Perceptrons remains one of my favorite books in many ways. Obviously the single-layer networks Minsky & Papert analyze are very primitive relative to today's d

Perceptrons remains one of my favorite books in many ways. Obviously the single-layer networks Minsky & Papert analyze are very primitive relative to today's deep neural networks, and the actual mathematical content may not be of direct use today. But the arguments were beautiful & insightful, and the exposition clear. For anyone interested in a mathematical understanding of computation in (artificial) neural networks, it's still not a bad place to start.