-
# One-Shot
- 将搜索空间定义为超级网络 `SuperNet` - 包含全部网络
- `One-Shot` 共享权重
- `One-Shot` vs `One-Stage`
- `One-Stage` - 搜索完之后模型可以直接使用,而不再需要retrain等操作
- `One-Shot` - 构造 `SuperNet`
- 大部分 `One-Stag…
-
# NAS Metrics
- Accuracy
- Efficiency
- GPU Days = N X D = `Number of GPUs` X `Number of days`
- Kendall Tau metric
Kendall Tau metric | ![image](https://user-images.githubusercontent.com/…
-
# DARTS
- `RL & EA` --> `DARTS`
- 离散领域的黑盒优化问题 --> 通过`可微的方式`进行结构搜索
- DARTS - Differentiable Architecture Search - 松弛(Relaxation)搜索空间使之连续 - 梯度下降 - AO
- CNN
- RNN
## Continuous Relaxati…
-
# ProxylessNAS
- `Proxy Task` -> `Target Task`
- Proxy - 小数据集(`Cifar`)/`Fewer Blocks`/`Fewer Epoch` -> 大数据 (`ImageNet`)/`Stacked Top-Performing Bloocks`/Retraining
- 把 NAS 视为 `path-level prun…
-
# ENAS
- 子模型共享参数, 提高效率,减少所需计算资源
- macro search space over entire convolutional models
- micro search space over sonvolutional cells
- Decision Block
- Conv Cell
- Reduction Cell
## Se…
-
# Search Space
Search Space | Description
-- | --
Entire Structure | 完整的网络结构,每层的结构从 search space 中选择,`skip-connection` 可选
Cell-Based | 先搜索表现好的 `Cell`, 再堆叠 `Cell`
Hierarchical 分层 | 增加 `Cell` 的…
-
# PNAS
- Progressive Neural Architecture Search
- 将大的搜索空间分解为小的搜索空间的乘积
- Operator -> Block -> Cell -> CNN
- 不区分 `Normal Cell` & `Reduction Cell`
- `Normal Cell` 功能实现 - stride=1
- `Reduction…
-
`AutoTS` is a popular autoML library centered around an all-in-one forecaster class of the same name.
https://winedarksea.github.io/AutoTS/build/html/source/intro.html#autots
It would be nice to hav…
-
DataRobot과 같은 AutoML (Automated Machine Learning) 도구는 기계 학습 모델을 개발하는 과정을 자동화하여 사용자가 쉽게 데이터를 분석할 수 있도록 합니다. 이러한 도구들은 데이터 사이언스 분야에서 큰 인기를 얻고 있으며, 여러 기업과 연구자들에게 유용한 자원이 되고 있습니다. 몇 가지 유명한 AutoML 도구들을 소개해 …
-
- [x] sklearn=0.23 support (PR intel-analytics/analytics-zoo#3304)
- [ ] update setup and doc
- [ ] test on jenkins with the latest version