A practical guide to machine learning

If you ever wanted to learn about Machine Learning but are sick of seeing presentations full of formulas and theory that never tell you how to actually use it, this mini course is for you. I will show you how to use Machine Learning without much fuzz and with practical advice how to get the most out of it.

Transcript

1. A practical guide to Machine Learning Christian Diener October 13th, 2016
2. Intro Learning by example Honorable mentions  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
3. Intro Learning by example Honorable mentions https://github.com/cdiener/xtraining  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
4. Intro Learning by example Honorable mentions What this talk is or is not... We will... not look at details not have to program anything not look at all possible machine learning algorithms actually use three industrial strength methods learn how to get decent performance have fun  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
5. Intro Learning by example Honorable mentions variables / features response samples prediction training  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
6. Intro Learning by example Honorable mentions genes (expression) panel cancer biopsies prediction training lung squamos cell carcinoma  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
7. Intro Learning by example Honorable mentions Urban dictionary of machine learning concepts overfitting Occurs when the model fits well to the training data but does not generalize to unseen data (we learn noise in the training data). hyperparameter Parameters of the machine learning method we have to choose by hand and that affect the outcome. regularization Using information not contained in the data set to improve the model.  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
8. Intro Learning by example Honorable mentions log P(ci) = βi · x − log Z If we have more features than samples we need regularization. Either L1 ( β ) or L2 (β2). Hyperparameters: regularization type (L1 or L2), regularization strength λ  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
9. Intro Learning by example Honorable mentions Decision trees all data class histogram x 1 > 11  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
10. Intro Learning by example Honorable mentions Random Forests vote Hyperparameters: Number of trees, Maximum depth of trees, Number of tested variables at each cut  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
11. Intro Learning by example Honorable mentions Deep Learning positive weight negative weight input layer output layer  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
12. Intro Learning by example Honorable mentions Deep Learning positive weight negative weight hidden layers in out ReLu tanh activation function Hyperparameters: soooooo many  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
13. Intro Learning by example Honorable mentions http://playground.tensorflow.org/  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
14. Intro Learning by example Honorable mentions Why Deep Learning is worth Billions  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
15. Intro Learning by example Honorable mentions And why its worth our attention... Predicting the sequence specificities of DNA- and RNA-binding proteins by deep learning. Alipanahi B, Delong A, Weirauch MT, Frey BJ. Nat Biotechnol. 2015 Aug;33(8):831-8. doi: 10.1038/nbt.3300.  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener
16. Intro Learning by example Honorable mentions Honorable mentions Gradient Boosting Machines Recurrent Neural Networks Reinforcement Learning Thank you!  CC BY-SA 4.0  cdiener.com  @thaasophobia  cdiener