Machine Learning

Course files and interactive tools.

Course CalendarStart here. Every class day, with links to that day's notebook and slides. Course FolderDatasets, the class-record notebooks, and the Block 1 slides. Classification WorldFit a classifier by hand. Train and test, and what overfitting looks like. Find the FabricationFour real names and one invented by a classmate's model. Can you tell which? Loss LandDrag the line through the data and watch where it puts you on the loss surface. Take gradient descent steps, see one point's loss on its own, and with three points see the whole dataset as one vector in ℝ³. SGD LabTrain a logistic classifier one step at a time and watch the walk through (w1, w2, b). Slice the cost with a plane, and see how batch size and learning rate change the path while λ changes the cost itself.

USAAIO PrepWednesday olympiad prep: the weekly slides. A1 KNN LeaderboardHow the class's k-NN classifiers scored on MNIST and CIFAR-10. A2 Test Set LeaderboardEvery A2 corpus, its entropy and bigram baselines, and every model ranked against them. A4 Weights BoardHow the class's linear classifiers scored on MNIST and CIFAR-10, with every model's weights as pictures. Quiz 4: One Batch StepQuiz 4's three-point batch, one gradient step at a time: each point's pull on the boundary, their average, and the probabilities in 3D.