Automated Machine Learning with Monte-Carlo Tree Search

Rakotoarison, Herilalaina; Schoenauer, Marc; Sebag, Michèle

doi:10.48550/arxiv.1906.00170

Cited by 5 publications

(13 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our hybrid models, we take the GP regression formulation to model both categorical and continuous variables; we use Monte Carlo tree search for categorical variables and acquisition maximization for continuous variables, which brings the best of two methods (Rakotoarison et al, 2019;Ru et al, 2020). For the search policy, we heuristically search the categorical space using MCTS.…”

Section: Hybrid Modelsmentioning

confidence: 99%

“…Our hybrid model architecture uses MCTS to generalize and unify several state-of-the-art mixed-variable models as shown in Table 2, especially Mosaic (Rakotoarison et al, 2019) and CoCaBO (Ru et al, 2020). This construction serve to unifies existing models.…”

Section: Monte Carlo Tree Searchmentioning

confidence: 99%

“…In terms of BO, we can treat the surrogates g of the black-box function f as the reward function. We adopt the notation f (s, •) which fixes categorical variables in f corresponding to the combination represented by the path s. Previously, Rakotoarison et al (2019) took UCB and constructed a GP surrogate model g s (•) independently that approximates the black-box function f (s, •). We also introduce the ε-greedy search policy to enhance exploration in place of the deterministic UCB policy, with a probability ε ∈ (0, 1), following .…”

Section: Monte Carlo Tree Searchmentioning

confidence: 99%

“…Hybrid models MCTS dependent GPs Mosaic (Rakotoarison et al, 2019) MCTS independent GPs GP (skopt) (Head et al, 2020) sampling independent GPs CoCaBO (Ru et al, 2020) 1-layer bandit dependent GPs EXP3BO (Gopakumar et al, 2018) 3…”

Section: Categorical Variables Continuous Variables Hybrid Modelsmentioning

confidence: 99%

See 3 more Smart Citations

Hybrid Models for Mixed Variables in Bayesian Optimization

Luo¹,

Cho²,

Demmel³

et al. 2022

Preprint

View full text Add to dashboard Cite

We systematically describe the problem of simultaneous surrogate modeling of mixed variables (i.e., continuous, integer and categorical variables) in the Bayesian optimization (BO) context. We provide a unified hybrid model using both Monte Carlo tree search (MCTS) and Gaussian processes (GP) that encompasses and generalizes multiple state-of-the-art mixed BO surrogates.Based on the architecture, we propose applying a new dynamic model selection criterion among novel candidate families of covariance kernels, including non-stationary kernels and associated families. Different benchmark problems are studied and presented to support the superiority of our model, along with results highlighting the effectiveness of our method compared to most state-of-the-art mixed-variable methods in BO.

show abstract

Section: Hybrid Modelsmentioning

confidence: 99%

Section: Monte Carlo Tree Searchmentioning

confidence: 99%

Section: Monte Carlo Tree Searchmentioning

confidence: 99%

Section: Categorical Variables Continuous Variables Hybrid Modelsmentioning

confidence: 99%

See 2 more Smart Citations

Hybrid Models for Mixed Variables in Bayesian Optimization

Luo¹,

Cho²,

Demmel³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…There is a variety of approaches that can be used to identify the optimal design of the data-driven model. For instance, AutoML solutions can be based on random search [5], Bayesian optimisation [6], reinforcement learning (RL) [7], Monte Carlo tree search [8], sequential model-based optimization [9], gradient-based approaches [10]. However, most of them are less flexible than evolutionary approaches to the model design (implemented e.g.…”

Section: Related Workmentioning

confidence: 99%

Multi-Objective Evolutionary Design of Composite Data-Driven Models

Polonskaia

Nikitin

Revin

et al. 2021

2021 IEEE Congress on Evolutionary Computation (CEC)

View full text Add to dashboard Cite

In this paper, a multi-objective approach for the design of composite data-driven mathematical models is proposed. It allows automating the identification of graph-based heterogeneous pipelines that consist of different blocks: machine learning models, data preprocessing blocks, etc. The implemented approach is based on a parameter-free genetic algorithm (GA) for model design called GPComp@Free. It is developed to be part of automated machine learning solutions and to increase the efficiency of the modeling pipeline automation. A set of experiments was conducted to verify the correctness and efficiency of the proposed approach and substantiate the selected solutions. The experimental results confirm that a multi-objective approach to the model design allows achieving better diversity and quality of obtained models. The implemented approach is available as a part of the open-source AutoML framework FEDOT.

show abstract