WaveNet follow up

7 views
Skip to first unread message

Stephen Roller

unread,
Nov 1, 2016, 1:26:14 AM11/1/16
to ut-flare
Recall me asking whether the conclusion of those papers was that maybe we should remove RNNs?

Well they did it.

https://arxiv.org/abs/1610.10099

Haven't read yet but I'm excited

Stephen Roller

unread,
Nov 1, 2016, 12:07:46 PM11/1/16
to ut-flare
this comment puts it best:


apology for poor english

when were you when lstm died?

i was sat in lab launching jobs in cluster

‘lstm is kill’

‘no’

Joel Iventosch

unread,
Nov 1, 2016, 11:47:12 PM11/1/16
to FLARE
Interesting concept/model, but I'm not sold yet...still not comfortable with replacing recurrence with convolutions...(not that I wouldn't certainly appreciate the speedup in training time!)

Joel

Peter Stone

unread,
Nov 2, 2016, 8:24:58 AM11/2/16
to FLARE

In case anyone on FLARE isn't aware of CNJC...

=================

Dear all,
The computational neuroscience journal club will be meeting Thursday at 2:30 in NMS 4.106. This week Kim will be
presenting the paper below:


Raghu M, Poole B, Kleinberg J, Ganguli S,  Sohl-Dickstein J. (2016). On the expressive power of deep neural networks.
[[https://arxiv.org/abs/1606.05336]]
Our website is at [[http://clm.utexas.edu/compjclub/]], if you want to see what we've discussed previously or if you
want to suggest papers.
Hope to see you there!
Rishi

Reply all
Reply to author
Forward
0 new messages