​Question 1: Complete the first three loss functions in the

Basic version. Do you still think larger step sizes are always
better? Or smaller ones? What do you think is the optimal
step size?

For the basic version, I do not think that the larger step sizes are always better. On the other hand the smaller ones are also not very beneficial, as they take way to long to reach the final destination. The most optimal one out of my test cases was usually the middle one as it found it the quickest without getting stuck.

Question 2: Complete the last two loss functions in the
Advanced version. Do you see how adding more starting
points boosts the chance of getting to the global minimum?

​

In the two images, we can observe how more starting points enable us to find the lowest point on a complex surface like the demon hill. The orange dots can easily find a fake/false valley and get stuck. If you have just one dot, it can get stuck in a small valley and never find the lowest point. But when you use more dots, like five, they spread out and explore more locations. This gives you a greater chance that at least one of them will find the absolute lowest spot. So, more starting points make it easier to find the best solution even when the path is confusing. The starting location of these dot's are also very beneficial and important to whether these dot's get to the correct location .