Autograd and backpropagationRapid Fire
Autograd & Backpropagation Fundamentals
Practise explaining how gradients move through a computation graph and what can break a backward pass.
Unlock with Full AccessThe round
This round covers reverse-mode differentiation, matrix gradients and gradient accumulation, alongside practical PyTorch behaviour. Some questions ask for an equation or tensor shape. Others ask you to explain what a change to the computation does to gradient flow.
What you'll practise
Gradient mechanics
Reason through matrix operations and contributions from different graph branches.
PyTorch behaviour
Explain the effects of in-place operations, detaching tensors and clearing gradients.
Checking derivatives
Use numerical gradient checks and account for their limitations.
Autograd & backpropagation
Answer seven questions
Get feedback on each answer when you finish the round.
All questions and feedback are included with Full Access.