← Quiz: Neural Networks from Scratch: all 10 questions
Quiz: Neural Networks from Scratch · Question 8 of 10
Why should you not start all weights at zero?
Why should you not start all weights at zero?
- A.Zero weights make the loss infinite
- B.All hidden neurons would compute the same value and get the same update, so they never learn different features
- C.NumPy cannot multiply by zero
- D.The bias would become negative
Answer & explanation
Correct answer: All hidden neurons would compute the same value and get the same update, so they never learn different features
Small random starting values break this symmetry.
Can you get the rest right without peeking?
Take the timed Quiz: Neural Networks from Scratch quiz free. Pass it to earn XP and count towards your certificate.
Start the quiz →