Have you ever opened a machine learning paper, only to feel overwhelmed by the language and jargon used? I know I’m not alone. It’s like hitting a wall on the very first page. It’s not the equations that get me, but the phrases like ‘Without loss of generality’ or ‘Convergence in distribution’. I find myself spending more time googling these terms than actually reading the paper.
Some people say, ‘Just push through, it’s how it works.’ But I’m not sure how you’re supposed to get value from a paper when 80% of the words are unclear. Others suggest only reading the intro and conclusion, but that feels like missing out on the meat of the paper.
And then there are the dependencies – citations, context, and all the underlying knowledge required to understand the paper. It’s like trying to drink from a firehose.
I’m curious, how do people actually read these papers without getting lost?
Take, for example, the Attention Is All You Need paper. There’s this expression: Attention(Q, K, V) = softmax(QK^T)V/root(dk). But the actual tensor process is much more complex, involving batch and layers before the tensor multiplications. Do domain experts really understand this, or do they have to read the code to get it?
Even the visual graphs don’t make it clearer. I know the authors try their best to explain, but the fact that I still don’t get it makes me feel even more frustrated.
So, How Do You Read ML Papers?
Do you have any tips or strategies for breaking down complex ML papers? How do you deal with the feeling of being overwhelmed by the language and jargon?
Share your experiences and advice in the comments below!