Experiments bench
The Lab
A few of my favorite concepts from my courses at emory and self study, with an attempt to make them intuitive to share with other people learning them!
Optimization
Gradient Descent
Drop a marble on a loss surface and watch it roll. Race SGD, momentum, and Adam.
Why the learning rate is the first knob you touch — and the first one that ruins a run.
Learning-Rate Schedules
Constant, step decay, cosine annealing, and warmup — side by side.
Big steps early, small steps late: why schedules beat any single learning rate.
Classic models
k-Means Clustering
Scatter your own points, pick k, and step through assign and update by hand.
How a good algorithm still lands in a bad answer when it starts in the wrong place.
SVM & the Kernel Trick
Drag points around a margin, then lift a tangle into a plane where a line works.
Why the widest street between classes generalizes — and how kernels make one exist.
Linear vs. Logistic
Place points and watch the fit chase them, residuals drawn in pencil.
The same machinery predicting a number versus predicting a probability.