Learning representations by back-propagating errors

Back-propagation asks how a hidden layer can learn useful features when a teacher specifies only the desired output. Read it to see how a forward pass stores activations, an output mismatch travels backward as chained derivatives, and each connection receives a local weight update that can organize hidden units around the task's structure.

Reading focus: Why hidden units create a credit-assignment problem: unlike output units, they have no teacher-provided target state. How the chain rule sends output-error derivatives backward through the same weighted connections and produces a gradient for every weight. What the mirror-symmetry and family-tree examples demonstrate, and why local minima, biological plausibility, historical priority, and missing reproduction details still limit the claim.

Nature 1986. David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams. 30 min read, easy difficulty.