Intro to Machine Learning - for Deep Learning talk at PyCon UK 2016

This is a very simple talk introducing some very basic machine learning ideas for my deep learning workshop at PyCon UK 2016.

Transcript

1. A very quick introduction to Machine Learning PyCon UK 2016 G. French University of East Anglia Image montages from http://www.image-net.org
2. A quick machine learning review
3. NOTE: This presentation is a mass-over- simplification of the topic It may also not be objectively very good
4. NOTE: There are other better sources out there Looking at the Scikit-Learn website at http://scikit-learn.org/ is IMHO a good start
5. Supervised learning
6. Given: Set of sample inputs X, e.g. images Corresponding outputs Y, e.g. classifications (dog/cat/banana/etc.)
7. Learn rules to assign output ′ to novel input ′
8. is the set of ground truths Often manually prepared (can be time consuming / expensive)
9. Unsupervised learning
10. Learn structure inherent in Not the focus of this talk
11. How does it work?
12. Each sample in is represented in vector form (a sequence of numbers)
13. Simple example Fisher’s Iris dataset – measurements of iris flowers
14. Fisher’s iris data set Each sample: 4 features: sepal length; sepal width; petal length; petal width classification is species: setosa; versicolor; virginica
15. Fisher’s iris data set Plot ‘sepal length’ and ‘petal width’ Represent classification (species: setosa, versicolor, virginica) as colour
16. 2D example Fisher’s iris dataset
17. Given a novel sample with known measurements but unknown species How would we predict the species?
18. We could divide our space into regions; one per class
19. Nearest neighbour classifier decision surface Fisher’s iris dataset
20. Linear SVM classifier decision surface Fisher’s iris dataset
21. Random Forest classifier decision surface Fisher’s iris dataset
22. Many more kinds of classifier Classifiers in Scikit-learn For more info, head over to: http://scikit-learn.org/