Dana-Farber / automl-in-healthcare-review

Automated machine learning: Review of the state-of-the-art and opportunities for healthcare

Home Page:https://doi.org/10.1016/j.artmed.2020.101822

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

AutoML in Healthcare Review

Automated machine learning: Review of the state-of-the-art and opportunities for healthcare

Selected highlights from the 2020 AutoML Review [https://doi.org/10.1016/j.artmed.2020.101822] that reviewed over 2,160 works related to the field of automated machine learning.

The curated list of automated feature engineering tools for Automated Machine Learning

Full details in https://www.sciencedirect.com/science/article/pii/S0933365719310437?via%3Dihub#tbl0005

Method Work Feature Engineering Technique Used by how many works
Deep Feature Synthesis LINK Expand-Reduce 151
Explore Kit LINK Expand-Reduce 53
One Button Machine LINK Expand-Reduce 32
AutoLearn LINK Expand-Reduce 16
GP Feature Construction LINK Genetic Programming 68
Cognito LINK Hierarchical Greedy Search 38
RLFE LINK Reinforcement Learning 21
LFE LINK Meta-Learning 34

Automated machine learning pipeline optimizers

Full details in https://www.sciencedirect.com/science/article/pii/S0933365719310437?via%3Dihub#tbl0010

Method Work Optimization Algorithm Data Pre-Processing Feature Engineering Model Selection Hyperparameter Optimization Ensemble Learning Meta-Learning Used by how many works
Auto-Weka LINK Bayesian Optimization (SMAC) ✔️ ✔️ ✔️ 703
Auto-Sklearn LINK Joint Bayesian Optimization and Bandit Search (BOHB) ✔️ ✔️ ✔️ ✔️ ✔️ 542
TPOT LINK Evolutionary Algorithm ✔️ ✔️ ✔️ ✔️ 84
TuPAQ LINK Bandit Search ✔️ ✔️ 94
ATM LINK Joint Bayesian Optimization and Bandit Search ✔️ ✔️ ✔️ 29
Automatic Frankensteining LINK Bayesian Optimization ✔️ ✔️ ✔️ 12
ML-Plan LINK Hierarchical Task Networks (HTN) ✔️ ✔️ ✔️ 24
Autostacker LINK Evolutionary Algorithm ✔️ ✔️ ✔️ 18
AlphaD3M LINK Reinforcement Learning/Monte Carlo Tree Search ✔️ ✔️ ✔️ 8
Collaborative Filtering LINK Probabilistic Matrix Factorization ✔️ ✔️ ✔️ ✔️ 29

Neural Architecture Search algorithms, based on performance on the CIFAR-10 dataset

Full details in https://www.sciencedirect.com/science/article/pii/S0933365719310437?via%3Dihub#tbl0015

NAS Algorithm Work Search Space Search Strategy Performance Estimation Strategy Number of Parameters Search Time (GPU-days) Test Error (%)
Large-scale Evolution LINK Feed-Forward Networks Evolutionary Algorithm Naive Training and Validation 5.4M 2600 5.4
EAS LINK Feed-Forward Networks Reinforcement Learning and Network Morphism Short Training and Validation 23.4M 10 4.23
Hierarchical Evolution LINK Cell Motifs Evolutionary Algorithm Training and Validation on proposed CNN Cell 15.7M 300 3.75
NAS v3 LINK Multi-branched Networks Reinforcement Learning Naive Training and Validation 37.4M 22400 3.65
PNAS LINK Cell Motifs Sequential Model-Based Optimization (SMBO) Performance Prediction 3.2M 225 3.41
ENAS LINK Cell Motifs Reinforcement Learning One Shot 4.6M 0.45 2.89
ResNet + Regularization LINK HUMAN BASELINE HUMAN BASELINE HUMAN BASELINE 26.2M - 2.86
DARTS LINK Cell Motifs Gradient-Based Optimization Training and Validation on proposed CNN Cell 3.4M 4 2.83
NASNet-A LINK Cell Motifs Reinforcement Learning Naive Training and Validation 3.3M 2000 2.65
EENA LINK Cell Motifs Evolutionary Algorithm Performance Prediction 8.5M 0.65 2.56
Path-Level EAS LINK Cell Motifs Reinforcement Learning Short Training and Validation 14.3M 200 2.30
NAO LINK Cell Motifs Gradient-Based Optimization Performance Prediction 128M 200 2.11

About

Automated machine learning: Review of the state-of-the-art and opportunities for healthcare

https://doi.org/10.1016/j.artmed.2020.101822