Transcriptions
Note: this content has been automatically generated.
00:00:00
hello i'm about to start it caught from alan turing was we want
00:00:06
or want to that someone is a machine that can learn from experience
00:00:11
so now they are here and they choose this stunning performances and in
00:00:15
fact in only last five years their performance is very improve by ten percent
00:00:20
what what is the reason behind the recent fast progress of machine
00:00:24
learning outwards so let's summarise the training site cycle of machine learning algorithm
00:00:29
first together some data then we choose the model that fitting uh it fits better the data we train it
00:00:35
on that we train it on the data and then me test it okay so now we have lots of data
00:00:42
and indeed the size of the data that we have
00:00:45
is growing began exponential rate this large data enables us to
00:00:50
train more complex models for example teach learning to ignore networks
00:00:55
but on the other hand this is not for free and we need to spend more computational resources
00:01:01
now let's construe to example of classifying docks so at each iteration v. choose
00:01:07
one data point one dog can be off data model based on that the
00:01:11
next iteration be choose another data point and the optic them all but this
00:01:15
is not the best idea wise that uh because we might have similar data points
00:01:20
or after some training normal model already learned some part of the data
00:01:25
but it's quite sure about some other part of the data in my research we give develop i'll go it's
00:01:31
actually fast celebrates that on the fly chan i identified
00:01:35
these data points that the model is i'm sure about them
00:01:39
and use only them top data model so all the show
00:01:43
you to show both critically and experimentally that our mess out
00:01:46
there is the model be much less a comp competition resource thank you