Applied machine-learning for science
Outline
The course consists of six two hour lectures, followed by one hour to discuss the week's homework assignment. Followed by a final project on real world data.
Current bucket list of topics to cover ((~) denotes: short introduction to):
- General problem statement and introduction
- Ensembles of trees: forests and gradient boosting
- Neural networks: convolutions aren't convoluted
- Model selection and evaluation: predict future performance
- PCA and t-SNE: lower dimensional embeddings and visualisation
- Bayesian optimisation for hyper-parameter tuning (~)
- Meet a GAN: cops and robbers for neural networks (~)
- Probabilistic datastructures: a bonus lecture
Course projects
Take a look at possible course projects.
Technicalities, installing, running code
All the code will be written in python. We will make use of the scientific python stack:
- python v3.6
- numpy v1.12.1
- scikit-learn v0.18.1
- keras
- matplotlib v2.0.0
- jupyter v5.0.0
All work submitted for credit has to run with these dependencies only.
Instructions on installing on windows, mac and linux.
License
Heavily inspired by ESL, ISL, Introduction to machine-learning with python, and lecture notes by Gilles Louppe.
All original work is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.