Open chriskolb opened 1 year ago
The "Deep dive on GD and optimality" video just popped up on my youtube feed and I think the definition of L-smoothness contains a typo. Instead of
|| \nabla \fx - \nabla f(\tilde{\xv}) || \le L ||\fx - f(\tilde{\xv}) || \quad \text{ for all } \xv, \tilde{\xv}
I think it should be
|| \nabla \fx - \nabla f(\tilde{\xv}) || \le L ||\xv - \tilde{\xv} || \quad \text{ for all } \xv, \tilde{\xv}
https://github.com/slds-lmu/lecture_optimization/blob/67a1b17e01f0a2376061e7da9d678481d656be0f/slides/04-multivariate-first-order/slides-multivar-first-order-3-gd-deepdive-optimality.tex#L49
(source e.g. here https://www.stat.cmu.edu/~ryantibs/convexopt-F13/scribes/lec6.pdf)
The "Deep dive on GD and optimality" video just popped up on my youtube feed and I think the definition of L-smoothness contains a typo. Instead of
|| \nabla \fx - \nabla f(\tilde{\xv}) || \le L ||\fx - f(\tilde{\xv}) || \quad \text{ for all } \xv, \tilde{\xv}
I think it should be
|| \nabla \fx - \nabla f(\tilde{\xv}) || \le L ||\xv - \tilde{\xv} || \quad \text{ for all } \xv, \tilde{\xv}
https://github.com/slds-lmu/lecture_optimization/blob/67a1b17e01f0a2376061e7da9d678481d656be0f/slides/04-multivariate-first-order/slides-multivar-first-order-3-gd-deepdive-optimality.tex#L49
(source e.g. here https://www.stat.cmu.edu/~ryantibs/convexopt-F13/scribes/lec6.pdf)