Automatic Composition of Complex Pipelines via End-To-End Learning and User Interaction

Abstract

Automated machine learning (AutoML) has recently gained attention due to the ability to help data science experts rapidly create prototypes of machine learning solutions. The existing frameworks have shown an improvement of performance on common tasks such as tabular classification and regression. In the real world, it is challenging to decide which AutoML framework to use for a given task since there is no standard method to evaluate them. This report presents our work on providing utilities for the standardization of AutoML components such as evaluation tools, generalization of a pipeline language, standardization of pipeline execution, and general-purpose code creation. Also, we present different AutoML systems created during the different stages of the program and the creation of specialized frameworks for Neural architecture search and time-series outlier detection.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Aug 01, 2021
Accession Number
AD1144426

Entities

People

  • Diego Martinez-garcia
  • Shuai Huang
  • Xia Hu

Organizations

  • Texas A&M University

Tags

Communities of Interest

  • Autonomy

DTIC Thesaurus Topics

  • Air Force
  • Air Force Research Laboratories
  • Artificial Intelligence Software
  • Bayesian Networks
  • Computational Science
  • Computer Languages
  • Computer Programming
  • Computers
  • Data Mining
  • Deep Learning
  • Detection
  • Information Processing
  • Information Science
  • Information Systems
  • Kernel Functions
  • Machine Learning
  • Neural Networks
  • Probabilistic Models
  • Supervised Machine Learning

Fields of Study

  • Computer science

Readers

  • Distributed Systems and Data Platform Development
  • Neural Network Machine Learning.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks