On taking advantage of opportunistic meta-knowledge to reduce configuration spaces for automated machine learning

Article Computer Science, Hardware & Architecture

VolcanoML: speeding up end-to-end AutoML via scalable search space decomposition

Yang Li et al.

Summary: This article introduces VolcanoML, an AutoML framework that addresses the scalability issues of existing systems when dealing with large and high-dimensional search spaces. VolcanoML decomposes the search space into smaller subspaces and guides the search process with an execution plan. Experimental results demonstrate that VolcanoML outperforms systems like auto-sklearn in search space decomposition and performance.

VLDB JOURNAL (2023)

Add to Collection

Review Computer Science, Artificial Intelligence

AutoML: state of the art with a focus on anomaly detection, challenges, and research directions

Maroua Bahri et al.

Summary: The past decade has witnessed rapid development in machine learning research, but the performance of multiple algorithms is sensitive to factors such as hyper-parameter tuning and data cleaning, requiring significant human effort to achieve good results. Automated machine learning (autoML) aims to make machine learning algorithms more accessible and reliable, with most research efforts focusing on supervised learning and few proposals for unsupervised learning.

INTERNATIONAL JOURNAL OF DATA SCIENCE AND ANALYTICS (2022)

Add to Collection

Article Computer Science, Artificial Intelligence

AutoWeka4MCPS-AVATAR: Accelerating automated machine learning pipeline composition and optimisation

Tien-Dung Nguyen et al.

Summary: Automated machine learning pipeline composition and optimisation aim to automate the process of finding the most promising ML pipelines within allocated resources. However, existing methods that evaluate pipelines by executing them often require a significant amount of time and prevent exploration of complex pipelines. To address this issue, a surrogate model named AVATAR is proposed to evaluate ML pipelines without execution, allowing for quicker rejection of invalid pipelines and evaluation of more pipelines within the same time budget. Integrating AVATAR into the SMAC algorithm configuration results in finding better solutions compared to using SMAC alone.

EXPERT SYSTEMS WITH APPLICATIONS (2021)

Add to Collection

Article Computer Science, Artificial Intelligence