COSC470

Home
Labs

COSC470: Special Topic - Machine Learning (2018)

Semester 2

Learning outcomes:

1. An understanding of, an ability to use, and an ability to implement classical machine learning techniques (Information Literacy, Critical Thinking)
2. An understanding of how AI agents play games and the most recent techniques for game playing (Information Literacy, Research, Critical Thinking, Lifelong Learning)
3. An understanding of the basics of machine learning theory and how it applies to the recent success of deep learning (Information Literacy, Research, Critical Thinking)
4. An ability to implement a deep learning model for a given task (Information Literacy, Critical Thinking, Research, Self-Motivation, Lifelong Learning)
5. An understanding of the fundamentals of image recognition (Information Literacy, Critical Thinking, Research)
6. An understanding of deep convolutional neural networks and an ability to use and implement such models (Information Literacy, Self-Motivation, Lifelong Learning)
7. An ability to conduct AI experiments and successfully communicate the results (Information Literacy, Communication, Research, Lifelong Learning)
8. An ability to find, read and understand recent research in the area (Information Literacy, Scholarship, Research, Lifelong Learning)
9. An appreciation for the ethical problems associated with AI (Information Literacy, Ethics, Lifelong Learning)

Lecture outline:

1. L1. Intro to classification and ML; decision trees
2. L2. AdaBoost
3. L3. SVMs
4. L4. Basics of Learning Theory (and a bit of deep vs. shallow)
5. L5. Deep Learning: classification/convnets/regularisation techniques etc.
   • Tensorflow implementation of the CIFAR-10 convnet.
6. L6. Faces (Viola Jones detection and Eigenface recognition)
• These notes are Jupyter notebooks -- here are some quick notes to get you started, and the zip files contain a notebook and some supporting images.
• Notebook on Viola-Jones face detction
• Viola & Jones Rapid Object Detection using a Boosted Cascade of Simple Features, Int. Conf. Computer Vision, 2001
• Viola & Jones Robust Real-Time Face Detection, Int. Jnl. Computer Vision, 57(2), pages 137-154, 2004
• Notebook on Principal Component Analysis
• Turk & Pentland Face Recognition using Eigenfaces, Int. Conf. Computer Vision 1991, pages 103-108
7. L7. Convnets and image-based object recognition +
• Notebook on EigenFaces and 500 images from CelebA [3.35MB]
• Notebook on Neural Networks in TensorFlow and the MIT labelled faces data set
• Notebook on Convolutional Neural Networks
8. L8. History of game playing, basic ideas, themes.
9. L9. Deep reinforcement learning
• Nice tutorial on RL with a neural network (doesn't use Tensorflow though)
10. L10. The alphago breakthrough (and alphazero)
11. L11. What next? [Are two player perfect information games just a solved case now? What are the more serious application cases?]
12. L12. Issues with Face Recognition
• Some controversial papers - note that these are linked to stimulate discussion, their presence here should not be interpreted as an endoresment or assessment of the work.
• Automated Inference on Criminality using Face Images X. Wu and Z. ZHang, unpublished, 2016
• Deep neural networks are more accurate than humans at detecting sexual orientation from facial images, Y. Wang and M. Kosinsi, unpublished, 2017
• Some commentry from Blaise Agüera y Arcas et al. on the criminality and sexual orientation work above
• Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification J. Buolamwini and T. Gebru, Proc. Machine Learning Research, 2018 - looking at bias in face recongition systems. Joy Buolamwini's TED Talk covers these issues as well
• Women Also Snowboard: Overcoming Bias in Captioning ModelsL. A. Hendricks, K. Burns, et al., ECCV, 2018 - an attempt to counter such biases
13. L13. Wrap-up, discussion of AI ethics.

Lecturers:

• Brendan McCane: L1-L3
• Michael Albert: L8, L10, L11
• Lech Szymanski: L4, L5, L9
• Steven Mills: L6, L7, L12
• Ali Knott and others: L13

Time:

Tuesday 9-11am