2. Aims
Aim 1 - Overview of Data science and AI
Artificial Intelligence / Machine Learning / Data Science
Aim 2 - Example of Data Science in health care
Aim 3 - Example of ‘state of the art’ machine learning
Aim 4 - How companies are using data science and AI
3. Who?
Director of Machine Learning @ Spencers Group
Lecturer in Intelligent Systems and Robotics @ De Montfort University
Teach Statistical Programming in R
C4Di Member
6 years experience of working in Machine Learning Projects
PhD in Computer Science/Machine Learning/Data Science (submit soon)
Clinical predictive modelling for patients with Chronic Heart Failure
Post-Graduate Certificate in Research
Statistical Modeling and Statistical Programming in R
MSc Intelligent Systems and Robotics
4. What is the difference between
AI and Machine Learning
AI
The perception
The presentation
The bits you see/interactive with
Term for Media / C-Suite
Machine Learning
The technical aspects
The ‘doing’ bits
The Maths!!!
Term for recruitment / technical teams
7. Who is
In Demand Data Scientist: The
Sexiest Job of the 21st
Century
Harvard Business Review (2012)
13 Top Tech Skills In
High Demand For 2018
Forbes (2017)
1. Experience With AI
3. Data Science Talent
9. Applied Machine Learning
11. Any Skills Related To Analytics
UK demand for AI
professionals has almost
tripled in three years
Computer Weekly (2018)
Million-dollar babies
The Economist (2016)
8. Data Scientist / Machine Learning Engineer
Skillset
Experience/Qualifications
MSc (in machine learning, AI or CS) with experience (+2 years)
PhD (in machine learning, AI or CS)
Languages/Technical
Python, R, Weka, Matlab
Scikit-learn, numpy, scipy, NLTK, Stanford CoreNLP, TensorFlow
C++, Java
ML Specific
Solid understanding of machine learning fundamentals
Strong opinions on a wide range of statistical approaches e.g. Logistic Regression, Decision
Trees, etc
10. Data Science
in action (health care)
‘what impact will <something> have on <something else>?
Real world example…
Philips Healthcare make a device call Motiva
It allows remote monitoring of patients
Question: what impact does it have?
Challenge
What has been done before?
Who has used it?
Can be compare with other people?
How do we measure ‘impact’?
11. Data Science
in action (health care)
Who?
150 people used it
Over 3,000 didn’t
How do we match and extract similar
patients?
Skills
Data Extraction (SQL)
Data Visualisation
Statistics
Analytical
Predictive
13. Data Science
in action (health care)
So what have we actually done?
A/B Testing (with a difference)
Does one <pathway/method> give better results?
How can we apply it?
Marketing strategy
Product Development
See ‘The Lean Startup’ by Eric Ries
Difference?
Findings – Broader and based on scientific methods
Our methods account for ‘chance’
14. Now we know it works can we
Identify High Risk Patients
(EARLY WORK, MORE TO COME AFTER PHD SUBMISSION)
Predict high risk patients using
Statistical Learning
Machine Learning
Results
How do you measure best?
Sensitivity vs Specificity
Impact of the decision
Caused an interesting debate
15. It is all about
The story
So now we know
HTM works in Hull and East Riding for patients with Heart Failure
We can identify patients who are at greatest risk of death or readmission
What we don’t know
Why?
Which elements?
Other outcomes
Peer reviewed papers
In-depth look at the algorithms used (new insights into their usage)
16. Examples of
Machine Learning
NEURAL NETWORK (ONLINE)
DEEP LEARNING (IMAGE CLASSIFICATION)
DEEP REINFORCEMENT LEARNING (SELF LEARNING AGENTS)
18. Example
DEEP LEARNING
Neural Networks +
More Layers
Combination of different types of layers
Example – Convolutional Neural Networks
Image classification
Object Detection
Loosely based on cat’s visual cortex [1]
[1] Deep Learning Tutorial - http://deeplearning.net/tutorial/deeplearning.pdf
Stamford. J, and Peach. B, (2016) “Scene Detection using Convolutional Neural Networks”, 2nd IET International
Conference on Technologies for Active and Assisted Living
Another Example (Python and TensorFlow):
https://www.youtube.com/watch?v=_zZe27JYi8Y
https://www.youtube.com/watch?v=mtu9_w8B984
19. Example
DEEP LEARNING
Stamford. J, and Peach. B, (2016) “Scene Detection using Convolutional Neural
Networks”, 2nd IET International Conference on Technologies for Active and
Assisted Living
20. Example
Deep Reinforcement Learning
Q-Learning
Q(s,a)
Learns – what is the best action (a) based on the current state (s)
States – convolution process
MSc Project - Playing Atari 2600 Games with Deep RL
Microsoft
Deepmind
Google acquires (£400 million)
Examples:
Deep Q Learning playing Atari (Link)
Unity (Github, ML Examples, Demo)
23. Next Steps
LEARN
Statistics (if DS)
Python
SciKit Learn
Kaggle
Udemy
DMU MSc Intelligent Systems
£6k (Link)
Distance Learning
Apply/Demo your knowledge
HIRE
Experience
Qualifications
Publications
Challenges
High demand / Low supply
How do you assess a candidate?
Expectations
R&D Tax Credits
SME: 230% (Link)
24. Recap
Aim 1 - Overview of Data science and AI
Artificial Intelligence / Machine Learning / Data Science
Aim 2 - Example of Data Science in health care
Aim 3 - Example of ‘state of the art’ machine learning
Aim 4 - How companies are using data science and AI
In medical diagnosis, test sensitivity is the ability of a test to correctly identify those with the disease (true positive rate), whereas test specificity is the ability of the test to correctly identify those without the disease (true negative rate).