CS231n: Convolutional Neural Networks for Visual Recognition | Hacker News Reader

CS231n: Convolutional Neural Networks for Visual Recognition

cs231n.stanford.edu

70 pointssdenton46 comments

One of the teachers of that class (Andrej Karpathy) has a blog with the best introduction to neural nets I've seen yet:

http://karpathy.github.io/neuralnets/

Those interested in CS231n may also find CS224d, Deep Learning for Natural Language Processing, to be interesting. There are notes and lecture videos online for that class as well (or at least there will be as the class proceeds):

http://cs224d.stanford.edu/syllabus.html

+1 for CS224d. Richard Socher is a great teacher and you can understand almost the entire class with very little background in NLP and Deep learning. However basic matuarity in LA will be helpful.

Both CS224d(NLP) and CS231n(vision) provides a great practical introduction to Deep learning. Thanks Stanford for proving such high quality material for free

With an IQ of just 125 and a BS in CS, I've struggled with this CS224d class. I've gotten half-way twice realizing that I don't know enough to learn more. I told myself, "Just keep going through the course and you'll eventually get it." This still isn't the case. I'll have to finish the videos and see what happens next. I get the general idea, but I definitely couldn't implement an artificial neural network after having just watched the videos. Perhaps reading the suggested readings would make a big difference.

Seriously, though. When I look at assignment 1, I don't even know what I'm supposed to do.

Some background in Linear Algebra and calculus will be really helpful.

If you want to implement backprop from scratch just look at the segment around https://youtu.be/I2TfdXfSOfc?list=PL05WXsDr_SWRGcuy5LA5eNjCq...

He explained really well how you can automatically get the derivation without going through painful calculation. Also i will suggest you to stick to the class as second half contains less maths and more fun techniques kind of like lego blocks :)

if you just want use neural net in production I highly recommend using something like theano/ keras

STAT212b: Topics Course on Deep Learning by UC Berkeley Statistics Department

https://github.com/joanbruna/stat212b

Reply on news.ycombinator.com