Alex S

Hands-on, math-first class in machine and deep learning. We'll start at supervised learning, loss functions, gradient descent, and the chain rule, then work forward and backprop by hand for a tiny model. We'll compare and contrast layer types (linear, conv; quick tour of transfoerms/RNNs) and training heuristics (like initialization, normalization, regularization, early stopping, and SGD vs. Adam). Hands-on time: create your toy net using NumPy, recreate it in PyTorch and get it to learn on a very small real dataset. Short whiteboarding blocks, live coding, and frequent check-ins where you get to do it yourself will occur. Colab links will be given out. Python familiarity is preferred, but not required; the math is graspable and is performed so it will stick, though basic knowledge of calculus is required.