(55 minutes to learn)
The perceptron is a simple algorithm for binary classification where the weights are adjusted in the direction of each misclassified example.
This concept has the prerequisites:
- Know the perceptron update rule
- Optional: show that the algorithm terminates if the data are separated by some margin
- Why can't the algorithm terminate if the data are not linearly separable?
Core resources (read/watch one of the following)
→ Stanford's Machine Learning lecture notes
Lecture notes for Stanford's machine learning course, aimed at graduate and advanced undergraduate students.
→ Pattern Recognition and Machine Learning
A textbook for a graduate machine learning course, with a focus on Bayesian methods.
Location: Section 4.1.7, "The perceptron algorithm," pages 192-196
Supplemental resources (the following are optional, but you may find them useful)
→ Machine Learning: a Probabilistic Perspective
A very comprehensive graudate-level machine learning textbook.
Location: Section 8.5.4, "The perceptron algorithm," pages 265-266
- The perceptron was proposed in the 50s, although it's still in use. More modern algorithms have a similar form, but are put on a more mathematical footing:
- logistic regression , which is formulated as a probabilistic model
- support vector machines (SVMs) , which are formulated as an optimization problem
- The perceptron convergence proof requires the assumption that the data are linearly separable by a nonzero margin. Support vector machines (SVMs) are geared towards the same case.
- The perceptron can be kernelized in order to capture nonlinear dependencies.