Machine Learning

From Deep Learning Course Wiki
Jump to: navigation, search

The field/study that aims to give "computers the ability to learn without being explicitly programmed" - (Arthur Samuel, 1959)

We're going to ELI5 Machine Learning by comparing it to a human's learning process:

  • Pretend you're a student and you have a test next week.
    • Your teacher hands you a bunch of material you need to learn for the test. (training data)
    • You only use flash cards to study. (machine learning model)
    • A week has passed and now you take the test. (testing data)
    • Some time later, you graduate and use what you've learned for your new job. (generalizing on new data)

This process is comparable to an oversimplified Machine Learning application:

  • Pretend you're a cucumber farmer and you're trying to automate the process of labeling good and bad cucumbers (pretend you know the difference).
    • You have a collection of labeled cucumber images (training data)
    • You train a machine learning model to identify the good and bad cucumbers in your collection (machine learning model)
    • You test your machine learning model on a smaller collection of cucumber images to see how it performs (testing data)
    • If you think your machine learning model is ready, you start using it on your cucumber production line (generalizing on new data)
Comparing the two
Machine Learning Student Cucumber
training data teacher's material collection of labeled cucumber images
model flash cards arbitrary machine learning model
testing data teacher's test smaller collection of cucumber images
generalizing on new data using what you've learned at your new job using your model on the cucumber production line