The “simple idea” of Maximum Likelihood is anything but.
There’s an epic history of a turbulent past, of surviving numerous assaults on the core idea, culminating in a beautiful yet complicated theory.
This paper is highly entertaining and educational.
https://t.co/sbyqNOx9nl
Added a new demo to minGPT that trains a GPT on pixels of CIFAR-10 images instead of text. Quite powerful that one can run the same training code/model on both domains. Notebook: https://t.co/nAt9VWXnrG . Produced reasonable samples after ~only 30 minutes on an 8-GPU V100 node:
Grant writing advice. Many people get too detailed and assume background knowledge of the reviewer too quickly. Your grant early on needs to answer (in my view) these questions:
I got really into a bunch of 1980s-era papers about histogram thresholding, and wrote a weird paper. I sent it to ECCV assuming the reviewer response would be "why are you writing a direct response to two 40 year old papers" but they loved it, so hey. https://t.co/BgZ9dETc8p
1/4 WTF guys I think I broke ML: loss & acc 🡅 together! reproduced here https://t.co/bK2XZm3Vxc. Somehow good accuracy is achieved *in spite of* classic generalizn theory (wrt the loss) - What's goin on? @roydanroy@prfsanjeevarora@ShamKakade6@BachFrancis@SebastienBubeck
1/4
Arithmetic mean vs. Geometric mean: Some little known big ideas.
First, the well-known: Arithmetic mean (M) is the average, whereas the geometric mean (m) is the rooted product of n numbers. M dominates m, but they are the same if the numbers are equal.
To beat COVID-19, we need contact tracing apps. But does that mean sacrificing our right to privacy?
HECK NO ✊
Here's a comic (collab w/ @carmelatroncoso) explaining how we can beat COVID-19 *and* Big Brother!
🧵COMIC THREAD BELOW 🧵
(as 1 page: https://t.co/RiZwE6bqwy )