Hacker News Reader

Comment by Jun8 | Hacker News Reader

Made me remember an interesting anecdote: During my PhD at Purdue I had the good luck to take the Pattern Recognition (nowadays would be called ML) course from Prof. Fukunaga, one of the pioneers in the field (https://en.m.wikipedia.org/wiki/Keinosuke_Fukunaga). In one of his classes he mentioned that years ago a company came to him, requesting his expertise in optimizing movement of a bank of elevators. He thought it would be trivial :-) After telling us that all his approaches failed miserably, he had one of his impish laughs and told us that he never worked on that problem again. Miss that guy. Get a copy of his book (https://books.google.com/books/about/Introduction_to_Statist...) if you can, it’s written in the “old style” but very informative.

Interesting fact: Gen Fukunaga (https://en.m.wikipedia.org/wiki/Gen_Fukunaga), founder of anime company Funimation, is his son.

View on HN

There is this anecdote in the text "Operations Research, An Introduction." by Hamdy Taha, which I consider to be a very good example of picking the right tools for the job. An OR group was hired to fix elevator wait times in an office building. After studying the problem they installed full-length mirrors at the entrance to the elevators. Complaints around elevator service disappeared as people were now engaged in watching themselves and others while waiting.

I first heard of this from Dr. Russell Ackoff's Systems Thinking Speech. I had a mini mind blown moment.

Reminds me a parable from Are your lights on by Gerald Weinberg.

I’ve always wondered if the main computer in Star Trek has to be so big just to manage the turbolifts. Three dimensional elevators with multiple exits per level, probably prioritizing routes dynamically based on the passengers and the military alert level. Imagine the algorithms.

I think it would fundamentally be a constraint based programming problem, eg the sudoku solvers, travelling salesman implementations, software that tells eg Amazon warehouses which parcels go in the van first based on both size and the route it’s planned so that the driver never had to dig all the parcels out to get the buried ones.

In my ML class some students did a project on using reinforcement learning to optimize elevators. They told me that after 100 years of operation the algorithm would have enough data to start working really well.

Boiler Up!

I had Prof Fukunaga for EE301 I think (Might have been EE202). I was too young and dumb to recognize his brilliance.

Awesome, it’s a small world. I was his teaching assistant during the times he taught EE301 (Signals & Systems) so chances are I’ve corrected your exams!

I started in physics before switching to CS. I'm glad the ECE department is good, because the CS department was terrible.

I pounded my head on this problem in a multi-agent system course in graduate school. Needless to say, I could not outperform a simple naive algorithm however hard I tried.

Reply on news.ycombinator.com