Reading deep learning papers

I noticed that during the course, @jeremy made many references to academic papers. Knowing that a given architecture or technique can be useful before it becomes mainstream can give you an edge over competitors in kaggle, for example.

I assume that to keep track of the Deep Learning world, one has to read papers every now and then. So my question is: how to do it. I can imagine one can go to Google Scholar and search for ‘convolutional networks’ and keep track of Hinton’s new publications on a weekly or monthly basis, but is there any better way? Like a newsletter of latest useful developments of something like that.

1 Like

Hey,

So the way i do it is by checking out what Jeremy likes on twitter. Those are mostly useful stuff. And I check out newsletters like fastML or ImportAI. Good way to see what’s coming.

As for papers I check out arxiv-sanity. I import papers into this app called Mendely and read/highlight them. If i dont understand something I google/ask in the forums! :slight_smile:

8 Likes

As it happens, we just discussed this in our in-person part 2 lesson. Here’s the links I provided (some of which @karthik_k314 mentioned already):

16 Likes

@jeremy
Do you use something else for organizing web pages like this? Mendely just stores URLs. But we can lose that info in case the page is removed. I can add that to evernote but do you have a better suggestion?

I just use Mendeley. It’s worked fine for me so far.

At work we use Paperpile to organize papers, and it works pretty well. (limited to Chrome browser, though)

Zotero is an open source alternative to Mendeley. The latter was bought by Elsevier in 2012. I use Zotero on Firefox, it has 1click downloads with metadata and needs little tweaks to add highlighting.

If you get overwhelmed & scared of math, my recommendation is to actually print one paper at a time, go to a coffee shop with a pen, highlighter, notebook, + no laptop, and slowly go through them one sentence at a time.

my tendency with this stuff specifically is to freak out and give up as soon as I see math, but when in a focused environment & being patient with myself it’s not that scary after all!

Seconding Mendeley on desktop to keep track of what I’ve read / what to read!