← All skills
Skill · Neural nets, built by handLive

Deep Learning

Build the network from first principles in NumPy — neurons, activations, forward pass, backprop, and gradient descent. The intuition every framework hides.

  1. Task 1A single neuronEasyBuild the smallest unit of a network: weighted sum plus bias.~11 min · 1 blanks→
  2. Task 2Activation functionsEasyAdd the non-linearity that lets networks learn curves.~11 min · 1 blanks→
  3. Task 3A layer of neuronsMediumRun a whole layer at once with one matrix multiply.~12 min · 1 blanks→
  4. Task 4Forward passMediumChain two layers into a full forward computation.~13 min · 1 blanks→
  5. Task 5Loss functionsMediumPut a single number on how wrong the network is.~12 min · 1 blanks→
  6. Task 6Gradients by handHardEstimate a derivative numerically — the slope of the loss.~13 min · 1 blanks→
  7. Task 7BackpropagationHardChain derivatives through a neuron to get dL/dw.~15 min · 1 blanks→
  8. Task 8Gradient descentHardStep downhill repeatedly to reach a minimum.~14 min · 1 blanks→
  9. Task 9Train an XOR netHardSolve the problem a single neuron famously can't: XOR.~18 min · 1 blanks→
  10. Task 10A full training loopHardTie it together: predict, measure, and descend until it fits.~16 min · 1 blanks→