on Mayfirst

     in reply to

     • Temporary Marjolein
     @mk Yes, that was another of my slides :-) Anyway, that's basically the problem I worked with in my thesis.
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 20:28:01 UTC sazius on Mayfirst

     • lnxw48 (Linux Walt)
     • Temporary Marjolein
     @lnxw48 @mk Yeah, that's basically what is usually done. If you have several examples of the object you have an algorithm tries to learn the distinguishing features (low-level statistics of pixels etc). This is still far from human understanding of the contents of a picture.
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Sunday, 14-Dec-2014 20:32:06 UTC sazius on Mayfirst

     • lnxw48 (Linux Walt)
     • Temporary Marjolein
     @lnxw48 @mk Well, it might associate that word thanks to metadata on the web. But that doesn't really represent any deeper knowledge :-)
   • David Benfell (benfell@cybernude.org)'s status on Monday, 15-Dec-2014 04:12:21 UTC David Benfell

     in reply to
     By contrast, *my* defense will be over the phone. I could do it in my bathrobe.
     Claes Wallin (韋嘉誠) repeated this.
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Monday, 15-Dec-2014 10:20:51 UTC sazius on Mayfirst

     • lohang
     @lohang Thanks! :-)
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Monday, 15-Dec-2014 10:21:17 UTC sazius on Mayfirst

     • David Benfell
     @benfell :-)
   • Charles Roth MPC (encycl@vox.a2c3.co)'s status on Monday, 15-Dec-2014 11:39:03 UTC Charles Roth MPC

     in reply to

     @sazius Marvelous!

   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Monday, 15-Dec-2014 19:37:19 UTC sazius on Mayfirst

     • lnxw48 (Linux Walt)
     @lnxw48 I think you did, but thanks! I keep extracting new congratulations by posting each step :-)
     Claes Wallin (韋嘉誠) repeated this.
   • sazius on Mayfirst (saziusmayfirst@social.mayfirst.org)'s status on Monday, 15-Dec-2014 19:38:48 UTC sazius on Mayfirst

     • lnxw48 (Linux Walt)
     • morph
     @morph @lnxw48 There's actually one more step: on Wednesday I'll be officially awarded the degree by the dean :-) Not officially doctor yet.