A short historical past lesson on machine studying origins that proved to be a helpful studying train.
In 1959, in a small workplace at MIT, scribbled on a blackboard was a weight replace algorithm. It was step one in direction of machines that might study. Bernard Widrow and a PhD candidate named Ted Hoff had simply talked via an issue in Widrow’s analysis [1]. At this second, nevertheless, they weren’t centered on synthetic intelligence.
What they have been speaking about was a sign denoising technique. A technique that might study from noisy knowledge. It got here from a area that’s a lot forgotten about within the historical past of machine studying — adaptive filters.
What resulted was one of many first concepts in direction of one of the broadly used studying algorithms, the least means squared (LMS) algorithm.
The work right here on the LMS algorithm ultimately led them to coach the primary neural community, ADALINE [1]. It additionally set the framework for future work in backpropagation.
I first heard this story within the guide, Why Machines Study. I might have simply taken as only a historic notice. However, I usually discover helpful insights when taking a look at how authentic inventors approached an issue. It usually removes a number of the complicated facets seen in additional developed…
