1
00:00:07,200 --> 00:00:13,480
My research focuses on learning
from multiple sources of data for better prediction.

2
00:00:13,480 --> 00:00:21,680
We develop translational AI technologies for better
analysing multi-modal data in health care,

3
00:00:21,680 --> 00:00:24,680
including medical images, genomics data,

4
00:00:24,720 --> 00:00:28,400
chemical data and electronic health record data.

5
00:00:28,400 --> 00:00:31,320
From this data, we learn useful features

6
00:00:31,320 --> 00:00:34,720
and also analyse their complex relationships.

7
00:00:34,720 --> 00:00:38,160
This PyKale project aims to develop

8
00:00:38,160 --> 00:00:41,520
a software package for machine learning

9
00:00:41,520 --> 00:00:46,480
from multiple sources of data
for interdisciplinary research.

10
00:00:46,480 --> 00:00:50,120
We formulated green machine
learning principles

11
00:00:50,120 --> 00:00:55,200
to reduce repetitions, reuse
existing resources

12
00:00:55,200 --> 00:00:59,760
and recycle machine
learning models across disciplines.

13
00:00:59,760 --> 00:01:04,520
We designed a pipeline based API to standardise

14
00:01:04,520 --> 00:01:10,000
all machine learning workflows
into the same six steps.

15
00:01:10,000 --> 00:01:15,680
We take the FAIR approach
because we rely on open source software

16
00:01:15,680 --> 00:01:20,200
in both our teaching and research activities.

17
00:01:20,200 --> 00:01:23,960
Therefore, we want to become part of it.

18
00:01:23,960 --> 00:01:29,680
In September 2021, our PyKale software package

19
00:01:29,680 --> 00:01:32,280
was officially approved

20
00:01:32,280 --> 00:01:34,800
to become an official member

21
00:01:34,800 --> 00:01:38,400
of the world-leading PyTorch ecosystem.

22
00:01:38,400 --> 00:01:41,720
There are many barriers and challenges.

23
00:01:41,720 --> 00:01:45,360
The most challenging one is the time

24
00:01:45,360 --> 00:01:49,120
and efforts needed to make things happen.

25
00:01:49,120 --> 00:01:56,200
To overcome this,
we have to align this with our other tasks.

26
00:01:56,200 --> 00:01:59,720
What we did is to blend this

27
00:01:59,720 --> 00:02:04,680
with our other teaching
and research activities.

28
00:02:04,680 --> 00:02:09,360
In retrospect, I would like to have started

29
00:02:09,360 --> 00:02:14,880
some of the necessary tasks
much earlier, such as testing.

30
00:02:14,880 --> 00:02:20,400
It seems to be quite difficult
before we actually start doing it.

31
00:02:20,400 --> 00:02:23,640
But once we have done a few examples,

32
00:02:23,640 --> 00:02:26,840
it appeared to be relatively easy.

33
00:02:26,840 --> 00:02:30,280
We should spend less time worrying

34
00:02:30,280 --> 00:02:32,960
and more time coding.

35
00:02:32,960 --> 00:02:37,520
Having built this open source
software package,

36
00:02:37,520 --> 00:02:42,960
it becomes much easier
for us to share our research.

37
00:02:42,960 --> 00:02:46,760
For example, anybody with a Google account

38
00:02:46,760 --> 00:02:51,840
will be able to run our real world
machine learning example

39
00:02:51,840 --> 00:02:54,640
on the cloud in their browser

40
00:02:54,640 --> 00:02:57,360
without any local installation.

41
00:02:57,360 --> 00:03:01,320
This enables us to bring more students

42
00:03:01,320 --> 00:03:04,080
and researchers on board

43
00:03:04,080 --> 00:03:10,040
to either contribute to its development or make use of it.

44
00:03:10,040 --> 00:03:13,800
In the collaboration with AstraZeneca,

45
00:03:13,800 --> 00:03:19,320
we have agreed from the beginning
that all our development

46
00:03:19,320 --> 00:03:23,720
will be made open source under the PyKale framework.

47
00:03:23,720 --> 00:03:29,880
The FAIR principles will continue
to play a vital role in my research.

48
00:03:29,880 --> 00:03:34,720
I embed these principles
not only to my research,

49
00:03:34,720 --> 00:03:37,800
but also teaching and services.

50
00:03:37,800 --> 00:03:45,280
All my teaching materials are made open source 
and freely available on GitHub.

51
00:03:45,280 --> 00:03:50,880
In my role as the Turing Network
Development Award lead,

52
00:03:50,880 --> 00:03:55,120
We created a Sheffield AI website

53
00:03:55,120 --> 00:03:58,600
as an open source project

54
00:03:58,600 --> 00:04:03,240
where everyone can contribute

55
00:04:03,240 --> 00:04:07,280
and participate in its development.

56
00:04:07,280 --> 00:04:12,840
If you are a researcher
wishing to generate real world impact

57
00:04:12,840 --> 00:04:18,720
from your research, you should definitely
follow the FAIR principles.

58
00:04:18,720 --> 00:04:22,520
Making your data and software FAIR

59
00:04:22,520 --> 00:04:25,280
is the most effective

60
00:04:25,280 --> 00:04:29,200
and efficient way to generate impacts.

61
00:04:29,200 --> 00:04:33,360
If you are not sure about how to do it,

62
00:04:33,360 --> 00:04:40,520
the best way is to find good examples,
sharing similarities

63
00:04:40,520 --> 00:04:43,360
with what you want to do

64
00:04:43,360 --> 00:04:46,640
and to learn from these good examples

65
00:04:46,640 --> 00:04:51,720
to become one of them over time.