# Least Mean Squares Algorithm
**Entity class:** Algorithm / adaptive filtering
## Definition and history
The least mean squares algorithm updates weights in response to an observed error, making an adaptive filter respond to changing signal conditions. [[articles/Intelligence Inside|Intelligence Inside]] follows the Widrow–Hoff rule from ADALINE into telephone equalization and echo cancellation. Its operational name in communications engineering is LMS, an iterative approximation to minimizing mean squared error. Continuous coefficient adjustment is machine learning even when the product documentation calls the task filtering.
## Mechanism and limits
Robert Lucky's original adaptive equalizer used zero forcing, so the invention of the equalizer and its later LMS update rule have different authorship. LMS can update each tap at a sample rate in a running filter, whereas a perceptron branch predictor updates weights under a gated training condition. Counts of updates across these systems expose scale, while the operations have different precision, cost, and purposes.
## Relationships
**Co-authors:** [[wiki/Bernard Widrow|Bernard Widrow]] and [[wiki/Marcian Hoff|Marcian Hoff]]. **Early hardware:** [[wiki/ADALINE|ADALINE]]. **Later use:** [[wiki/Adaptive Equalizer|Adaptive Equalizer]] and [[wiki/Echo Cancellation|Echo Cancellation]]. **Article route:** [[articles/Intelligence Inside|Intelligence Inside]].
## Sources
[Widrow, *Thinking About Thinking*](https://isl.stanford.edu/~widrow/papers/j2005thinkingabout.pdf); [[articles/Intelligence Inside|Intelligence Inside]].