Question 1
Why is Xavier initialization preferred in neural networks?
Question 2
Why are CNNs considered more powerful than traditional neural networks for image processing?
Question 3
What problem does batch normalization primarily address?
Question 4
Why is learning rate decay applied during training?
Question 5
What is a key characteristic of CNN architectures inspired by biological vision systems?
Question 6
Which model update approach is considered a standard/best default choice when uncertain?
Question 7
How does inverted dropout differ from traditional dropout?
Question 8
In which scenario is model ensembling particularly beneficial?
Question 9
What advantage does Nesterov Momentum have over standard momentum?
Question 10
What is the primary objective of using activation functions in a neural network?
Question 11
What is the purpose of dropout in a neural network?
Question 12
What role do historical gradients play in the AdaGrad optimization technique?
Question 13
Which parameter update method combines the elements of both momentum and RMSProp?
Question 14
What is a significant downside of using AdaGrad?
Question 15
What are the implications of using second-order methods in training neural networks?