May 22, 2017

Nice article on how momentum works

http://distill.pub/2017/momentum/

Modeling individual then average vs modeling average (using Deep Neural Network)

"Our belief is that forcing a neural network to model the individual doctors and then averaging at test time should give better generalization than simply training a neural network to model the average of the doctors."
Reference: https://arxiv.org

May 18, 2017

When love dies!



May 15, 2017

Supernatural event and religion


If I don't understand something now then it must be wrong. Case dismissed.
Really!

Reference



May 11, 2017

Elon Musk

1. The success of expert-generalists throughout time shows that this is wrong. Learning across multiple fields provides an information advantage (and therefore an innovation advantage) because most people focus on just one field.

2. Learning transfer is taking what we learn in one context and applying it to another. It can be taking a kernel of what we learn in school or in a book and applying it to the “real world.” It can also be taking what we learn in one industry and applying it to another.
 deconstructs to learned the core -> reconstructs them in a new way

Ask following to questions when you read something:
1. “What does this remind me of?” and
2. “Why does it remind me of it?”

Reference

May 2, 2017