Foundations · Foundational · Beginner
Perceptron
The earliest neural unit: a weighted sum of inputs passed through a threshold to produce a binary decision.
What Perceptron is
Introduced in 1958, the perceptron is a single linear classifier. Its inability to solve non-linearly separable problems such as XOR was a famous early limitation, resolved by stacking layers with non-linear activations.
How it works
Weights are updated whenever the unit misclassifies an example, nudging the decision boundary towards the correct side.
Why it matters
It is the historical and conceptual seed of everything in deep learning, and explains why depth and non-linearity were necessary.
Common uses
- →Teaching linear classification
- →Simple baselines
Strengths
- ✓Extremely simple
- ✓Convergence guaranteed if data is linearly separable
Watch for
- ✓Cannot represent non-linear boundaries
Continue exploring
More in this collection
Browse all AI Concepts