4.2 Accelerated Gradient Descent

[@sandippaul100]

The proof is very intricate and this lecture is much hard than the previous ones 😢

[@ZUHRI_gaming14]

Yes, the proof is a little taugh. But if you look at the papers e.g Beck and Teboulle it will always be something like this. Prof Caramanis in my opinion, has done a great job explaining each steps of the derivation.

[@余淼-e8b]

27:59 maybe \lambda_t should be greater than or equal to 1 to avoid multiplying a negative number on both sides of the inequality of (a)

[@Stochine]

At 30:16, why were you able to replace x_t with x_{t+1}, \lambda_{t} with \lambda_{t+1}, and d_{t} with d_{t+1}?

[@余淼-e8b]

38:41 \lambda_t = \frac{1 \pm \sqrt{1 + 4\lambda_{t-1}^2} }{2}, maybe the constant term in the numerator should be 1

[@santhoshpeddi7562]

How convex combination is equals to that value ?