Optimization (ACM 40990 and ACM 41030)

Description: For Optimization in Machine Learning (ACM 40990, Spring 2025), refer to the Brightspace page. For Optimization Algorithms (ACM 41030, Spring 2025), refer to this page.

 

 

 

Lecture Notes:

Course Documents:

Examinable results - Part 2:

  • Week 8: The projection operator, Section 12.4
  • Week 8: First-order optimality conditions in case of a single inequality constraint, Section 13.2
  • Week 9: Nothing!
  • Week 10: Showing that if the LICQs are satisfied, then the Lagrange Multipliers are unique - Section 15.1, Theorem 15.1
  • Week 10: Showing that the tangant cone is inside the set of LFDDs; showing that the tangant cone and the LFDDs are the same when the LICQs are satisfied - Section 15.2, Theorem 15.3
  • Week 11: Assuming Farkas's Lemma, prove the necessary conditions (KKT conditions) for a feasible point x_{star} to be a minimizer - Section 16.4
  • End of list!



Exercises #1: Line-search methods.

Code repository:



Exercises #2: Newton iteration and the Strong Wolfe Conditions.

Code repository:



Exercises #3: BFGS revisited and the Trust-Region Method.



Exercises #4: Global Optimization and Simulated Annealing

Code repository:



Exercises #5: Constraints (ACM 41030 only)



Exercises #6: More Constraints (ACM 41030 only)



Foundations of Data Science: Materials for Short Course on 8/12/2023: