The oldest ML algorithm: Perceptron (1957)

In Day 10 of the ML: Teach by Doing project, I learnt about one of the oldest ML algorithms: The Perceptron.

The old ways have a lot to teach us!

The Perceptron is quite cool because of the following reasons:

(a) It was developed in 1957, when no one knew or cared about ML.

(b) It was not just an algorithm, people even built a machine to implement it! Here’s a photo of that machine.

(c) It was one of the first algorithms which learnt from it’s mistakes. No one called it machine learning then, but it was.

(d) It is just a 9 line algorithm, but each of those nine lines is quite profound.

I realised that perceptrons provide an intuitive understanding of how machines learn from data. In that sense, they are the stepping stones to ML as we know today: neural networks, transformers etc.

Modern ML courses miss out and omit this crucial algorithm of ML history.

Like in life, we should not discard the old ways. They have a lot to teach us, and we have a lot to learn from them.

Here’s the video I made. It turned out to be quite long: 1 hr 15 mins. Nothing worthwhile comes quick!

Students who are not engaged by theory, are engaged by strong visuals. I have tried to make the algorithm description as visual as possible. I have also tried to explain the intuition behind each and every step of the algorithm.

That way, even if you forget the algorithm later, the intuition stays with you. That should be the aim of every ML teacher.

My lecture notes can be accessed here: Link

Stay tuned for Day 11. More exciting material coming up soon!