Principles of Neurodynamics
Rosenblatt gathered perceptron theory into a book: multilayer schemes, the convergence theorem, the limits and the open questions.
Why it matters
The full statement of the 1960s connectionist programme, and the main text Perceptrons was written against seven years later.
The book covers more than the single-layer case: Rosenblatt considers perceptrons with several layers and cross connections. For the three-layer series-coupled scheme (section 13.3) he gives a probabilistic error-correction procedure that propagates corrections back from the response end towards the sensory end, and proves (theorem 2) that it reaches a solution in finite time with probability 1 if one exists; it is not gradient back-propagation. For multi-layer and cross-coupled systems in general he writes himself that a great deal still remains to be done. That gap, the training of arbitrary deep networks rather than the single-layer limitation, is what stayed unresolved until the 1980s.