A working knowledge of machine learning in 45 minutes

tl;dr I built a tiny thing and gave a tiny talk

I am currently working at the MIT Media Lab on a project studying news around the world. New graduate students in the Media Lab take a pass/fail course that tours them through the various labs, and my lab (Civic Media) covers how to build software.

I had the opportunity to give a short talk on

  1. What is machine learning?
  2. How does machine learning go right?
  3. How does machine learning go wrong?
  4. How can machine learning be incorporated into software projects?

This was an ambitious amount to cover in 45 minutes, but I approached it with a tour of scikit-learn for “traditional” machine learning, spacy for NLP, and keras for deep learning. I used a Jupyter notebook with working demos instead of slides or board to emphasize how easy open source software makes implementing “working” machine learning. You can find this notebook here.

I also opened and concluded with a tiny working demo of a natural language processing app, which doubled as instructions for running the notebook, and references for students to use afterwards. This site can be found here.