Transcriptions
Note: this content has been automatically generated.
00:00:01
hello my name is then you are only pages you in in in the lab um i do my research
00:00:06
machine learning and one of the problems i'm working on is how to search in a database when it's hard to
00:00:12
describe what you're looking for in details indeed most of us here would struggle to draw
00:00:17
a picture of a friend accurately enough that it can be used as a aquariums for search
00:00:22
however if i show you two pictures of your friend of someone else you would be delayed
00:00:26
some increase to this is to not only for pictures but for many other things like videos music
00:00:33
of um fluid and so on so people are good at making comparisons now suppose
00:00:38
that there is a database of some comparable objects anyone's find a particular object there
00:00:43
but without explicitly describe what it is just to be more concrete let's say
00:00:47
there is um database of movie actors and you want to find this particular
00:00:52
i'm late on the slide if you don't remember her name but you remember how she looks like
00:00:57
um than what we propose in our search algorithm is to review
00:01:02
example from this database and ask you to choose around the object from this sample which is the most
00:01:09
similar to your target so which one of these four people looks more like that lady on the previous slide
00:01:16
and then you give us your answer would even other sample and so on until we find your target uh so that's we
00:01:23
um our algorithm allows you to navigate through the database by only comparing objects from
00:01:29
this database to target but without explicitly describe it you target um our algorithm also
00:01:37
uh at lawrence its own feature presentations but again only
00:01:40
from the comparison data and without any feature extraction methods
00:01:44
and in fact if you look at the two d. um features
00:01:49
a stadium batting of four of the features that this out with learns
00:01:53
uh you'll see that people who looks alike they're actually being and um clustered together
00:01:59
so we build up this website was that um these uh um go there and visited a um
00:02:05
if you want to look up forks movie actors and or just check out our search algorithm you're welcome thank