AutoML¶
Survery¶
Taking Human out of Learning Applications: A Survey on Automated Machine Learning
Automated model selection¶
Simple Search¶
- Grid search
- Random search
Derivative-Free Optimization¶
- Heuristic Search
- Model-Based Derivative-Free Optimization
- Bayesian Optimization: builds a probabilistic model
- Gaussian process
- Tree-based model
- Deep network
- Bayesian Optimization: builds a probabilistic model
Neural Architecture serach¶
Automated Feature Engineering¶
Feature Enhancing Methods¶
Dimension reduction¶
- PCA -Principal components analysis [On lines and planes of closest fit to systems of points in space (1901)]
- LDA -Linear Discriminant Analysis [The use of multiple measurements in taxonomic problems]
- Auto-encoders Extracting and composing robust features with denoising autoencoders
Feature generation¶
Feature encoding¶
Papers¶
MnasNet (CVPR 2019)¶
MnasNet: Platform-Aware Neural Architecture Search for Mobile by Google
NAS-FPN (CVPR 2019)¶
NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection
EfficientNet (ICML 2019)¶
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks by Google Brain
Blog |
Code: tensorflow/tpu/models/official/efficientnet
如何评价谷歌大脑的EfficientNet? - 知乎
used in EfficientDet
Others¶
Google’s AutoML: Cutting Through the Hype Neural architecture search vs. transfer learning: two opposite approaches
~[D] Google’s AutoML: Cutting Through the Hype grad student descent lol~