Hacker News new | past | comments | ask | show | jobs | submit login
Ask HN: Where to start learning about neural networks?
6 points by sarreph on June 28, 2015 | hide | past | web | favorite | 5 comments
Hey HN,

I'm a high-level (language-wise) programmer, who is fascinated by the current developments in neural networks and deep learning. However, in a bid to understand more about this fascinating technology, I've realised that I have no idea about where to start leaning about neural nets.

I am certainly not hoping to (anytime soon) get to the point where I can build one myself, but would be hugely grateful if anybody with experience in the field could provide me with an overview or pointer of how to start getting into understanding deep learning, AI, and neural network.

Reading suggestions, article or even course links would be greatly helpful!

Cheers HN, and happy Sunday! ;)




As a mostly self-taught programmer, I honestly didn't grok the maths until I started doing my Master's, which is why I highly recommend going back to school. Take a machine learning course at the best university you can afford.

Speaking of which, while programming is a nice skill to have, most of the toughest bits of modern AI approaches, including neural networks, is in the maths. For neural nets, you need to have a solid understanding of multivariate calculus, statistical methods, and probability theory. You should have at least an intuition of what the error function is, and how error backpropagation works.

In my course, we used the Bishop book (Pattern Recognition and Machine Learning, https://books.google.com.au/books/about/Pattern_Recognition_...). It's dense, and very math-heavy, but it's definitely worth slogging through.


I'm an undergrad who's taken a single linear algebra course, gotten through the first round of multivar calculus and statistics. I found the link below to be a really good outline, and impetus for further googling. Basically, I'll read through it, and whenever I don't recognize a term, I'll spend an hour or two going through that term, and then return to this link for some contextualization of whatever the hell it is that I just read!

http://www.iro.umontreal.ca/~pift6266/H10/notes/mlintro.html


I highly recommend Hugo Larochelle talks:

https://m.youtube.com/playlist?list=PL6Xpj9I5qXYEcOhn7TqghAJ...


I started with Andrew Ng's Machine Learning course on Coursera [1]. He presents the entire subject, including NNs and prerequisite theory, in a non-intimidating, intuitive fashion. There are simple coding exercises, such as digit recognition using NNs.

Once you're familiar with the basics, you can go deeper into the subject with the books suggested here.

[1]: https://www.coursera.org/learn/machine-learning


Hi, check out Jeff Heaton's books http://www.heatonresearch.com/




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: