[CaCL] memory network papers

[William Schuler]

hi all,

i don't see us discussing four papers in one day.  evan, can we perhaps triage a couple of these papers (say #3 and #4) and discuss when we meet whether to reorganize our schedule to cover them on a subsequent day?

wm

> On Sep 18, 2017, at 3:13 PM, Evan Jaffe <jaffe.59 at buckeyemail.osu.edu> wrote:
> 
> Hey all, remember we're doing 4 memory papers this week? Yeah, I forgot
> too - start reading now ;)
> 
> Sukhbaatar et al 2015: End-to-end Memory Networks
> https://arxiv.org/abs/1503.08895
> 
> Kaiser et al 2017: Learning to Remember Rare Events
> https://arxiv.org/abs/1703.03129
> 
> Kumar et al 2016: Ask Me Anything: Dynamic Memory Networks for NLP
> https://arxiv.org/abs/1506.07285
> 
> Miller et al 2016: Key-Value Memory Networks For Directly Reading
> Documents https://arxiv.org/abs/1606.03126
> 
> 
> I swapped the original Weston out for the Sukhbaatar, which seems a
> little more polished and readable.  Everyone cites it too, so it's #1
> priority for discussion.  (feel free to look at Weston 2015, Memory
> Networks as well)
> 
> The Kaiser is the most recent and has some nice decay-like properties,
> and that'll be priority 2.
> 
> Kumar is 3rd and extends earlier Weston models to sequences,
> incorporating positional information. Haven't looked at it in detail yet.
> 
> The Miller adds some encoding hashing something to the Sukhbaatar, so it
> seems like an incremental improvement given my short skim - lowest
> priority unless I get lots of angry mail with convincing arguments to
> the contrary.
> 
> See you guys Thursday,
> 
> Evan
> 
> 
> 
> 
> 
> 
> 
> 
> _______________________________________________
> CaCL mailing list
> CaCL at lists.osu.edu
> https://lists.osu.edu/mailman/listinfo/cacl