Could ML predict the lottery?
I had a great time presenting last week during the Argyle Digital “PANEL DISCUSSION: Language Arts – Crafting Effective Techniques to Advance NLP” as part of the DATAx conference. You can watch the panel discussion for free via the link below...
Could ML predict the lottery?
Trying to predict the lottery is not a novel effort. People have considered this in a variety of ways. You can go out and Google this question and the results are pretty interesting. I ended up watching a video on YouTube from 2018 by Raj Ramesh.[1] My next searching effort was aligned to trying to find some academic work on the subject. That effort involved taking a few quick strolls around Google Scholar then heading over to ArXiv to see if maybe something was in pre-print.[2] Some research does exist and seems to surround a concept called the lottery ticket hypothesis. That research will not help you predict the lottery using machine learning or artificial intelligence. It actually seems to be explaining why the introduction of luck with overfitting is interesting, but problematic. A lot of people (349 citations worth) seem to like to reference a paper out of MIT CSAIL by Jonathan Frankle and Michale Carbin called, “The Lottery Ticket Hypothesis: Finding sparse, trainable neural networks,” that was sent to pre-print in 2018 and published in 2019 officially in a journal.[3]
From the abstract: “...we articulate the lottery ticket hypothesis: dense, randomly-initialized, feed-forward networks contain subnetworks (winning tickets) that—when trained in isolation— reach test accuracy comparable to the original network in a similar number of iterations. The winning tickets we find have won the initialization lottery: their connections have initial weights that make training particularly effective.”
What pretail does that really tell us about how machine learning could predict the lottery?
A lottery like the Powerball or some of the other large lottery programs are generally based on numbers. You have a defined number space and your selection of x numbers needs to match the randomly selected set of x numbers from the lottery result. You can 100% look for patterns over time and try to figure out the most likely numbers to be selected. That effort might even make you feel better, but assuming the machine designed to select the numbers accurately enforces randomness of selection the likelihood of any number being selected is going to be the same. It will be a fraction created by taking the one number being selected against the total number of balls in the number space being evaluated. Based on that little bit of logic you have probably discerned that you don’t need machine learning to predict the lottery. The best way to win the lottery is to explicitly buy a ticket for every number combination. That effort would be expensive, time consuming, and logistically problematic to achieve.
Links and thoughts:
Check out the latest episode of Machine Learning Street Talk from this week “#53 Quantum Natural Language Processing - Prof Bob Coecke”
I’m going to sneak in a guitar link this week… “Ernie Ball Music Man x John Petrucci: 20th Anniversary Signature Collection”
I sort of enjoyed this video from Linus this week on an old Dell computer from 2019
Top 5 Tweets of the week:
Footnotes:
[1] This is a video by Raj Ramesh called “Can AI Pick Your Next Winning Lottery Number?”
[2] https://arxiv.org/pdf/1906.02768.pdf is a paper tilted “Playing the lottery with rewards and multiple languages”
[3] Paper by Frankle & Carbin from 2019 on ArXiv https://arxiv.org/abs/1803.03635
What’s next for The Lindahl Letter?
Week 19: Fear of missing out on ML
Week 20: The big Lindahl Letter recap edition
Week 21: Doing machine learning work
Week 22: Machine learning graphics
Week 23: Fairness and machine learning
I’ll try to keep the what’s next list forward looking with at least five weeks of posts in planning or review. If you enjoyed reading this content, then please take a moment and share it with a friend.