Memory, attention, sequences

towardsdatascience.com

102 pointsjonbaer3 comments

Before I went to bed I checked HN and this was on the front page.

Now many hours later it's still on the front page with 74 upvotes and absolutely zero comments. What's going on here? I've never seen this happen.

I saw it yesterday too. I'm interested in attention models, but I'm not a skilled enough practitioner to say anything meaningful about them yet.

Someone linked me the author's more recent article [1] yesterday which claims hierarchical attention models totally outclass traditional RNNs/LSTMs. I really want to read an empirical comparison or a HN thread with experts chiming in.

[1] https://towardsdatascience.com/the-fall-of-rnn-lstm-2d1594c7...

Someone is probably cheating at HN and giving boosts to topics that seem machine learning related. Or perhaps a hidden number of clicks is used.

Reply on news.ycombinator.com