On Machine Learning :
Article from 2 months back on Machine Learning. Answering the big question : can you do it without the mambo jumbo of academia?
His answer :
Neural network math is simply matrix multiplication. All of it. And if you know it, that's all you need.
And back propagation calculus calculates automatically from neural network. You don't need all those 10 pages of derivative proves commonly seen in academic papers. And even if you need it, you will only need derivatives of square functions.
But surely an academic will outperform the experienced through his academic knowledge. Right?
His answer : Machine learning is an empirical field. Meaning people work here with raw data giving raw results, and not with finding what theoretical calculations say. Therefore experience is everything. Not the theory.
Further : he says he now learns more from "outsider"s on twitter than from peer-reviewed academic papers.
That's all. The world hasn't chagned much.
.
https://marksaroufim.substack.com/p/machine-learning-the-great-stagnation
- Comments:
- Who knows? He sounds like he knows things, and that's all matters. Sounding confident enough to impress others.