-
It doesn’t look like the temperature is annealed in your gumbel softmax. Is there a reason for this as it is not standard? @tkipf
-
For many bayesian scientists [and for one of my recent application domains] there's recently been a lot of hype around [this article](https://arxiv.org/abs/1611.01144) about how to learn categorical v…
-
I would like to test your code with the transformer architecture in fairseq.
Have you ever tried?
Could you please suggest me the better way to do that?
-
Hi author, thanks for your work!
I want to ask you about the idea of the article, that is, why you should use Gumbel-softmax - if I understand correctly, the input in this article is not sampling.E…
ithok updated
2 years ago
-
- https://arxiv.org/abs/1611.01144
- 2016
カテゴリー変数は、世界の離散的な構造を表現するための自然な選択です。
しかし、確率的ニューラルネットワークでは、サンプルをバックプロパゲートすることができないため、カテゴリー型の潜在変数を使用することはほとんどない。
本研究では、カテゴリー分布からの非微分サンプルを、新しいGumbel-Softmax分…
e4exp updated
2 years ago
-
## 一言でいうと
離散値の予測(=クラス分類など)を行う際の出力について。クラスの特定に使うargmaxなどは微分可能でないためうまく誤差伝搬できない。なので、連続/離散分布の特性を調整する変数を導入した方法を提案。その名はガンベル・ソフトMAX
### 論文リンク
http://blog.evjang.com/2016/11/tutorial-categorical-vari…
-
# Categorical Reparameterization with Gumbel-Softmax - Reading Collections
[https://owen-liuyuxuan.github.io/papers_reading_sharing.github.io/Building_Blocks/GumbelSoftmax/](https://owen-liuyuxuan.…
-
From my understanding, the input of `F.gumbel_softmax` (i.e., the` logits` parameter) should be the \log of a discrete distribution. However, I didn't see any softmax or log_softmax before the gumbel_…
-
Is it possible to use Gumbel softmax trick with a beam search decoder in the existing implementation of Texar. I want to differentiate through the generated beam search samples into the generator netw…
-
Hi, @HEmile , I test the example in examples/vae/discrete_vae.py, but find the gumbel softmax performs much better than rebar, relax and reinforce (testing loss after 10 epochs: 98 for gumbel, 165 for…