Being better at Deep Learning isn't a feat achievable in few days or weeks. It might take months or years as the field keeps evolving at a rapid pace.
The solution is simple...
Read research papers and keep up with the new trends and emerging advances.
But where to start?
If you are beginner (like me) and want to get your foundations gain strength, this article is for you!😀
First off, Begin Compilation
Compile a list of papers that interests you from various sources.
Try taking a shot at one paper, skim it; don't like it, skip. Try next one, skim it; like it, read a related paper; like it a lot? Nice, go ahead read it more. Check other papers like this one which interests you and get started.
Don't read it the bad way by reading from first word to the last word.
Read it the Good way
- Take multiple passes through the paper.
- Read heading/abstraction/figures first to get an abstract of what the author wants to convey.
- Then read Intro + Conclusion + Figure; the author tries their best to put a very clear summary in these sections.
- Skip/Skip math.
- Read the whole thing but skip parts that don't make sense. Maybe even try reading maths.
After you have finished reading (and hopefully understanding!😀) the paper, try answering these questions:
- What did the authors try to accomplish?
- What were the key elements to the approach?
- What can you use yourself?
- What are other references do you want to follow?
If you really want to test that you have thoroughly understood everything, try deriving the derivations from scratch.
Code and implementation
Instead of simply running open-source code, try implementing it from scratch. This might be more useful for ML practitioners who might not want to derive the math equations but want to test themselves by implementing it.
Steady and consistent reading is more important than short burst reading.
Keep Learning ✌