Skip to content

Instantly share code, notes, and snippets.

@jxnl
Last active August 12, 2017 16:56
Show Gist options
  • Save jxnl/719678ded76d15ffd0606dc5b9b11a36 to your computer and use it in GitHub Desktop.
Save jxnl/719678ded76d15ffd0606dc5b9b11a36 to your computer and use it in GitHub Desktop.

2cents on diversity.

Some friends have asked me about my thoughts on google memo.

Stereotype threat is an observable phenomena. 

It doesn’t explain every performance gap, and eliminating it won’t close all performance gaps, however, it’s net loss and suboptimal.

tl;dr

Large part of the diversity funnel issue is a lack of role models or ‘people who seem to be like me’ Thinking diversity is locally optimal over group performance is most likely false when taking generational benefits  I use math to shows that diverse algorithms and models work better together than multiple versions of the similar models.  

1. Role Models and Diversity Hires

Having a diverse set of leaders provides us a diverse set of role models. With this, young people can find common ground and aspire to become the next greats.

Humans are so stupidly good at pattern recognition that we probably overfit on these patterns as a kid. We are also so stupid that once we see these patterns of who success looks like, we become stubborn.

No one is going to tell a Chinese or Indian dude that they can’t success in STEM even if the they are failing courses and has shit performance. They will probably apply to jobs they can’t get or retake courses they fail. This is from experience.

I can’t say the same for those who have no one to look up to. :/

People say diversity hires are suboptimal. From my experience, it’s rare to see locally optimal systems like the hiring process become globally optimal, without the presence of strong feedback loops.

If “diversity hires” help inspire the next generation of hard working folks to come into a field that I think is awesome, that’s a fucking win.

2. Diverse teams

Everyone knows that and ensembled method of experts only outperform the best expert of each model is uncorrelated. Duh

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment