Experimental Design on a Budget for Sparse Linear Models and Applications.

Area of Science:

• Statistics
• Machine Learning
• Experimental Design

Background:

• Optimal experimental design is crucial but challenging under budget constraints.
• Sparse linear models are prevalent in high-dimensional machine learning.
• Existing strategies are limited for budget-constrained sparse linear model design.

Purpose of the Study:

• To propose novel strategies for budget-constrained optimal design of experiments.
• To address the challenge of sparse linear models in high-dimensional settings.
• To develop tractable algorithms applicable to a broader class of sparse models.

Main Methods:

• Introduced two new strategies: one geometric and one algebraic.
• Developed algorithms that are tractable and generalizable.
• Validated methods through extensive experiments on benchmarks and a neuroimaging study.

Main Results:

• Proposed geometric and algebraic strategies effectively address budget constraints in sparse linear models.
• Algorithms are computationally efficient and applicable to various sparse models.
• Experimental results demonstrate practical effectiveness, particularly in a neuroimaging context.

Conclusions:

• The novel strategies offer efficient solutions for budget-constrained experimental design in sparse high-dimensional settings.
• The developed algorithms are practical and demonstrate effectiveness in real-world applications.
• Findings may inform future enrollment strategies for scientific studies.