From Research to Practice

Estimated time: approximately 6 hours

Machine learning research is most useful when you can translate ideas from the literature into concrete decisions in your own work.

In this activity, you will read an accessible machine learning paper, identify useful principles, and apply at least one of them to a machine learning project.

After completing the activity, you should be able to:

  • read a machine learning research paper more effectively;
  • identify ideas that are relevant to an applied problem;
  • translate a research principle into a concrete modeling decision;
  • evaluate whether the change actually improves your approach.

Read — approximately 1 hour

Read:

A Few Useful Things to Know About Machine Learning

Take notes as you read.

Rather than trying to remember every detail, identify principles that could influence how you develop or evaluate a machine learning system.

Apply

Choose at least one principle from the paper and apply it to a machine learning project.

Document:

  1. which principle you selected;
  2. why it is relevant to your project;
  3. what you changed;
  4. how you evaluated the effect;
  5. what you learned from the result.

If your project contains several trained models, ensembling is one useful idea to explore.

You could combine predictions from multiple models, compare the ensemble against your best individual model, and investigate whether—and why—the result changes.

The goal is not to prove that the paper’s recommendation always works.

The goal is to move from:

“I read an interesting idea.”

to:

“I tested the idea in my own context and collected evidence about whether it helped.”