ML-Carp Recap and Outlook

What’s next?

Dr. Jens Brinkmann

The University of Auckland

September, 2023

Introduction to Machine Learning with Scikit Learn

• Nelis introduced us to the topic
• Mike built on that
• Regression and Classification
• Ensemble methods, Clustering, Dimension reduction

Introduction to Deep Learning

• Nidhi switched from generic ML to Deep Learning (DL)
• Some aspects (there are more:)
  • ML: is non-stoachstic; similar if not the same result
  • DL: non-convex and complex
  • DL: only estimated/non-deterministic results/every run can differ a bit
  • DL: of uses neural networks, but also DBN, VAEs, etc.
  • DL: tends to use way more resources (CPU/GPU)
  • ML: Can still solve a lot of real-world issues using limilted resources

Workflows for heavier computation

• Google Colab vs. locally installed Jupyter Notebooks or Plain Vanilla Python
  • Do all use Colab ‘in the real world’ all the time? No!
    • VSCode
    • Python locally installed
      • Virtual Environment
      • pip install
    • using GPU/Nectar/…

Recommended next steps

• This was a lot of content!
• Go through the examples in the near future / try to practice these techniques in your own research context
  • Set yourself a deadline
  • not: I will do that one day; that day never comes!
• Please fill out the Survey before leaving this session
• Join the Hacky Hour channel
  • Hacky Hour - Tuesdays 3-4pm on Zoom
  • Our instructors/helpers will be on hand for the next 4 Hacky Hour sessions to help you apply what we’ve covered this week.

Kaggle

Kaggle is an AI community aimed at getting a better understanding/practice. There are competitions that we can recommend:

• Regression:
  • House Prices 🏠
• Classification:
  • Titanic 🚢
  • Spaceship Titanic 🛰️ 🚢
  • mnist digit recognition 🖊️ (probably quite a bit harder just because of the data input/processing)

(Alternative to Kaggle: Scikit Toy Datasets)

Links to Teaching Material and Slides

• ML content
• DL content
• DL slide deck
• This Deck of Slides

Good luck

• Stay in touch with us on your ML/DL/AI journey
• Use it or lose it
• ML/DL tools are valuable to your research now and your research or other career path

ML Carp Recap and Outlook