Machine Learning

Module 4



Topic Checkpoints
Machine Learning & Image Classification (Introduction) [Lecture Video, Slides]  
Intro to Python [Lecture Video, Slides] Python Pocket Calculator
Nearest Neighbors [Lecture Video, Slides]  
Python Matrices & NumPy [Slides] Image Distance
Practice: Intro to Colab and NumPy
Optimization & Neural Networks [Lecture Video, Slides] Contribute to the HelloRob Handwriting Dataset!
[Instructions, HelloRob Dataset Checker Tool, Submission Form]

The HelloRob Handwriting Dataset

You may notice that the accuracies that we can achieve in Project 4 on the MNIST dataset are extremely high. Can we expect these results to transfer to the robot?

To see how well our learned models transfer to data outside of the MNIST dataset, we will collect our own dataset of handwriting samples and test on those. These will look a bit different from the MNIST images. Here is a sample of some images from Fall 2022 students:

HelloRob Digit Dataset

Instructions for Contributing

  1. Create an image of your handwriting sample for each digit from 0 to 9 and save them as individual images. You may use a tool like Paint or a similar drawing tool to create these images on a computer, tablet, or phone. The images must follow these rules:
    • Make the canvas a perfect square,
    • Use black ink on a white background,
    • Make sure the digit is well centered and takes up the full canvas,
    • Save the image as a JPEG file.
  2. Use the HelloRob Dataset Checker Tool to visualize your images in the small form factor they will be converted to and make sure they look good before submitting your images.
  3. Follow this link to submit your image samples. Please use your school email when filling out the form!

Please follow the instructions on the form carefully! If your images are not usable, they will have to be removed from the dataset.