Derive loss landscapes, backpropagation Jacobians, manifold hypotheses, and transformer attention mechanisms from scratch using rigorous multivariable calculus and linear algebra.
Enroll in Lecture SeriesVector spaces, orthogonal projections, spectral decompositions, and Principal Component Analysis as low-rank approximation.
Hessian matrices, gradient descent dynamics, saddle points, Adam optimizer momentum physics, and learning rate schedules.
Kullback-Leibler divergence, cross-entropy bounds, rotary positional embeddings (RoPE), and softmax temperature scaling.