menu
Logo
99+
Bobcat Attack Menu
Logo
99+
Ohio Basketball Topic
Topic: A look back on season one of the NET rankings - from Busting Brackets
Page: 1 of 1
mail
OU_Country
8/26/2019 12:09 PM
Some discussion on the NET rankings:

https://bustingbrackets.com/2019/08/26/ncaa-basketball-lo... /
mail
TWT
8/27/2019 9:27 PM
NET gives mid majors more of a chance to have a high seed if they run the table and dominate. Quadrant records sort out the Top 100 but how accurate is that if one team is 4-2 while the other is 8-4? RPI was a good system but it was devalued so much in the eyes of the committee that it had increasingly little weight.
mail
OU_Country
8/28/2019 11:02 AM
Uncle Wes wrote:expand_more
NET gives mid majors more of a chance to have a high seed if they run the table and dominate. Quadrant records sort out the Top 100 but how accurate is that if one team is 4-2 while the other is 8-4? RPI was a good system but it was devalued so much in the eyes of the committee that it had increasingly little weight.
I agree - I like the NET now that we've seen it used over time. My only complaint about NET is that they haven't been willing to release how different things are weighted. I've always been a proponent of the idea that winning games should be the most important thing when it comes to at-large selection, with a huge emphasis on road/neutral wins. Because it's been proven time and time again that home court means more in college hoops than almost any other sport.
mail
person
NewAthenian
8/28/2019 12:55 PM
Don't quote me here, cuz I've only picked up a few things here and there (no expert), but that's rarely stopped me from shooting from the hip. The article says the formula was "optimized with machine learning." That tells me they used "AI" to suss it out. This will stand in direct contrast to transparency for a number of reasons.

First instinct is to say that any AI R&D firm is never going to tell their secrets. The truth is that many times, the firm has no idea how the AI did what it did. If they knew in the first place, they wouldn't need the AI. After the fact it is too late because these "deep learning" things are pretty tough to reverse engineer. I mean, in some cases you can go back and see what it did, but it would make little sense to the human brain.

Best guess in this case is that they used the AI to figure out those weights - it's not like T-2000 will be picking the field come March. And they could even publish those weights, but if asked to defend why this stat was weighted more and that stat was weighted less they couldn't say anything other than "Skynet said so, all hail our new robotic overlords."
mail
OU_Country
8/29/2019 10:33 AM
NewAthenian wrote:expand_more
Don't quote me here, cuz I've only picked up a few things here and there (no expert), but that's rarely stopped me from shooting from the hip. The article says the formula was "optimized with machine learning." That tells me they used "AI" to suss it out. This will stand in direct contrast to transparency for a number of reasons.

First instinct is to say that any AI R&D firm is never going to tell their secrets. The truth is that many times, the firm has no idea how the AI did what it did. If they knew in the first place, they wouldn't need the AI. After the fact it is too late because these "deep learning" things are pretty tough to reverse engineer. I mean, in some cases you can go back and see what it did, but it would make little sense to the human brain.

Best guess in this case is that they used the AI to figure out those weights - it's not like T-2000 will be picking the field come March. And they could even publish those weights, but if asked to defend why this stat was weighted more and that stat was weighted less they couldn't say anything other than "Skynet said so, all hail our new robotic overlords."
Because I'm sometimes skeptical of "AI", this seems to fit in here:
https://youtu.be/w9dwSh9q7-Q
mail
person
Brian Smith (No, not that one)
8/29/2019 5:00 PM
This message board is slowly but surely becoming an engineering forum. Flood plains. Zoning restrictions. Self-driving cars. Artificial intelligence. In 2031, people are going to wonder why the biggest forum for engineering talk is called BobcatAttack.com.

Everyone is going to assume Tim O'Shea was the world's greatest engineer, because why else would he be mentioned so often.

I'm okay with all of this.
mail
GraffZ06
8/29/2019 11:49 PM
Not to completely engineering nerd thread derail but...there's a big difference between "machine learning" and "AI". Machine learning is just a very small subset, or basic building block of AI.

Machine learning can take many forms from simple rule based reasoning, to Bayesian learning to neural networks or decision trees - and can be either supervised or unsupervised. But, it's goal is to fit data and find patterns based on ingesting lots and lots of data with which it trains or "learns" from.

Using ML to do regression analysis to form statistical models for analysis is something that is done commonly - and a perfectly acceptable use for developing predictive performance algorithms for something like, say, oh....comparing college basketball teams.

This, however, has absolutely NOTHING to do with A.I. or the machines from Skynet or the Matrix taking over the world. It's mostly just applied mathematics and statistics.

/end Engineering rant
Last Edited: 8/29/2019 11:51:31 PM by GraffZ06
mail
person
NewAthenian
8/30/2019 11:01 AM
Ah but come on, AI fear-mongering and dropping pop-culture references are fun. If we can't go off topic, what do we come here for!
mail
OhioCatFan
8/30/2019 11:48 AM
I've yet to see an AI system where the emphasis should not be put on the word "artificial."
mail
GraffZ06
9/2/2019 1:17 AM
OhioCatFan wrote:expand_more
I've yet to see an AI system where the emphasis should not be put on the word "artificial."
QFT. Especially considering true "A.I." doesn't really even exist yet. The field of bioinformatics is making some strides, but it's a long LONG way away yet.
Showing Messages: 1 - 10 of 10
reply
vertical_align_top
vertical_align_center
vertical_align_bottom
add_comment
MAC News Links



extra small (< 576px)
small (>= 576px)
medium (>= 768px)
large (>= 992px)
x-large (>= 1200px)
xx-large (>= 1400px)