DOC: How does it work?

[MartinThoma]

It would be amazing if you could create documentation / a YouTube video / a podcast where you explain on a high level how it works.

I know and understand how CNNs work, but I haven't worked with attention-based models and I think I miss a couple of papers to understand Attend, Copy, Parse End-to-end information extraction from documents. Maybe you could just tell me what you think are good resources to understand this paper? Or blog posts about this paper?

[naiveHobo]

I have been trying to write a blog post explaining the working of the Attend, Copy, Parse Framework and the internal working of InvoiceNet, things like data preparation and input pipeline in detail. I've been a little swamped at work for the past few weeks so a final issue is still not ready. Need some more time to get everything together, I hope that's alright! But do keep an eye on this repo in the coming weeks, I shall link the blog post whenever it's out.

I've not come across any good resources on the original paper. However, I do still think that the original paper itself is extremely well explained though it does require a reading on attention as a prerequisite.

[MartinThoma]

I love the original paper, but I also felt that I didn't understand crucial parts.

I have been trying to write a blog

Cool, I'm looking forward to it! If you use Medium and want somebody to review it, you can sent me a draft link ([email protected])

[victor-ab]

I also felt like the paper was too fast/superficial on the crucial parts.

Please let us know whenever you do that explanation.