Automated Machine Learning in Practice: State of the Art and Recent Results
read more
Citations
Benchmark and Survey of Automated Machine Learning Frameworks
Automated Machine Learning: The New Wave of Machine Learning
Applications of machine learning algorithms for biological wastewater treatment: Updates and perspectives
The Road to Explainability is Paved with Bias: Measuring the Fairness of Explanations
Automated Machine Learning—A Brief Review at the End of the Early Years
References
Long short-term memory
Support-Vector Networks
Neural Architecture Search with Reinforcement Learning
Sequential model-based optimization for general algorithm configuration
A Tutorial on Bayesian Optimization of Expensive Cost Functions, with Application to Active User Modeling and Hierarchical Reinforcement Learning
Related Papers (5)
Frequently Asked Questions (10)
Q2. What is the way to tune a support vector machine?
For instance, the trained optimizer is used to tune the hyperparameters of a Support Vector Machine [29] without accessing the gradients of the loss function with respect to the hyperparameters.
Q3. What are the three main problems targeted by the literature?
Feature engineering: Feature preprocessing, representation learning and selecting the most discriminant features for a given classification or regression task are problems targeted by the literature.
Q4. What is the main idea of a complete pipeline?
A complete pipeline includes data cleaning, feature engineering (selection and construction), model selection, hyperparameter optimization and finally building an ensemble of the top trained models to obtain good performance on unseen test data.
Q5. What research lines have grabbed the attention in the past years?
The problem of manual hyperparameter tuning [13] inspired researchers to automate various blocks of the machine learning pipeline: feature engineering [14], meta-learning [15], architecture search [16] as well as full Combined Model Selection and Hyperparameter optimization [17] are the research lines which grabbed a great deal of attention in the past years.
Q6. What is the main idea behind Explorekit?
Explorekit [19] not only iteratively selects the features but also generates new feature candidates to obtain the most discriminant ones.
Q7. What is the main point of the paper?
In this paper, the authors have presented an independent evaluation of current approaches in the field of shallow AutoML, and have presented their own version of Portfolio Hyperband that shows promising results in terms of computational efficiency while being on par with the state of the art accuracy-wise.
Q8. What is the architecture for a network?
Real et al. [31] propose an evolutionary architecture search based on a pairwise comparison within the population: the algorithm starts with an initial population as parents, and every network undergoes random mutations such as adding and removing convolutional layers and skip connections to produce offspring.
Q9. What is the process of generating the top 20% of the pipeline?
it selects the top 20% of the generated population based on cross-validation accuracy, and produces 5 descendants from each by randomly changing a point in the pipeline.
Q10. What is the main idea of learning to optimize?
future work aims at the idea of learning to optimize, but the meta-training paradigm is changed to reinforcement learning to enable training on more realistic, i.e. non-smooth objective functions.