# Frank Rosenblatt > 1928–1971 · Psychologist, Creator of Perceptron > > **Recorded contribution:** Developed the trainable perceptron and Mark I Perceptron, building on earlier neural models and advancing connectionist learning ## How to use this dossier Read for a causal chain, not a hero story: inherited problem → contribution → mechanism → downstream capability → limit. Then close the page and complete the reconstruction exercise from memory. ## 1. Historical orientation Frank Rosenblatt (1928–1971), a psychologist at Cornell Aeronautical Laboratory, developed the perceptron as a learning model and built the Mark I Perceptron. Earlier neural models by McCulloch and Pitts and learning rules associated with Hebb were essential precedents. Rosenblatt's perceptron adjusted weights from labeled examples rather than requiring every classification rule to be hand-coded; for linearly separable data, the convergence theorem guarantees it will find a separating hyperplane. Public claims that it was the first neural network are too broad. Its real importance is the union of a trainable mathematical classifier, a proof, experimental hardware, and an ambitious research program. ## 2. The problem inherited Pattern recognizers depended on hand-designed templates and rules, while researchers wanted machines that improve classification behavior from examples. ## 3. The central contribution Rosenblatt formalized and experimentally developed the perceptron: a weighted threshold classifier with an error-driven update rule and provable convergence on separable training data. ## 4. Reconstruct the mechanism 1. Encode an example as numerical features and compute the weighted sum of those features plus a bias. 2. Apply a threshold to produce one of two predicted classes. 3. When the prediction is wrong, move each weight toward the correctly labeled example by a learning-rate-scaled update. 4. Repeat over examples until classification stabilizes or continued errors reveal that no linear separator has been found. ## 5. What changed downstream - The perceptron established supervised weight adjustment as a concrete machine-learning procedure. - Its convergence analysis linked learning behavior to geometry and separability. - The line of work became a predecessor of multilayer neural networks and modern linear classifiers. ## 6. Attribution, limits, and uncertainty - A single perceptron cannot represent XOR or any classification that is not linearly separable in its feature space. - Rosenblatt built on earlier neural and learning theories and worked with an experimental engineering team. - Convergence on training data does not by itself give calibrated probability, robustness, fairness, or good generalization. ## 7. Reconstruction lab Implement the perceptron update and train it on AND, OR, and XOR. Plot decision boundaries and update counts; prove why XOR fails in the original two-dimensional features, then add one engineered feature or hidden layer and explain what changed. ## 8. Evidence trail - [The Perceptron: A Probabilistic Model for Information Storage and Organization](https://doi.org/10.1037/h0042519) — Psychological Review - [Frank Rosenblatt and the Perceptron](https://news.cornell.edu/stories/2019/09/professors-perceptron-paved-way-ai-60-years-too-soon) — Cornell University --- *Research checked 2026-08-09. Dates, roles, and claims about living people are historical snapshots. Linked sources remain the authority; this dossier is original instructional synthesis.*