WEBVTT
Kind: captions
Language: en

00:00:00.000 --> 00:00:04.319
our ability to learn and get better at

00:00:04.319 --> 00:00:06.299
tasks through experience is part of

00:00:06.299 --> 00:00:07.319
being human

00:00:07.319 --> 00:00:09.330
when we are born we know almost nothing

00:00:09.330 --> 00:00:11.219
and can do almost nothing for ourselves

00:00:11.219 --> 00:00:13.320
but soon we're learning and becoming

00:00:13.320 --> 00:00:15.990
more capable every day but did you know

00:00:15.990 --> 00:00:18.990
that computers can do the same. Machine learning

00:00:18.990 --> 00:00:20.640
brings together statistics and

00:00:20.640 --> 00:00:22.680
computer science to enable computers to

00:00:22.680 --> 00:00:24.449
learn how to do a given task without

00:00:24.449 --> 00:00:27.180
being programmed to do so. Just as your

00:00:27.180 --> 00:00:29.640
brain uses experience to improve a task

00:00:29.640 --> 00:00:33.030
so can computers. Say you need a computer

00:00:33.030 --> 00:00:34.590
that can tell the difference between a

00:00:34.590 --> 00:00:36.360
picture of a dog and a picture of a cat

00:00:36.360 --> 00:00:39.300
you could begin by feeding it images and

00:00:39.300 --> 00:00:39.989
telling it

00:00:39.989 --> 00:00:42.540
this one's a dog, that one's a cat.

00:00:42.540 --> 00:00:44.460
A computer programmed to learn will seek

00:00:44.460 --> 00:00:46.320
statistical patterns within the data

00:00:46.320 --> 00:00:48.210
that will enable it to recognize a cat

00:00:48.210 --> 00:00:51.030
or dog in the future it might figure out

00:00:51.030 --> 00:00:53.190
on its own that cats have shorter noses

00:00:53.190 --> 00:00:55.199
and dogs come in a larger variety of

00:00:55.199 --> 00:00:57.420
sizes and then represent that

00:00:57.420 --> 00:01:00.210
information numerically organizing in space

00:01:00.210 --> 00:01:03.120
but crucially it's the computer not the

00:01:03.120 --> 00:01:05.010
programmer that identifies those patterns

00:01:05.010 --> 00:01:06.960
and establishes the algorithm by which

00:01:06.960 --> 00:01:09.360
future data will be sorted. One example

00:01:09.360 --> 00:01:10.619
of a simple yet highly effective

00:01:10.619 --> 00:01:12.869
algorithm is to find the optimal line

00:01:12.869 --> 00:01:15.000
separating cats from dogs. When the

00:01:15.000 --> 00:01:16.890
computer sees a new picture it checks

00:01:16.890 --> 00:01:18.420
which side of the line it falls on and

00:01:18.420 --> 00:01:22.500
then says either cat or dog. But of

00:01:22.500 --> 00:01:26.549
course there can be mistakes. The more

00:01:26.549 --> 00:01:28.049
data the computer receives the

00:01:28.049 --> 00:01:29.549
more finely tunes its algorothm becomes

00:01:29.549 --> 00:01:31.470
and the more accurate it can be

00:01:31.470 --> 00:01:34.979
in its predictions. Machine learning is

00:01:34.979 --> 00:01:37.650
already widely applied. It's the technology

00:01:37.650 --> 00:01:39.840
behind facial recognition, text to speech

00:01:39.840 --> 00:01:43.259
recognition, spam filters on your inbox,

00:01:43.259 --> 00:01:45.149
online shopping or viewing

00:01:45.149 --> 00:01:48.270
recommendations, credit card fraud

00:01:48.270 --> 00:01:50.790
detection and so much more. At the

00:01:50.790 --> 00:01:52.619
University of Oxford machine learning

00:01:52.619 --> 00:01:54.899
researchers are combining statistics and

00:01:54.899 --> 00:01:56.369
computer science to build algorithms

00:01:56.369 --> 00:01:58.710
that can solve more complex problems,

00:01:58.710 --> 00:02:00.960
more efficiently, using less computing

00:02:00.960 --> 00:02:01.630
power

00:02:01.630 --> 00:02:03.789
From medical diagnosis to social media

00:02:03.789 --> 00:02:06.820
the potential of machine learning to

00:02:06.820 --> 00:02:08.560
transform our world is truly

00:02:08.560 --> 00:02:13.390
mind-blowing. To find out more about machine

00:02:13.390 --> 00:02:15.819
learning visit www.OxfordSparks.ox.ac.uk

00:02:15.819 --> 00:02:17.890
follow us on Twitter and find us on

00:02:17.890 --> 00:02:23.340
Facebook