Principal Data Scientist, Apple. ML/DS researcher and author of the book above. Dyadic, normative, partially hydrogenated. I don't speak for my employer, ever.
Cupertino, CA
Joined May 2009
- Tweets430
- Following0
- Followers444
- Likes184
Tom Fawcett retweeted
Fox News cut away from Trump's rally immediately after his brain short-circuited
An introduction to artificial life for people who like #AI : thegradient.pub/an-introduct…
The aphorism "All models are wrong, some are useful" is like "Drastic times call for drastic measures". It's often used to justify something really awful.
Please stop saying that #Python closely resembles the English language. Just stop. It's a ludicrous thing to say and the comparison doesn't help anyone.
Slides worth looking at: How to recognize #AI snake oil.
cs.princeton.edu/~arvindn/ta…
"For predicting social outcomes, AI is not substantially better than manual scoring using just a few features"
Case study of unfortunate evaluation methodology. Yes, deep learning predicts earthquake aftershock patterns (blog.google/technology/ai/fo…) --- but no better than logistic regression (arxiv.org/abs/1904.01983).
These are classification performance curves from two models (called datasets here), followed by a table of common metric values. Which model is better? #DataScience
"Please Stop Explaining Black Box Models for High-Stakes Decisions", a new paper by @CynthiaRudin, argues the importance of inherent interpretability of learned models. arxiv.org/pdf/1811.10154.pdf
Tom Fawcett retweeted
2037. AI-powered Amzoogle drones scan the charred landscape, reading hundreds of uncommented YAML files that determine which humans to hunt. In a cave, a child huddles next to a blind old man. He whispers, "Objects are combinations of data and methods." Her Java grows stronger.
Interviewer (Human): What is you greatest strength?
Deep Neural Network (DNN): I am an expert in Machine and Deep Learning
Human: What's 10 + 2?
DNN: Zero
Interviewer: Nowhere near, it's 12.
DNN: OK, 12
Interviewer: What's 10 + 5?
DNN: 12
\H.Zenil\
Wieland Brendel shows that ImageNet is simpler than we thought, and NNs are using surprisingly superficial evidence. medium.com/bethgelab/neural-…
What could go wrong? newscientist.com/article/218…
That's not the way it's used. "Democratization is defined as the action/development of making something accessible to everyone, to the 'common masses.'" kdnuggets.com/2018/01/democr…
I'm confused why people talk about "democratizing" #AI. Is requiring a knowledge&skill undemocratic? Do we complain that car repair, piano playing, football, archeology, and horseback riding haven't been democratized?
How to delete your #quora account: Login, click on your account image in upper right, click Privacy in upper left, go to bottom, click Delete Account. Confirm.
A clear, concise explanation of proof-of-work from @JohnDCook : johndcook.com/blog/2018/11/1…
Tom Fawcett retweeted
Just started reading @sejnowski's book, "The Deep Learning Revolution". Looks really interesting, but I got stuck on the very first paragraph. (See below.) The highlighted part is far from being true!
The man who should have won the 2008 Nobel Prize in Chemistry drives a courtesy van for a car dealership in Alabama. Fascinating article by @barabasi . nautil.us/issue/66/clockwork…