Perceptron
The perceptron is the original trainable linear classifier: a weighted sum plus a threshold. It is the atom of neural nets, and it cannot learn XOR without a hidden layer.
Definition
Rosenblatt (1958): ŷ = sign(w·x + b), with a mistake-driven update. The perceptron convergence theorem says linearly separable data are learned in finite steps. Minsky–Papert showed the single-layer version cannot represent XOR.
Why it matters
Every dense layer is a bundle of perceptrons (with a smoother activation). The XOR lesson is why we stack layers.
Case
A linearly separable credit screen can be a perceptron. A two-factor XOR-like regime screen cannot — you need a hidden layer or a feature map.
How to read it
If someone says ‘neural net’ and means one linear layer, they mean a perceptron. Ask for depth.
Key papers
Landmark references with real publisher or arXiv links — not token-overlap noise.
- The Perceptron: A Probabilistic Model for Information Storage and Organization in the BrainRosenblatt · 1958 · Psychological ReviewThe perceptron paper.