Lecture 9 - Optimization and Leaning for Robot Control - Intro to Numerical Optimization

Описание: Gradient descent, slow convergence with ill-condition functions. Optimality conditions for unconstrained optimization. Newton's method for root finding. Newton's method for optimization. The limitations of vanilla Newton's method: non-monotonic convergence, oscillations. Line search. Hessian regularization. Optimality conditions for equality-constrained optimization. Line search with merit function.

0:00 Gradient descent
7:15 Second order optimization
36:07 Limitations of Newton's method
46:46 Line search
58:19 Hessian regularization
1:13:00 Equality constraints