Preporato

Neural networks and backprop

Layers, activations and backpropagation, written out by hand.

Start 6 problems

Problems

  1. 1Forward pass of a two-layer networkEasy · 10 min · 10 hidden testsSolution free
  2. 2Softmax that never overflowsEasy · 8 min · 11 hidden tests
  3. 3Mean cross-entropy for a batchMedium · 14 min · 10 hidden tests
  4. 4Inverted dropout, training and evalEasy · 8 min · 11 hidden tests
  5. 5Xavier and He initialization scalesEasy · 7 min · 9 hidden tests
  6. 6Backprop through a two-layer ReLU networkHard · 25 min · 10 hidden tests

Practice it at full scale

Hands-on labs on real tooling, in your browser.

PyTorch in 45 MinutesA PyTorch Training Loop From Scratch

Where interviews test it

Cases from our interview prep that lean on this step.

Are you ready? Readiness check