[PPT] - 1 https://trallard.github.io/Talks/RSE-shefeld The state of machine PowerPoint Presentation

SLIDE 1

1

SLIDE 2

The state of machine learning The state of machine learning

RSE seminar, University of Shefeld Tania Allard, PhD

https://trallard.github.io/Talks/RSE-shefeld

2 . 1

SLIDE 3

Tania Allard Tania Allard

Developer advocate Research Software Engineer Data expert  trallard  ixek

2 . 2

SLIDE 4

Machine learning Machine learning everywhere everywhere

 ixek

3

SLIDE 5

Machine learning Machine learning everywhere everywhere

So much that it is starting to not make sense anymore... like when you say a word 50 times in a row

 ixek

3

SLIDE 6

For good or for bad it is everywhere:

 ixek

4

SLIDE 7

For good or for bad it is everywhere:  Deployed in healthcare and warfare

 ixek

4

SLIDE 8

For good or for bad it is everywhere:   Deployed in healthcare and warfare In the creative industry (from music to books)

 ixek

4

SLIDE 9

For good or for bad it is everywhere:    Deployed in healthcare and warfare In the creative industry (from music to books) Reading CVs and judging your creditworthiness

 ixek

4

SLIDE 10

For good or for bad it is everywhere:     Deployed in healthcare and warfare In the creative industry (from music to books) Reading CVs and judging your creditworthiness Making us more Instagram worthy

 ixek

4

SLIDE 11

The big players:  Apple  Facebook  Google IBM Intel  Microsoft Nvidia Open AI  Twitter

 ixek

5

SLIDE 12

Machine learning generalised in two workflows Machine learning generalised in two workflows

Model development (R&D) Model serving (production for customers consumption)

 ixek

6

SLIDE 13

 ixek

7

SLIDE 14

What are these giants' issues? What are these giants' issues?

 ixek

8

SLIDE 15

What are these giants' issues? What are these giants' issues?

Mainly scale...in multiple areas

 ixek

8

SLIDE 16

If we have a small team we have a smaller number of issues... right?

 ixek

9

SLIDE 17

If we have a small team we have a smaller number of issues... right?  Small number of models to maintain

 ixek

9

SLIDE 18

If we have a small team we have a smaller number of issues... right?   Small number of models to maintain People have the knowledge in their heads

 ixek

9

SLIDE 19

If we have a small team we have a smaller number of issues... right?    Small number of models to maintain People have the knowledge in their heads They have their own methods to track progress

 ixek

9

SLIDE 20

That is the small team performance fallacy That is the small team performance fallacy

We still need processes and best practices in place... so let me get back at this later

 ixek

10

SLIDE 21

As the team As the team demand demand grows the problems grow grows the problems grow

    Increased complexity of data ow Larger number of workows Managing complexity of ows and scheduling becomes a nightmare Resource allocation has to be on point

 ixek

11

SLIDE 22

Serving models becomes harder Serving models becomes harder

 ixek

12

SLIDE 23

 ixek

SLIDE 24

13

SLIDE 25

How do they serve How do they serve millions of millions of

 ixek

SLIDE 26

customers across customers across the globe? the globe?

14

SLIDE 27

Three main players:    Infrastructure / resources Processes People

 ixek

15

SLIDE 28

 ixek

SLIDE 29

16

SLIDE 30

 ixek

17

SLIDE 31

Infrastructure as a code Infrastructure as a code

 ixek

18

SLIDE 32

 ixek

19

SLIDE 33

Everything as a code Everything as a code

Version control Less ambiguity on the congurations Shorter turnarounds Deterministic environments

 ixek

20

SLIDE 34

Processes Processes

 ixek

21

SLIDE 35

 ixek

SLIDE 36

22

SLIDE 37

Data and code as first class citizens Data and code as first class citizens

 ixek

SLIDE 38

23

SLIDE 39

 ixek

SLIDE 40

24

SLIDE 41

People People

Data scientist Data engineer ML Engineer

 ixek

25

SLIDE 42

What does academia have to What does academia have to

ffer?
ffer?

 Much more than you think

 ixek

26

SLIDE 43

People People

Researchers Research software engineers Librarians

 ixek

27

SLIDE 44

Resources and Infrastructure Resources and Infrastructure

We still need to gure this out... it is pretty much an ad-hoc case

 ixek

28

SLIDE 45

Processes Processes

Scientic rigour Peer review Data management

 ixek

29

SLIDE 46

Which areas could benefit from academic Which areas could benefit from academic collaborations? collaborations?

 ixek

30

SLIDE 47

Meta-learning Meta-learning Humans learn across tasks (learn from experience)

 ixek

SLIDE 48

31

SLIDE 49

If prior tasks are similar then we can carry prior knowledge

 ixek

32

SLIDE 50

AlphaGo uses some sort of meta-learning

 ixek

33

SLIDE 51

Algorithmic fairness Algorithmic fairness

It has become increasingly important to ensure that models are making justied calls that are free from unintended bias.

 ixek

34

SLIDE 52

Algorithmic fairness Algorithmic fairness

It has become increasingly important to ensure that models are making justied calls that are free from unintended bias. The one way to make progress is through interdisciplinary collaboration

 ixek

34

SLIDE 53

Towards model explainability Towards model explainability

Address the trade-off between performance and interpretability

 ixek

35

SLIDE 54

Reinforcement learning deadly triad Reinforcement learning deadly triad

Following nature's paradigms RL agents receive awards and then learn to maximise success by performing optimal actions.

 ixek

36

SLIDE 55

How to keep an algorithm learning if there are far too many potential variables or outcomes to be evaluated without being fed ridiculous amounts of data.

 ixek

37

SLIDE 56

In brief In brief

Focus on the 3 pillars:    People Infrastructure Processes

 ixek

38

SLIDE 57

Thank you Thank you

 ixek  tania.allard@microsoft.com

39