Benevolent machine learning and ai
Download as PPTX, PDF0 likes422 views
This document discusses machine learning and artificial intelligence. It begins with headlines about jobs being replaced by automation but also new jobs being created that require new skills. The document then discusses the speaker's personal journey working with machine learning techniques like neural networks, evolutionary algorithms, and deep learning. It provides examples of machine learning applications and overview of basic machine learning concepts like the biological neuron, McCulloch-Pitts model, and multi-layer perceptrons. The speaker concludes that machine learning will be a companion to humans and that human factors are important when developing these systems.
![IS THISTHE
FUTURE?
Qniksefat [CC BY-SA 4.0
(https://creativecommons.org/licens
es/by-sa/4.0)], from Wikimedia
Commons
2](https://image.slidesharecdn.com/benevolentmachinelearningandai-200616082938/85/Benevolent-machine-learning-and-ai-2-320.jpg)
![ISTHISTHEFUTURE?
• stephen bowler from wakefield, united
kingdom [CC BY 2.0
(https://creativecommons.org/licenses/by/2.0)
], via Wikimedia Commons
3](https://image.slidesharecdn.com/benevolentmachinelearningandai-200616082938/85/Benevolent-machine-learning-and-ai-3-320.jpg)
Benevolent machine learning and ai
- 1. DrScottTurner Computing UniversityofNorthampton BENEVOLENTAI AND MACHINE LEARNING Are they coming for you?
- 2. IS THISTHE FUTURE? Qniksefat [CC BY-SA 4.0 (https://creativecommons.org/licens es/by-sa/4.0)], from Wikimedia Commons 2
- 3. ISTHISTHEFUTURE? • stephen bowler from wakefield, united kingdom [CC BY 2.0 (https://creativecommons.org/licenses/by/2.0) ], via Wikimedia Commons 3
- 4. • 800 Million Global Workers will be replaced by robotic automation • McKinsey Global Institute https://www.mckinsey.com/feature d-insights/future-of-work/jobs- lost-jobs-gained-what-the-future- of-work-will-mean-for-jobs-skills- and-wages • 14% of global workforce will need to transistion to new occupational categories and learn new skills 4 HEADLINES https://www.bbc.co.uk/news/world-us-canada-42170100
- 5. • WEF: 75 million jobs globally by 2022 but a 133 million new ones created. • More data analysts, software developers, social media specialist, teachers, customer service workers. • ‘Vastly improve’ productivity of existing jobs by better algorithms. 5 HEADLINES https://www.bbc.co.uk/news/business-45545228
- 7. SOBENEVOLENTMACHINELEARNING? A phrase I adapted from Tegmark (2018) description of Benevolent AI “… a good outcome not as guaranteed, but something that needs to ensured by hard work in the form of AI-safety research”
- 8. MACHINE LEARNING IS ALREADYAROUND US. •• A few Examples • Neural Networks in fraud detection • Clustering algorithms for data analysis • Some ANPR systems • SIRI came out of DARPA research
- 9. MYOWN PERSONAL JOURNEY • Producing an noninvasive glucose monitor • Selecting filters for Neurophysiological Signals • Detection of Fraudulent Counting Evolutionary Algorithms • Simple neural approach to model an insect eye – movement detection • Leather classification • Current: Deep Learning in Detecting Facial Emotions Neural Network
- 12. Children 11110000 00001111 Parents (crossover point at half way along sequence) 0000 0000 1111 1111
- 13. Neural Network
- 14. BIOLOGICAL NEURON Taken from http://hepunx.rl.ac.uk/~candreop/minos/NeuralNe ts/neuralNetIntro.html
- 15. MCCULLOCH-PITTS MODEL X1 X2 X3 W1 W2 W3 T Y Y=1 if W1X1+W2X2+W3X3 T Y=0 if W1X1+W2X2+W3X3<T
- 16. INTRODUCETHEBIAS • Y=1 • if W1X1+W2X2+W3X3 +W0X0 0 • Y=0 • if W1X1+W2X2+W3X3 +W0X0 <0
- 17. LOGIC FUNCTIONS - OR X1 X2 1 1 Y Y = X1 OR X2 X0 -1
- 18. LOGIC FUNCTIONS -AND X1 X2 1 1 Y Y = X1 AND X2 X0 -2
- 19. LEARNING AND GENERALISING • Learning is achieved • by adjusting the weights • Generalisation is achieved • because similar patterns will produce an output
- 21. MULTI-LAYEREDPERCEPTRON • Feedback network • Train by passing error backwards • Input-hidden-output layers • Was the most common
- 22. MULTI-LAYERED PERCEPTRON (TAKEN FROM PICTON 2004) Input layer Hidden layer Output layer
- 23. MULTI-LAYEREDPERCEPTRON Feedback network Train by passing error backwards Input-hidden- output layers Most common
- 24. Deep Learning
- 27. Current Projects
- 29. USEFUL SOURCES • An introduction to Neural Network - Nielson MA (2015) “Neural Networks and Deep Learning” Determination Press http://neuralnetworksanddeeplearning.com/chap1.html • http://aiinschools.com/resources/
- 30. Round Up
- 31. PLAYTIME • You can view videos about build the neurones in a spreadsheet at: • A Network in a spreadsheet https://www.youtube.com/watch?v=yieX99Y_Phg • Training a neurone - https://www.youtube.com/watch?v=3993kRqejH
- 32. CONCLUSIONS • We need to take a balanced view. • Price for taking part is dropping dramatically at the moment • Machine Learning in it various guises has a lot to offer. • Conclusion after the discussion • – I think machine learning is going to a companion • - Human factors are important.
- 33. . 33 Picton P(2004) Neural Networks https://amzn.to/2Er6iUZ TEGMARK, M. (2017). Life 3.0: being human in the age of artificial intelligence. https://amzn.to/2GGiu6j Turner S (2015) Artificial Neural Network – Single Neurone produced with Excel https://youtu.be/CZwKUyNePvg One of my favourite books in this area at the moment Fry H (2018) Hello World: How to be Human in the Age of the Machine https://amzn.to/2VnnxvN