Watching this video in the parking lot. Thank you and keep the good work. Short, precise, applied and to the point. This is how applied stat and econometrics should be taught.
Nobody I know can use a children's marker on paper or a white board with tons of stats on it but STILL make it so understandable and clear. A+ my friend.
I think a linear combination of metrics is not that suitable even though it is very common. We should use the function of constraints that suits the kind of optimization. Like satisfier kind of constraints should have exponential costs if not satisfied and optimizer constraints should be linearly combined. Have seen this to work much better in papers for pose estimation where they need to get elbows and knees right
Glad that you are back Ritvik 😊, isn’t this similar to lasso or ridge regression or in general regularised loss where the coefficient lambda determines the weight we give to regularisation vs minimising actual loss function and is dynamically tuned based on cv data
It feels so much easier to learn from you than from 500 pages books. I would also enjoy some more safety-oriented videos: adversarial robustness, interpretability, trojans, alignment, black swans...
@@jasdeepsinghgrover2470 These books look great, but it's a big time investment to read. I personally enjoyed the Center for AI safety's course, it's pretty thorough and includes exercises.
What, no math!!! I think you should do a follow up where you develop the equations. What is the relationship of dynamic multi-objective optimization to Lagrange multipliers or other convex optimization topics?
How do we perform multi-variate sorts? Suppose you are trying to choose a portfolio of assets. I want to sort on return and risk. How do I get top 10 high return and low risk assets. P.S. I am an economist and know the theoretical way, but I can seem to easily apply it in this scenario. 🇿🇼🇿🇦