Hello, here is a Multiple Choice Question and its solution
We add a regularization term because
this sometimes renders the minimization problem of the cost function into a strictly convex/concave
problem
this tends to avoid overfitting
this converts a regression problem into a classification problem
Solution: 1 and 2 are correct
I find the above 1. answer counter-intuitive as the purpose of regularization is purely to fight overfitting...
When have we seen that adding a regularization term renders the problem into a strictly convex/concave problem ?
I guess the explanation for this is the case of the logistic regression on linearly separable data, for which the weights will converge to infinity (which means the optimum is infinity, which means the function is not strictly convex (there are many infinities)) .
Regularization makes sure we converge to a finite solution, which somehow is unique ?
Mock Exam 2018 - Multiple Choice - Regularization
Hello, here is a Multiple Choice Question and its solution
We add a regularization term because
problem
Solution: 1 and 2 are correct
I find the above 1. answer counter-intuitive as the purpose of regularization is purely to fight overfitting...
When have we seen that adding a regularization term renders the problem into a strictly convex/concave problem ?
Thank you in advance for your help
Following up on my own question...
I guess the explanation for this is the case of the logistic regression on linearly separable data, for which the weights will converge to infinity (which means the optimum is infinity, which means the function is not strictly convex (there are many infinities)) .
Regularization makes sure we converge to a finite solution, which somehow is unique ?
Well confused over this...
2
Add comment