During one training step a foghorn-log classifier first produces a guess from the current weights, then uses the mismatch to nudge those weights. Which half is which?
Select an answer to reveal the explanation.
Short Explanation
The guess is the forward pass. Sending the mismatch backward to nudge the weights is backpropagation. Inference is later scoring of new examples, not that training split.
Full Explanation
The forward pass computes a prediction from current weights. Backpropagation uses the loss to send the error backward and update parameters. Inference is later scoring of new examples, not that training split. CUDA kernels and cluster schedules are out of scope.