-
- https://arxiv.org/abs/1608.08710
- 2017 ICLR
様々なアプリケーションにおけるCNNの成功は、計算コストとパラメータ保存コストの大幅な増加を伴っている。
これらのコストを削減するための最近の取り組みとして、元の精度を損なうことなくさまざまな層の重みを刈り込んだり圧縮したりすることが行われている。
しかし、重みの大きさに基づくプルーニングでは…
e4exp updated
3 years ago
-
After configuring Tailwind in a project created with `hyperapp-starter` I noticed 2 things:
1. Autoreload has a delay of ~1.5s
2. Couldn't use Tailwind classes globally, such as `Hello`, but inste…
-
Pretty much a necessity, so best figure out an approach.
-
There are a few places where the trackshots is iterated over framewise and then regrouped person-wise, which is quite bad since it's actually already person-grouped. The reader should provide person g…
-
Hi Johan,
I would like to have an efficient way of serializing (with cereal) HashMaps and HashSets. I think going through lists is inefficient because the hashes have to be recomputed (although I hav…
-
## What was wrong?
The ORM data model is setup with the following *loose* constraints.
1. A `Header` has a **nullable** foreign key to it's parent
2. A `Block` **must** point to a header
3. A …
-
# 論文リンク
https://arxiv.org/abs/2001.04451
# 公開日(yyyy/mm/dd)
2020/01/13
# 概要
・計算コストの高い Transformer の計算効率効率のための手法を 2つ提案
・内積注意 → 局所性鋭敏型ハッシュを使用して計算量オーダーを O(L^2) から O(L log L ) に削減 (L はシーケンス長)
…
-
I think it's necessary to make ZooeyBot a bit more efficient.
For my NM95 runs, the difference between using full auto and ZooeyBot is not even a minute.
While the bot refreshes, which is probably…
-
```
Currently it always creates a Key with unknown exact TypeId (except for
single type hierarchies). But it must use nearly the same logic as
ItemMaterializer - i.e. it must add exact TypeId to the K…
-
Probably I'm just missing something obvious, but given I have a component tree like:
```
|---|
| R |
|---|
|
|------|------|
| |
…