Not meant as an issue, but a little perplexed by feature_dim... #15
Comments
Hi, |
Hi, Yes, having such a bottleneck is a very crucial architectural choice. I did try a bunch of other variations (different size, having an mlp, etc.), but couldn't really come up with anything that works better than this. |
Very interesting! I suppose it helps generalization. I'm surprised by the massive de-compression of the |
Initially I used |
Yeah, that makes sense. I hadn't thought of it that way. |
I noticed
feature_dim
is set to50
, which is quite a bottleneck from the encoding dim of32 * 35 * 35
and the downstreamhidden_dim
of1024
. Very interesting. Do you think the bottleneck helps create some kind of better compression for learning?The text was updated successfully, but these errors were encountered: