What do you "think" gradient descent will do?

Updates the parameters from backpropagation

What does the loss function evaluate for neural networks?

The current performance of the network

In backpropagation, how does the network know how to update its parameters?

By calculating gradients layer by layer

What is the main goal of backpropagation?

To find how to make the network's performance improve

What does the gradient vector tell us?

The steepest ascent direction of the loss function

Why is it not feasible to solve all derivatives for neural networks explicitly?

Neural networks have too many parameters

studbud

Ai Week 2

11 termsStudBud Administration8 plays

Imported from external platform

Flashcards11 cardsFPractice quizPHost liveL

Questions 11

Answers

1
What do you "think" gradient descent will do?
It will guarantee perfect predictions from the model
Calculates the exact global minimum of a function
Finds the closest local maximum by adjusting parameters incrementally
Updates the parameters from backpropagation
2
What mathematical concept is central to backpropagation?
Chain rule in calculus
Eigenvectors in Linear algebra
Conditional Probability in Probability theory
The Monster in Group Theory
3
What is the output of a loss function when predictions = labels?
-1
Infinity
1
0
4
What does the loss function evaluate for neural networks?
The current performance of the network
The stabilitiy of the network
The differnce between an output and an desired output of the network
The neuron activity of the network
5
In backpropagation, how does the network know how to update its parameters?
By solving all derivatives explicitly
To trial and error for lower loss values
By guessing new parameter values
By calculating gradients layer by layer
6
What is the main goal of backpropagation?
To find how to make the network's performance improve
Make the network's performance improve
7
In the student example, the desired output for the tricky student is:
1
0.5
0.87
0
8
Why do we square the differences in mean squared error?
To increase the loss value
To simplify derivative computation
To make the function non-linear
To prevent negative values
9
What is the gradient vector symbol in math?
Ω
Σ
∇
Δ
10
What does the gradient vector tell us?
The steepest descent direction of the loss function
The steepest ascent direction of the loss function
The maximum value of the loss function
The vector field of the loss function
11
Why is it not feasible to solve all derivatives for neural networks explicitly?
Neural networks have too many parameters
Activation functions prevent calculations
Derivatives don’t work with large numbers
Weights and biases are constant

Ai Week 2 Flashcards | StudBud